samples and cellSize were validated but isoValue and origin were not:
iso = NaN or +inf silently returned an empty mesh, and origin = NaN gave
NaN positions. Extend Extract's existing top-of-method validation authority
to reject non-finite isoValue and non-finite origin components with
ArgumentException, matching the existing style, with tests for each
component.
Overflow at extreme magnitudes (cellSize ~3e38, or finite samples whose
interpolation overflows to inf) is a separate, float-domain limit rather
than a validated-input error; documented in the XML doc and provenance
notes, deliberately not range-checked.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
TryGetOtherEndpoint's bool return was dead control flow: when a mutant
flipped its `return false` guard to `return true`, `other` stayed a clone
of `coord` (only one component would have been incremented on the true
path that never runs for that guard), so the caller compared a sample
against itself and behaved identically either way. That produced the
equivalent survivors at Extract's :92 continue and the guard's :198
return.
Bound the edge loop with the same condition (coord[axis] < cellCounts[axis])
directly, computing `other` inline once the bound holds. One path instead
of two that happened to agree; the helper is gone, and so is the `continue`
that used to depend on it. Behaviour is unchanged: verified with a
before/after run of the existing suite plus a temporary harness comparing
Extract's Positions/Quads/QuadEdges against ca220a5's code on sphere,
non-cubic, open-plane (each axis), checkerboard, and minimal 2x2x2
fixtures - all byte-for-byte identical.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Mutating obj is CultGeometryGridEdge other && Equals(other) to && true
passed the suite: nothing exercised object.Equals directly with a mismatched
X/Y/Z/Axis, null, or an unrelated type. Add a behavioural test that checks
each component independently, rejects null and other types, and agrees with
the typed Equals and GetHashCode.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The provenance doc still described the deleted geometric winding rule
(diagonal cross product must have a positive dot product with the
inside->outside direction) as the spec. That rule was replaced by the
inside-flag times per-axis handedness table (OrientationSign: +1, -1, +1;
axis 1's perpendicular pair (X, Z) is left-handed) when the geometric
fallback was deleted. State that as the actual rule, with the
outward-cross-product property following only on non-degenerate quads.
Also record that non-finite isoValue/origin are rejected.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The `==`->`!=` and `< 0`->`> 0` mutants on Extract's orientation check are
both Killed. The one survivor, `OrientationSign[axis] <= 0`, is an
equivalent mutant, not a coverage gap: OrientationSign only ever holds ±1,
so `< 0` and `<= 0` agree on every value it can produce. No vstest
coverage-based false survivor has actually been confirmed for
GameCult.Geometry; remove the withdrawn claim and keep the correct notes
about full-SHA --since, safe.directory, and config-change scope fan-out.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Stryker's diff-scoped run against this cut reported CultGeometrySurfaceNets
.Extract's `if (inside == (OrientationSign[axis] < 0))` - the sole authority
for quad winding after F4's fix deleted the geometric cross-product fallback
- as a Survived Equality mutation (== -> !=). Hand-mutating it and running
dotnet test directly failed 4 of 38 SurfaceNetsTests (the three per-axis
corner-order tests and the outward-winding test), confirming it is a real
vstest coverage-based test-selection miss, not a genuine gap. Recorded this
as a second confirmed case alongside the mtp/static-readonly-initializer
caveat, per this cut's verification requirement to hand-confirm any survivor
that looks like it must be covered.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The three TryGetOtherEndpoint survivors were triaged as a vstest coverage
false-survivor, mirroring the confirmed mtp/static-readonly caveat above.
That triage was wrong: the actual mutant is `return false` -> `return true`
inside each per-axis guard, which is equivalent (it makes the method report
an edge going nowhere, so the caller compares a sample against itself and
skips it exactly as it would with a correct `false`). Rewrote the paragraph
to say so, and removed the unsupported claim that this project has a
confirmed vstest false-survivor case. Also recorded, per this cut's
verification requirements, that `--since` needs a full commit SHA and that
`safe.directory` must be set in a container.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Fixed An_extremely_small_edge_delta_falls_back_to_the_edge_midpoint (renamed
A_tiny_but_normal_edge_delta_...): it pinned the wrong value (1/6) for the
fallback that F3's production commit deleted; now asserts the exact
interpolated value, hand-derived in double precision independently of
Interpolate. Added a genuinely subnormal-delta sibling test.
- Added two more vertex-placement fixtures (crossing counts 4 and 6) to kill
the sum/count -> sum/3 mutant, each with a hand-derived expected mean.
Counts of 1 and 2 from the finding are not used: a single cell's 12 edges
form the cube graph Q3 (3-regular, 8 vertices), whose edge-cut size for any
corner sign pattern is never 1 or 2 for any nonzero cut (minimum nonzero
cut is 3, already covered by the existing symmetric/unequal fixtures) -
see the fork note in this cut's report.
- Added three corner-order pinning tests (one per axis), each asserting the
exact welded vertex at each of Quads[0..3] for a chosen edge against a
hand-computed position, including the axis-1 v1/v3 swap that F4's fixed
OrientationSign now performs. These kill the axis-0-mirrored, c0/c2-swapped
and equivalent per-axis corner-order mutants.
- Added four degenerate-quad orientation tests (sample exactly at iso,
large-origin/tiny-cellSize sphere, octahedron at iso, tiny-tolerance
sphere) asserting every shared undirected edge is traversed in opposite
directions (net zero), via a new AssertEdgeDirectionsBalance helper that
also covers the 4-use case. Restricted the existing
Every_quad_winds_outward_from_inside_to_outside test's geometric check to
non-degenerate (non-zero cross product) quads, per spec.
- Added Grid_edge_inequality_is_the_negation_of_equality (F6: != is compiler-
mandated alongside ==, and needed its own direct test) and
A_non_finite_sample_is_rejected (F7: NaN and +-Infinity samples).
All 38 SurfaceNetsTests pass; the 7 pre-existing GeometryDocumentTests
failures from CultLib commit e420410 are untouched.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
F1: Extract's locals (position, v0..v3) were assigned through pattern-match
and short-circuited out-var chains whose definite-assignment analysis breaks
under a single mutated operator, so Stryker's Safe Mode dropped all 98
mutants in Extract as compile errors. Restructured to plain null checks and
a for-loop over a uint[4] so every local is unconditionally assigned; no
mutation of a boolean operator can make it fail to compile.
F3: deleted the `<= 1e-20f` fallback in Interpolate. A straddling edge always
has delta > 0 (.NET preserves subnormals), so the guard only ever replaced a
correct interpolated value with a wrong 0.5 midpoint.
F4: deleted the per-quad geometric cross-product winding test. Orientation
now comes only from the topological inside/outside flag plus a fixed
per-axis handedness sign (OrientationSign; axis 1's (X, Z) perpendicular
pairing is the one left-handed case, X x Z == -Y). This is now determined
purely by which endpoint of the crossing edge is inside and the axis, never
by quad geometry, so degenerate (zero-area) quads orient consistently with
their neighbours instead of falling back to an unstable dot-product sign.
F7: Extract now rejects any non-finite sample (NaN or +-Infinity) with
ArgumentException, in the same style as the existing shape/cellSize checks.
F8: collapsed three duplicated (edge, axis) mappings - TryGetOtherEndpoint's
per-axis switch, the three Axis*EdgeStarts tables plus the axis==0?1:0
ternaries in AccumulateAxisEdges, and TryGetQuadCells' three axis-specific
corner blocks - into one PerpAxes/PerimeterLoop table pair, used by all
three. Replaced the three duplicated `<= isoValue` inside tests with one
IsInside predicate. Deleted the per-axis trailing `InRange(x, cellsX)` /
`InRange(y, cellsY)` / `InRange(z, cellsZ)` checks (provably always true,
since TryGetOtherEndpoint already guarantees the edge's own-axis coordinate
indexes a valid cell) and the explicit TryGetValue+throw invariant guard
(the four cells around a crossing edge are proven to always have a vertex;
a direct dictionary index now fails loudly via KeyNotFoundException if that
invariant is ever actually violated, with no reachable branch to maintain).
F6: deleted CultGeometryGridEdge's <, >, <=, >= operators (CultGeometryQuadMesh.cs).
Nothing in production or tests used them; ordering goes through CompareTo.
!= is kept only because C# requires it whenever == is defined (CS0216).
Builds clean; does not touch CultGeometryIsoSurface.cs, CultGeometryDocuments.cs,
or ISOSURFACE-PROVENANCE.md.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Three TryGetOtherEndpoint boundary-guard mutants (per-axis "if (x + 1 >=
sizeX) return false;" forced to "if (true) return false;") were reported
Survived despite being exercised by the X/Y/Z open-boundary tests. Hand-
mutating all three and running dotnet test directly failed 11 of 24
SurfaceNetsTests, confirming a vstest-runner coverage-attribution false
survivor (same class of caveat already recorded for CultMath's mtp runner,
here under the default runner). Recorded under Known tool caveat.
Separately, adding stryker-config.json in the same cut that first uses it
defeats --since line scoping: Stryker fell back to mutating the whole
GameCult.Geometry.csproj glob ("Non-CSharp files in test project were
changed"), not just the new files, which is why the run's survivor list
includes CultGeometryIsoSurface.cs, CultGeometryDocuments.cs, and
CultGeometryPrimitives.cs. Recorded under Follow-ups so a later diff-only
cut isn't surprised by the difference.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A --since:29e50ad Stryker pass found real gaps, all inside this cut's own
files:
- vertex-placement and normals fixtures anchored a vertex at the world
origin, so a mutated (+ instead of -) or dropped cellSize multiply was
invisible against a zero operand; moved them off the origin;
- the cube normals test only checked length, and three orthogonal
+-2-length face contributions summed to the same length no matter which
one a mutant negated; now checks the exact expected direction per vertex;
- the open-boundary rule only had an X-axis test, leaving TryGetQuadCells's
Y- and Z-axis guards (and TryGetOtherEndpoint's Y/Z cases) uncovered;
added the Y- and Z-axis twins of the existing plane test;
- the 1e-20 near-zero-delta fallback in Interpolate was never exercised;
added a fixture with an edge delta of ~3e-21;
- CultGeometryGridEdge's Equals and ToString had no direct test.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The 4x4x4 pure-checkerboard field had no fully-interior cell: every active
cell bordered the array edge, so some crossing edges lost a quad to the open-
boundary rule and edge usage came out odd (observed: all-3) for reasons
unrelated to the ambiguity the test means to exercise. Wrapped the
checkerboard in a one-sample uniform-outside margin (5x5x5 grid, checkerboard
only on samples 1..3) so every crossing edge under test keeps all four
surrounding cells in range, isolating the even-but-not-2 claim from
open-boundary effects.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Two spots failed to build: a method-group reference to float.IsFinite and an
`is 1 or 2` pattern inside a lambda that FluentAssertions compiles as an
Expression<Func<T, bool>>; expression trees can't contain pattern-matching
`is`, and a method group isn't implicitly convertible to an expression tree
either. Rewritten as explicit lambdas with `==`/`||`.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Adds SurfaceNetsTests.cs and QuadNormalsTests.cs covering the sign
convention, edge naming, outward winding, vertex placement (midpoint and
unequal-crossing means, origin/cellSize), closed-sphere welding and
Euler characteristic, an ambiguous checkerboard cell's even-but-not-2 edge
usage, open-boundary quad suppression, determinism/ordering, and
face-weighted normals (sliver-vs-large-quad area weighting, unit cube
normals, zero-area/untouched-vertex degenerate case). Adds
stryker-config.json (vstest runner; this project is NUnit 4, not xunit.v3)
and records GameCult.Geometry as scoped, diff-only, in
docs/mutation-testing.md's follow-ups.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Implements CultGeometrySurfaceNets.Extract (welded CultGeometryQuadMesh, one
quad per interior crossing grid edge, outward-oriented winding) and
CultGeometryQuadNormals.FaceWeighted (unnormalized diagonal-cross-product
area weighting, matching CultGeometryIsoSurface's sign convention and
degenerate-normal rule). Fresh implementation per SURFACE-NETS-PROVENANCE.md;
no code or tables copied from the cited literature.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Stryker.NET becomes CultLib's mutation tool (CultMath config, baseline 2243
mutants, 350 survived). Behavioural tests close the three areas Aetheria
leans on: the BatchMath SIMD body, first_order_intercept's branches, and
equality and comparison across the vector types. 350 -> 264 survivors;
Soul passed the cut after one accumulation test was added.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Soul's pass on 41440cb found BatchMath.cs:115
(AddRadialFalloffAcceleration2D(ReadOnlySpan<float2>, ...)) accumulates
onto the caller's acceleration buffer, but no test ever seeded that
buffer with nonzero values: mutating += to = passed all 133 tests. The
float-array overload already had this guard
(RadialFalloffAccumulatesOntoExistingAcceleration); add the float2
equivalent, with a nonzero, per-element-distinct starting buffer, mixing
in-range and out-of-range particles so both the accumulated and
unchanged-seed cases are checked.
Checked every other += site in BatchMath.cs for the same gap: the
array-overload's scalar tail and vector body, and both
IntegrateSemiImplicitEuler2D overloads, all already seed their
accumulator buffers (vx/vy/px/py, velocity/position) with nonzero,
per-element values in their existing tests. Only this one site was bare.
Verified by hand-mutating BatchMath.cs:115 (+= to =): the new test fails
(expected 6.78, actual 0.48); reverted the source byte-exactly after.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The vector-vector self-tie added in the previous commit does not cover
float2/float3/float4's vector-scalar and scalar-vector overloads: their
existing no-tie fixture (a distinct vector against a single scalar) never
had a component equal to that scalar, so their boundary-flip mutants
(< to <=, > to >=) kept surviving. Add a uniform vector (every component
the same value) against a scalar equal to that value, tying every
component of both scalar overloads at once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The no-tie fixture in each *ComparisonOperatorsDifferPerComponent test
correctly kills direction mutations (< swapped for >) but leaves every
strict/non-strict boundary swap (< to <=, > to >=) surviving, since with
no equal component the two behave identically. That default is
appropriate for a float-threshold epsilon check, but these are
general-purpose vector comparison operators (CultMath's own
RectIntersectionTreatsTouchingEdgesAsContact test compares literal
endpoints), and for the int types an exact tie is the routine case, not
the rare one. Add a complementary "compared against itself" call per type
(vector-vector for double2/double3/int2/int3/int4, plus vector-scalar and
scalar-vector for float2/float3/float4) so every component ties and the
boundary swaps are actually exercised.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`Precision` was declared `const float`, so `Assert.Equal(a, b, Precision)`
bound to xunit's absolute-tolerance overload (allowing any two values
within 4.0 of each other) instead of the intended 4-decimal-place
overload. Hand-mutating the `t2 > 0 ? min(t1, t2) : t1` ternary to always
return `t1` (math.cs:287) exposed it: the assertion silently accepted a
result almost 0.27 away from the expected value. Declaring it `const int`
selects the correct overload; confirmed the same hand mutation now fails.
Also add the float2-overload euler-integration length-validation test the
mutation report showed missing (BatchMath.cs:208's ValidateEqualLengths
call had no test at all).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Soul-style self-check against the actual mutation report (not just green
tests) found several fixtures whose values coincidentally produced the
same output on the buggy path as on the correct one:
- BatchMath's scalar-tail fixtures put the one remainder particle/step
either outside the falloff radius or mask-frozen, so the y-axis
arithmetic a mutation would touch was never reached with an observable
effect. Pin the remainder particle in range (radial falloff) and dynamic
(Euler integration) so the scalar tail's own math is exercised, not just
the vector body's.
- The float2-overload of AddRadialFalloffAcceleration2D had no no-op
(radius <= 0 / strength == 0) test at all.
- first_order_intercept_time's negative-determinant test used a
perpendicular relative velocity, making b = 2*dot(vel, pos) exactly
zero; a mutant that drops the "no solution" branch entirely and falls
through to the linear fallback coincidentally computes 0 too. Use a
non-perpendicular fixture so the fallback's nonzero value is
distinguishable from the correct zero.
- No fixture ever reached the `t1 > 0` branch at all (a > 0, both
quadratic roots real and positive), so the boundary/negate checks
guarding it and the `t2 > 0 ? min(t1, t2) : t1` line were never
covered. Add that case.
- bool2.@false/@true, bool3.@true and bool4.@false have no consumer
anywhere in the suite (bool3.@false and bool4.@true are read by
HlslSemanticsTests); pin all six vector constants directly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Soul: a git-URL package lands read-only in Unity's cache, where a file with
no meta is warned about and ignored on every import.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Hand-verified that weakening float3.Equals from && to || passes the whole
suite. The same shape recurs with no shared base type across float2/3/4,
double2/3, int2/3/4, bool2/3/4, quaternion, float2x2 and float3x3. Add
data-driven tests, per component position, that a value differing at
exactly one component is unequal under Equals/==/!=, that GetHashCode
stays consistent with Equals, and that </>/<=/>= (including the
vector-scalar and scalar-vector overloads on the float types) each report
a different result per component. float2x2/float3x3 are covered at row
granularity since they have no == operator; quaternion and bool2/3/4 have
no ordering operators and are covered for equality only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
math.first_order_intercept_time had a single closing-target test. Add
coverage for zero relative velocity, the a-approx-zero linear fallback, a
target that outruns the projectile (negative determinant), a receding
target with both quadratic roots non-positive, and a closing case with
off-origin, off-axis shooter and target state. The closing cases assert
that the projectile (inheriting the shooter's velocity) and the target
actually coincide at the returned time; the no-solution branches pin the
value the source's own contract specifies.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Both existing BatchMath tests used 2-3 particles while LaneCount >= 4, so
only the scalar tail ever ran, and both held y at 0 for every particle, so
every y-axis mutation was invisible. Add fixtures sized LaneCount * 3 + 1
with distinct nonzero x/y per particle, pinned against a scalar reference
computed independently in the test, plus explicit coverage of the early
return, negative deltaTime, length-mismatch, and empty-span guards on all
four BatchMath overloads.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
math.erf/math.erfinv (source, tests, README) already landed on main via
87b95ae, b688cb2, fc9ce97, 4cdc02b, 6e8362b, cc71f17. This is the release
step 0.2.3 skipped needing: bump the package version, rebuild the tracked
Unity DLL/pdb deterministically, and add the CHANGELOG.md this package
never had (docs/semver-policy.md names this exact gap).
- packages/cultmath/src/CultMath/CultMath.csproj: Version 0.2.3 -> 0.2.4.
- packages/cultmath/unity/org.gamecult.cultmath/package.json: 0.2.3 -> 0.2.4.
- packages/cultmath/unity/org.gamecult.cultmath/CHANGELOG.md: new, [0.2.4]
entry only (Added: erf, erfinv). Earlier versions are not backfilled, per
the semver policy's own ruling on this package's pre-existing gap.
- packages/cultmath/unity/org.gamecult.cultmath/Runtime/Plugins/CultMath.dll
and .pdb: rebuilt via packages/cultmath/scripts/build-unity-package.ps1
(dotnet build, netstandard2.1, Release, --no-incremental, no Source Link).
Built on Windows; the assembly is platform-neutral IL. The script's own
determinism check (fresh build hash == tracked hash) and its version-parity
check (package.json version == assembly FileVersion) both pass.
- No HLSL mirror change: erf/erfinv are C#-only, and
packages/cultmath/shaders/CultMath.hlsl already matches the copy under
unity/org.gamecult.cultmath/Shaders/CultMath.hlsl byte for byte.
Verified independent of build-unity-package.ps1's own checks: a scratch
console app loaded the tagged cultmath-unity-v0.2.3 DLL and this build's
fresh DLL into separate AssemblyLoadContexts and diffed their public
surface. Only two members were added (math.erf(float), math.erfinv(float)),
none removed; AssemblyVersion/FileVersion/InformationalVersion match apart
from the 0.2.3 -> 0.2.4 bump; TargetFrameworkAttribute is unchanged
(.NETStandard,Version=v2.1). erf/erfinv round-trip and domain-edge behavior
(erfinv(+-1) = +-Infinity, erfinv(1.5) = NaN) verified by invoking the
loaded methods directly.
dotnet test packages/cultmath/tests/CultMath.Tests -c Release: 102 passed,
0 failed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What lands:
- Cut 3: the native MsQuic bridge, with the Windows release leg and F5
closed on both platforms, and the closerace fixture fix.
- Cut 4: the koffi Node binding, consumer, session manager and trust.
- Cut 5: the TS provider. It reaches the C# native connector, which is
the Unity overlay's path, on Windows.
- Fix batches 1-4: use-after-free, handler isolation, the
handshake-timer eviction, the koffi allocation leak, one C#-shaped
inbox, encode-once, advertisedHost, publish-only
(acceptClientFrames: false), and maxConnections.
- Retention: the provider seeds late joiners with the newest
latest-only frame per (channel, body), in TS and C#.
- Cut 6, narrowed: native binaries committed under
packages/cultmesh-ts/native, built by Actions. A second from-scratch
build reproduced them byte for byte.
Gates: four Opus Soul passes on the runtime and one Sonnet pass on the
packaging. The last runtime pass found that the connection death under
load was the old handshake eviction, which fix 3a had removed, and it
confirmed StreamPixels' load (60 Hz latest-only for 10 minutes) with
flat memory.
Also carried, per the operator's ruling: CultMath erf/erfinv (Cut 12.0
and its fix batch) from the branch base.
Map conflicts were resolved in favour of main, which owns the maps. The
only branch-side lines main lacked were a Cut 3 "still outstanding"
block that main's closure supersedes.
Open follow-ups, recorded on the map: IPv6-literal endpoints cannot be
dialled; consumer-side native queues are unbounded while Unity is not
polling; a client can see connect() resolve past maxConnections before
the refusal arrives; subtraction (about -130, mostly comments) is
deferred.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Soul found that on a Windows clone (core.autocrlf=true) SHA256SUMS and
MANIFEST.txt were written with CRLF, so the documented
`sha256sum -c SHA256SUMS` failed on every line: each filename ended in
\r. CI stayed green only because it checks out on Linux. After this
rule and a fresh checkout, the check passes on Starfire for all four
binaries.
Provenance, recorded here because verify-committed only checks the
committed files against the committed sums: run 35912901846 rebuilt
all four binaries from scratch after they were committed, and every
artifact is byte-identical to the committed file (libmsquic.so.2
c09a72f9..., libgamecult_mesh_quic_native.so 0ee3d283..., msquic.dll
c17c6581..., gamecult_mesh_quic_native.dll 9dae391a...). Two
independent CI builds produce the same bytes, which shows both where
they came from and that the build is reproducible.
Self edited this one-line config directly instead of dispatching Hands.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`cultmesh-quic-native-{win32-x64,linux-x64}.tar.gz` collided by substring
with the negative grep for the parked `@gamecult/cultmesh-quic-native-*`
platform package names (11.9, not built in this campaign). Rename to
`quic-native-{win32-x64,linux-x64}.tar.gz`, matching the upload-artifact
names already used. The release job has not run (no tag pushed), so this
has no runtime effect yet.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The binaries are committed and verify-committed went green on the trigger's
own run (https://github.com/GameCult/CultLib/actions/runs/35912901846).
Reverts quic-native.yml to the triggers named in cut 6 (docs/typescript-
quic-realtime-cut.md:2153-2156): push to main and pull_request run
verify-committed only, workflow_dispatch runs the two builds, a
cultmesh-quic-native-v* tag runs all four. Diff against 4d75d90 (the file
before the temporary trigger was added) is empty.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Downloaded from the TEMPORARY push-triggered run
https://github.com/GameCult/CultLib/actions/runs/35912386052 (commit
e5bf1c9), built by build-win32-x64 and build-linux-x64 in quic-native.yml.
Every file's digest was verified against the job logs before this commit
(win32-x64/msquic.dll's also matches build-quic-native.ps1's own pinned
expectedRuntimeSha256, confirming it is the pinned MsQuic 2.5.9 runtime).
win32-x64/gamecult_mesh_quic_native.dll 9dae391a44d59f12e19baa08a44d86c7e8b4916ca5d40007a42d65913ce0896f
win32-x64/msquic.dll c17c65813fe007eb486a5270d244f3659261355795b27abb99bd19ff77539263
linux-x64/libgamecult_mesh_quic_native.so 0ee3d28324c79a5b0135368187e7bd75c1969ad94fc6985755e81de4ec96e430
linux-x64/libmsquic.so.2 c09a72f9b3bfd9119044b305c6064731c1e1879411e6f54276474fb58b524b68
SHA256SUMS is generated sha256sum output over exactly these four relative
paths, committed verbatim. MANIFEST.txt records the pinned MsQuic sources,
the bridge commit, the run URL, and each build job's printed host/compiler
line; build.release notes this run used the TEMPORARY push trigger rather
than a tag, since no release exists yet.
Cut 6 (packaging), 11.5/11.7, Q8 Option 1: the operator (via this session)
downloads, verifies and commits; no workflow writes to the tree.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
workflow_dispatch cannot be dispatched against quic-cut6 until this workflow
file is registered from the default branch (a GitHub constraint, not a
permissions gap: `gh workflow run` and a raw dispatch API call both 404).
Rather than touch main or push a tag, add push: branches: [quic-cut6] and
extend verify-committed's and both build jobs' `if:` guards with a
`refs/heads/quic-cut6` arm, so this push itself runs the three jobs. The
release job's guard is untouched: it still matches only a
cultmesh-quic-native-v* tag, so this push does not create a release.
Every added line is marked TEMPORARY and paired with its match; both are
removed in one commit once the binaries are committed and verify-committed
goes green on that push.
Cut 6 (packaging).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Four jobs: verify-committed checks packages/cultmesh-ts/native/SHA256SUMS
against the tree on every push to main and every pull request; build-win32-x64
and build-linux-x64 run scripts/build-quic-native.{ps1,sh} on
workflow_dispatch or a cultmesh-quic-native-v* tag and upload their output as
artifacts; release, needing both builds and gated to the tag push, is the
only job with write permission and attaches the two platform archives plus a
committed-format SHA256SUMS and a draft MANIFEST.txt to a GitHub Release.
build-linux-x64 runs inside the pinned debian:13@sha256:f324c7ff… container
per Q6 A: the Actions container is the only permitted Linux release builder,
because the artifact links against that image's glibc and the deploy host
(Yggdrasil) runs the same OS. No release or tag has been pushed by this
commit; the release job exists but has not run.
Cut 6 (packaging), docs/typescript-quic-realtime-cut.md:2131-2156, narrowed
per StreamPixels cut map section 4.3.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The package directory must be self-contained once copied out of the repo
(pnpm materialises only `files`), so the licence text already committed
beside the Unity plugin binaries is copied here rather than referenced.
Same MIT text for both platforms. Cut 6 (packaging), 11.4.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The closure smoke already asserted dist/index.js and dist/index.d.ts for
every workspace package; extend it for cultmesh-ts to assert the four
committed native binaries plus SHA256SUMS, MANIFEST.txt and
MSQUIC-LICENSE.txt are in the packed file list, on every host. The tarball
carries both platforms; the host only decides which one loads. Cut 6
(packaging), 11.5/11.6.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
`* text=auto` already autodetects them by content, but that is a heuristic.
Say explicitly that packages/cultmesh-ts/native/**/*.dll and *.so(.N) are
binary so no future text-processing rule can apply EOL conversion and poison
the digests SHA256SUMS pins. Cut 6 (packaging).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The committed QUIC native binaries only reach a consumer if `files` carries
them: pnpm materialises a file: dependency from `files` alone, and the same
allowlist gates `npm pack` and the deploy `git archive`. Cut 6 (packaging).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The fixture fix (44ab376) lived only on quic-cut3-winrel. Cut 4 branched
from cc71f17 before it, so the QUIC line never carried it. This merges it
into quic-cut6 before packaging, so the release line runs the closerace
fixture whose poll timeout outlasts its own settle budget.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The synthetic connection P17 drives directly at acceptConnection() never
reaches CONNECTED (it was never dialed anywhere), so it never joins `peers`
either way — connectionCount alone couldn't tell "refused" apart from
"attached but still handshaking". Manually restoring the disposed-check's
removal and rerunning showed the original assertion passing regardless.
retain() is synchronous and only the attach path calls it, so comparing
CultMeshQuicNativeRuntime.refCount immediately around the acceptConnection()
call is what actually distinguishes the two: confirmed by hand that the
mutation now fails this test by assertion (2 !== 1) before restoring it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
P3 ("a latest-only send failure evicts the peer...") raced wall-clock
against a tight while-loop that does no real I/O per iteration (broadcast()
never awaits a latest-only peer's send), so it spun encoding and enqueueing
frames fast enough to starve the event-loop turns the native pump needs to
notice the dead connection — a self-inflicted race that any extra load in
the test process (mutation testing included, regardless of which line was
mutated) could push past the 2s bound with nothing to do with the mutant
under test. Fixed by decoupling how many attempts to seed (a small fixed
count, each with a real yield) from how long to wait for the result (one
waitUntil afterward).
New tests for mutants Soul found surviving by assertion:
- I1: an inbox key keeps its first-arrival queue position when a later
publish overwrites its value.
- L1: the outbox keeps one send in flight per peer, proven by
prototype-patching sendEncodedFrame to track concurrency (the provider
builds its peer transports internally, so this is the only vantage point
from outside the module).
- P15: broadcast() after dispose rejects.
- P17: an accept arriving after dispose is refused, not attached, driven
directly at acceptConnection() with a real (but otherwise unused) native
connection handle rather than a fabricated pointer.
Plus coverage for the new options: a publish-only provider faults a client
that sends to it and never queues anything, and a connection beyond
maxConnections is refused without disturbing existing peers.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul's probe proved a client can grow the provider's memory without limit
(20k reliable frames plus 20k distinct latest-only keys took
receiveQueueSize to 40000 and RSS from 97 to 318 MB) because StreamPixels
never calls receive() to drain it, and there was no cap on connections
either. The pinned certificate only authenticates the server, so any
internet client can dial in.
Add acceptClientFrames (default true, unchanged for existing callers and
interop peers, which all use receive()): when false, any inbound stream a
peer opens faults that connection at STREAM_STARTED, before a frame can be
decoded or queued, so no inbox or generation-map entry is ever created for
client data, and receive() itself rejects clearly.
Add maxConnections (default 1024): a connection beyond the cap is refused
at accept, shut down before any transport, outbox, or retained-frame
seeding is allocated for it. Existing peers are unaffected.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This brings fix batch 3c (per-runtime out-buffers, freed on the last
release, plus a steady-state RSS test in its own child process) onto
the branch carrying fix batches 3a and 3b and retention.
The merge was automatic. Checked by hand: release() frees the reused
buffers only after the reference count reaches zero and the pump has
exited. retain() (from 3a) is called only while the provider already
holds a reference, so it cannot resurrect a runtime that is closing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Hand-mutating CultMeshRealtimeInbox.receive() to check `completed`
before the ready queue (so dispose() rejects an already-queued frame
instead of draining it first) found no existing test that dies for it.
Add one: two reliable-ordered frames sit in the provider's receive
queue, the provider is disposed, and both frames must still come back
from receive() before the next call rejects.
Verified by hand against the final code and restored, per the fix
batch's mutation list; not committed as a harness.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Reproduces Soul's probe directly: 20,000 client-sent latest-only frames
on one key, none ever drained by receive(), leave exactly 1 pending
frame in the provider's receive queue, and that frame carries the
newest sequence sent. A second test sends 20 frames across 50 distinct
keys and checks the queue holds exactly 50 (bounded by key count, not
frame count). A third proves reliable-ordered client frames are never
coalesced and stay in order at the provider.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul's probe found readyFrames growing without limit: a publish-only
consumer of CultMeshQuicRealtimeProvider.receive() (StreamPixels' Node
service) let a single unauthenticated client key grow the provider's
receive queue by one entry per frame sent, forever.
Replace CultMeshRealtimeLatestOnlyCoalescer plus the three separate
ready-queue/waiter implementations built around it (the transport's
inbound path, the provider's outbox pump, and the provider's own
readyFrames array) with one CultMeshRealtimeInbox<T>, shaped like the
C# reference's CultMeshRealtimeInbox: latest-only entries coalesce to
at most one pending value per (channel, body) key, keeping that key's
first-arrival queue position; reliable-ordered entries always queue,
unbounded; receive() drains whatever is queued before a complete(error)
rejects further calls.
The provider's pumpReceivedFrom now publishes into this inbox under the
frame's (channel, body) key instead of pushing onto an unbounded array,
so a slow or absent receive() caller backs up to at most one pending
frame per distinct key. Added provider.receiveQueueSize (test-visible,
mirrors the existing trackedStreamCount precedent) to make that bound
observable.
realtime-quic.ts: 1298 -> 1273 lines (HEAD 1b0c71d), -16 against the
275010a anchor. Soul estimated -90 to -120 from unifying four ~209-line
sites into one ~80-line class; the actual saving is smaller because two
of those sites already need their own promise/waiter/AbortSignal
plumbing in TypeScript (no Channel with native CancellationToken
support), so consolidating them removes duplication but the surviving
single copy of that plumbing is not itself small. No further
correctness-neutral cut was found without code-golfing working logic.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Without sendOnStream's disposed guard, the queued send never settles (no
event or sweep is left to reject it), so a bare assert.rejects(send, ...)
would hang the whole file instead of failing this one test. Race the send
against a 3s timeout that rejects with a different message, so the mutant
is a loud assertion mismatch either way the native layer happens to fail
first.
Verified by hand: reverted sendOnStream's `if (this.disposed)` guard to
`if (false)` locally, rebuilt, and ran P8 alone — it died by assertion in
81ms (a native stream-send error on the already-shut-down connection,
still not matching /disposed/i), not a timeout. Restored the guard before
committing. Also hand-verified fix 1 the same way: commenting out the
`clearTimeout(handshakeTimer)` on CONNECTED reproduces the original
handshake-timer eviction, and the new fix-1 test dies by assertion
(0 !== 1) rather than a hang.
Full Windows pass on Starfire after restoring: 158/158 cultmesh-ts tests
and all 5 QUIC interop tests (including the native-connector and
late-join lanes), msquic.dll confirmed as the pinned build (4,181,856
bytes, SHA-256 C17C6581...).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
By the time the memory test ran in-process, ~150 earlier tests in the
suite had already grown and partly freed the glibc heap, so a real
per-call native leak could land in already-mapped free space and never
show up as RSS growth in that process: confirmed by hand, the
in-process version of this test kept passing even with `nextEvent`
mutated back to allocating fresh koffi buffers every call, once it ran
inside the full suite rather than alone.
Spawn a dedicated `node --expose-gc` child process for the measurement
instead. A fresh process has no such history, matching the ~8 MiB
(fixed) vs ~30 MiB (pre-fix) split measured in isolation. Confirmed by
hand again with the same mutation: the child-process version now fails
reliably (~28.7 MiB growth) inside the full 150-test suite, not just
standalone.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The prior version opened connection B only after A had already faulted,
then asserted a boolean flag set by B's own CONNECTED event. That flag
never flipped on Linux: something about accepting a second connection
right after the first one's forced shutdown made the assertion
unreliable, even though A's fault itself isolated correctly (visible in
the log: the onFault throw fired once, promptly).
Rewritten to establish B as a normal, fully connected loopback pair
first, fault A only afterward, then prove B's handler is still live by
sending an actual frame on B and confirming the server side still
receives it — a stronger assertion than a connected flag, and one that
does not depend on a second connection's accept timing relative to a
fault that just tore another one down.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
P8's test opened a fresh reliable stream on the first send, so disposing
before its microtask ran hit streamOpen() against an already-shut-down
connection (its own native error) instead of ever reaching sendOnStream's
disposed guard. Prime the stream with an awaited first send so the second,
raced send chains straight onto sendOnStream with no streamOpen call in
between.
P10b's test registered connection A's and B's server-side listeners after
an `await`, the exact async-registration gap fix 1 exists to close: a
connection's next native event can be dispatched, and silently dropped
(no listener registered yet), before that await's continuation runs.
Register both listeners synchronously inside their LISTENER_NEW_CONNECTION
callback instead.
The IPv6 fix 3 test asserted the wrong expectation: WHATWG URL's own
`hostname` keeps the brackets for an IPv6 literal, so the round trip
through parseQuicRealtimeEndpoint correctly returns "[::1]", not "::1".
Confirmed via a Linux verify run (cultlib-quic-native-dev image): before
this commit, exactly these three new tests failed and every other test
(155) passed, including every other fix/P3 test added alongside them.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The 5 MiB bound was a guess and false-failed on the fixed code: 200k
direct `nextEvent` calls carry their own ~8 MiB RSS noise floor from
thread-pool marshaling and promise bookkeeping, even with zero
`koffi.alloc` calls in the loop. Measured on Yggdrasil, same build and
host, same 200k-call loop:
fixed code (0 koffi.alloc calls in the loop): ~8.0 MiB RSS growth
pre-fix code (400k koffi.alloc calls, unfreed): ~29.9 MiB RSS growth
18 MiB sits well above the fixed baseline and well below the leaking
one, so the test stays robust to ordinary run-to-run noise while still
catching the regression.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The `.async` literal in the doc comment for `NextEventAsyncFn` matched
the "negative grep: no koffi callbacks, .async only on nextEvent" test's
`/\.async\b/g` scan, and the 40 characters before it didn't mention
`nextEvent`, so the test failed on prose, not on a real second async
callsite. Reworded to describe koffi's async export without the
literal `.async` token.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
realtime-quic-provider.test.ts: a peer stays connected and keeps
receiving well past handshakeTimeoutMs (must fail on the pre-fix code);
an invalid reliable-ordered or latest-only frame throws from broadcast()
without evicting any peer; advertisedHost is used in place of the bind
address and round-trips through the module's own parser; an IPv6 bind
advertises a bracketed literal; receive() with an already-aborted signal
rejects immediately; and a latest-only send failure evicts the peer
(P3 — the outbox's own send-failure path, not covered by the existing
reliable-ordered eviction test).
realtime-quic-consumer.test.ts: a reliable send still queued behind
reliableSendTail rejects instead of hanging when disposal wins the race
(P8 — sendOnStream's this.disposed guard).
realtime-quic-native.test.ts: a throwing onFault handler is isolated to
its own connection; the pump survives and a second connection's handler
still runs (P10b).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul's probes found every provider peer evicted at handshakeTimeoutMs
(10s by default): StreamPixels' Node service would drop every Unity
overlay 10s after it connected. Root cause: acceptConnection registered
a temporary buffering handler that was the only thing setting
state.connected, then attachPeer awaited CultMeshQuicNativeRuntime.open()
before swapping in the real handler and replaying buffered events. The
native bridge always emits NEW_CONNECTION before CONNECTED as separate
events; when CONNECTED landed after the swap, it went straight to
transport.handleConnectionEvent (which has no case for CONNECTED) with
nothing left to clear the handshake timer, so a fully healthy peer was
evicted on schedule regardless.
Fix: acceptConnection takes a synchronous retain() on the provider's own
already-open runtime instead of awaiting a fresh open(), so one handler
owns the connection from NEW_CONNECTION onward with no async window, no
handler swap, and no buffered-event replay. PendingAccept, its buffering
handler, and attachPeer are gone; dispose() now sweeps `accepted` (every
transport a connection ever got, whether or not its handshake reached
CONNECTED) instead of the deleted pendingAccepts set. Cut C's
retained-frame seeding still happens atomically before the peer joins
`peers`.
Also, while touching broadcast() and sendFrame():
- Encode and validate each frame exactly once in broadcast(), fanning the
encoded bytes out to every peer and outbox, instead of re-encoding once
per peer inside sendFrame(). An invalid frame (e.g. a 70,000-byte
channelId) now throws to the caller before touching any peer, rather
than failing the same way for every peer in turn and evicting all of
them while broadcast() itself reported success. The provider's retained
latest-only frames now cache their encoded bytes too, so seeding a newly
attached peer never re-encodes them either.
- Add an `advertisedHost` provider option, independent of the bind
address: `host` is typically 0.0.0.0, an empty string, or an IPv6
wildcard for binding, none of which describe a route a client can dial,
and the native listener only accepts IP addresses in the first place,
not a DNS name. IPv6 literals are bracketed correctly in the advertised
endpoint. The uppercase cert-sha256 pin is unchanged (the operator ruled
StreamPixels pins a long-lived self-signed certificate).
- provider.receive() with an already-aborted signal now rejects
immediately instead of registering a waiter an already-fired abort event
can never reach.
- The duplicated "datagrams not supported" message (sendFrame and
broadcast each had their own copy) is now one shared constant.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A caller that already holds a live runtime instance (the QUIC realtime
provider, handing a reference to each newly accepted peer) needs one more
reference without crossing an async boundary. `open()` itself has no I/O
to await, but being `async` still forces every caller through a microtask
hop; `retain()` just increments the shared refcount and returns `this`,
synchronously, so the caller stays inside the same turn.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Calls the runtime's private `nextEvent` directly (0 ms timeout, no
listener/connection open, so each round trip returns immediately with
status 0) 200,000 times after a warm-up, then asserts RSS growth
across the run stays under 5 MiB with `global.gc()` forced before and
after. The old per-call `koffi.alloc` code allocates a ~64 B event
struct plus a 4 B int32 every iteration and never frees either, so it
clears this bound by roughly 2x; the fixed code, which allocates
nothing per call, stays flat. Skips when no native bridge is built or
when Node wasn't started with `--expose-gc`; `npm test` now passes
`--expose-gc` so the check runs by default instead of always skipping.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The `LoadedNativeLibrary` interface just re-listed the sixteen fields of
the object literal `loadBindings` already built, one liability the
methods below never needed a second time. Split the library-building
body into `buildLibrary()` and let `LoadedNativeLibrary` be
`ReturnType<typeof buildLibrary>`; `loadBindings` is now a thin cache
wrapper around it. `nextEventAsync`'s signature stays explicit as
`NextEventAsyncFn`, since `promisify` can't infer one from koffi's
untyped `.async`.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A Soul-verified probe found every koffi out-parameter allocation
(runtime open, listener open, connection open, stream open, and both
per-`nextEvent` buffers) was never freed. `nextEvent` runs in the pump's
tight poll loop, so this leaked roughly 53 bytes per idle cycle even
with no traffic, and grew unbounded under real frame volume (214 MB of
RSS over 4M allocations on the pinned koffi 3.3.0 build).
Allocate `nextEvent`'s out-event and out-required buffers once per
runtime instance and reuse them for every call instead of once per
call: the pump never has more than one `nextEvent` in flight, so
decode-then-reuse is safe, and they are freed once in `release()`
after the pump loop exits. Every other allocation (`open`,
`listenerOpen`, `connectionOpen`, `streamOpen`) is synchronous and
short-lived, so those are freed with `koffi.free` right after decode.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Cut 5's TS provider reaches the C# native connector on Windows (the Unity
path). Fix batch 2 closed Cut 4's use-after-free, handler and leak
findings, and fixed the root lockfile. Retention seeds late joiners in
both runtimes.
The Soul gate blocked a merge. Every peer is evicted at the handshake
timeout, koffi allocations leak, client frames pile up without bound,
and the chain-validated connect path does not exist. For the last one
the operator ruled a pinned self-signed certificate.
Fixes 3a (provider) and 3c (binding leak) are in flight as separate
per-file batches.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cut C's headline cross-process proof. The TS `serve` peer already tore its
listener down the instant it finished broadcasting, so nothing could dial it
late; it gains `--linger-ms` to hold the listener open afterward instead. The
interop harness starts that provider, runs a throwaway `dial` peer that
receives every frame and disconnects (the provider requires a first peer
before it broadcasts at all, so this is also what proves broadcasting
finished), then only after that dials with the C# native connector and
asserts the frame it gets is the retained one, by payload, not just
non-empty. New `NativeConnectorReceivesTheRetainedFrameWhenJoiningLate` in
GameCult.Mesh.Quic.Native.Tests carries the assertion; the existing
`NativeConnectorReceivesFromExternalProvider` is untouched. Windows-only,
matching the native connector under test.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
AReliableOrderedFrameIsNeverRetainedAndIsNotDeliveredToALateJoiner disposed
the first session and then waited on server.ConnectionCount == 0 before
attaching a second one, which timed out under Yggdrasil's ygg-verify
container (dotnet test, cultlib-quic-native-dev image): the retention rule
under test does not depend on the first peer disconnecting, so the fix keeps
both sessions connected concurrently and drops the flaky wait.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
One sentence each in the C# QUIC README and the transport-planes doc: a
provider retains the newest latest-only frame per (channel, body) and
delivers it to a peer as soon as it attaches, bounded by the number of
distinct keys.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CultMeshQuicRealtimeServer mirrors the TS provider: a `_retained` dictionary
holds the newest latest-only frame per (channel, body) key, updated in
BroadcastAsync with the same CultMeshGeneration comparison the receive-side
filter already uses. AcceptLoopAsync seeds a newly accepted client with the
retained frames. Both the retained update+fan-out and the client
registration+seed run under one lock (`_retainedGate`), so a seed and a
concurrent broadcast are totally ordered and a client can never end up seeded
with a frame older than one it already received via broadcast. Reliable-
ordered frames are events and are never retained.
CultMeshGeneration is now internal so the server can reuse the transport's
own generation comparison instead of a second implementation.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CultMeshQuicRealtimeProvider now keeps a `retained` map of the newest
latest-only frame per (channel, body) key, updated on broadcast using the
same generation comparison the receive-side filter already uses. Every peer
attached after that point is seeded with the retained frames before it is
added to the peer set, so a StreamPixels overlay that connects or reconnects
after state was published converges on it instead of waiting for the next
change. Reliable-ordered frames are events and are never retained.
Operator ruling (2026-09-23, OQ1 = A): the provider retains, not a service
heartbeat or a per-peer attach hook.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The mid-handshake test still hung under M2 in the full workspace test run
even after the previous two fixes (pendingAccepts sweep, try/finally).
Diagnosed by printing the stuck value: refCount settled one below its own
"before" snapshot, not above — the opposite of a leak. Root cause:
CultMeshQuicRealtimeProvider.dispose() releases its runtime reference
fire-and-forget (`void this.runtime.release()`), so when M2 makes every
preceding test's own dial fail during the handshake too, their releases
can still be landing when this test starts and snapshots "before". One
lands mid-test, after the snapshot, and refCount never returns to a value
that already excluded it.
Added waitForRefCountToStabilize(): polls until refCount stops moving for
a short quiet window before treating it as a baseline. Verified against
the actual M2 mutation, in the full `npm run test` workspace run (not
just this one file, which had already stayed clean in isolation): the
file now exits in the low tens of seconds with 9 expected failures (every
test that dials through the now-broken cert check) and none of the
"Promise resolution is still pending" hangs from the three earlier
attempts. Confirmed clean on the unmutated tree first.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Same hygiene sweep as the previous commits, applied to the three tests
whose own subject is a fault or cleanup path that, if it regressed, would
leave a receiveFrame() call with nothing to ever resolve or reject it:
the two "fix 1" throwing-handler tests (a malformed frame that should
fault the connection, and a peer shutdown that should run cleanup()) and
the "fix 3" peer-shutdown test. Each now wraps its transport in
try/finally and uses receiveFrameOrTimeout() instead of a bare
receiveFrame() behind assert.rejects.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Re-running M1 after the previous fix: the assertion now correctly failed
(assertRejectsAndDisposes disposed the resolved client transport), but
the file still hung. Root cause: under M1 the handshake actually
completes (the connector accepts any certificate), so the provider's
peer is fully attached rather than left mid-handshake, and the test's
`provider` was never wrapped in try/finally — the failed assertion threw
past `provider.dispose()`, leaking the listener and the now-live peer.
provider.dispose() is idempotent, so wrapping the whole body costs
nothing on the normal path.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The "dispose() evicts a connection still mid-handshake" test (previous
commit) used a bare assert.rejects(connector.connect(...)) to force the
handshake-rejection path it needs, which is exactly the pattern that hung
realtime-quic-consumer.test.ts under M1: found re-running M1 against the
full workspace after the previous fix — it hung again, this time in
realtime-quic-provider.test.ts, at this new test, for the same reason.
Ported the same assertRejectsAndDisposes() helper into this file rather
than importing it across test files (each file already keeps its own
small helpers).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Found running the M2 mutation (certificate pin compared case-sensitively)
on Yggdrasil: it broke the handshake for every provider test that dials
with the provider's uppercase-advertised pin, each test correctly failed
its own assertion in under 40ms, but the whole file then hung for two and
a half minutes past its last visible test result.
CultMeshQuicRealtimeProvider.dispose() only disposed connections already
in `this.peers` (attached after reaching CONNECTED). A connection whose
handshake never gets there — a rejected certificate, a stalled client —
sits in acceptConnection()/attachPeer()'s own state with its own
handshakeTimeoutMs timer and, once attachPeer's own
CultMeshQuicNativeRuntime.open() resolves, its own runtime reference,
entirely outside dispose()'s reach. Nine tests each leaked one such
pending accept under M2, each held open for up to its 10s default
handshake timeout, serialized enough to add up to the observed hang.
Fixed by tracking every in-flight accept's `evict` closure in a
`pendingAccepts` set that dispose() sweeps alongside `peers`; each accept
removes itself once it settles, whether attached or evicted, so the set
never grows for a healthy, long-lived connection. Added a regression test
that pins a handshake to reject on a long (30s) handshakeTimeoutMs and
asserts dispose() releases the runtime reference in under 2 seconds, not
after the timeout.
Also fixed the same underlying test-hygiene gap the M2 investigation
surfaced directly: the M4 and M4b tests (a stream-kind/delivery mismatch
should fault a connection) left `transportA` undisposed with an unbounded
`receiveFrame()` behind a bare `assert.rejects`; if the fault path
regresses, that call hangs forever instead of failing. Added
receiveFrameOrTimeout() (a bounded receiveFrame, in the same spirit as
assertRejectsAndDisposes()) and a `finally { transportA.dispose(); }` to
both tests.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
packages/cultmesh-ts/package.json has depended on koffi 3.3.0 since Cut 4
landed the native QUIC binding, but the root package-lock.json was never
regenerated to match: it carried no koffi entry at all, so `npm ci` at the
repo root failed with EUSAGE. The StreamPixels deploy host runs exactly
`npm ci && npm run build:ts`, so this blocked deployment.
Regenerated on Yggdrasil (Linux, the cultlib-quic-native-dev container),
not on the operator's workstation, via `npm install`. Verified from a
clean clone on the same host: `npm ci && npm run build:ts` both succeed
against the regenerated lockfile.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Found running the M1 mutation (bypasses the certificate-accept check) on
Yggdrasil under cgroup-limited CPU: three tests correctly failed their
assertion (`assert.rejects(connector.connect(...))` / `manager.connect(...)`
observed a resolve instead of a reject), but realtime-quic-consumer.test.js
then hung instead of exiting — "Promise resolution is still pending but the
event loop has already resolved". Each of those assertions discarded the
connect() call's return value on the assumption it would always reject, so
the transport that unexpectedly connected under the mutant was never
captured, never disposed: a live native connection, an active
onConnectionEvent registration and an outstanding runtime reference kept
the process alive after node:test had already recorded the failures. This
is the exact scenario the batch names by history ("Soul saw M1 and M3 fail
their assertion and then hang the file").
Added `assertRejectsAndDisposes()`, a drop-in replacement for
`assert.rejects()` that disposes whatever the promise resolves to before
failing the assertion, and used it at all eight call sites in this file
that assert a connect() call rejects (two direct connector.connect() calls,
six through the session manager). Verified on Yggdrasil: M1 re-run now
fails the same three assertions and the file exits on its own.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Real bug found while writing the "dispose with a pending send in flight"
regression test, not a pre-scoped fix: sendOnStream() registered a
pendingSends entry unconditionally. If the reliable-send microtask queued
by sendFrame() (chained onto reliableSendTail) ran after cleanup() had
already executed — cleanup()'s own reject-sweep runs once and is guarded
against re-entry, and offConnectionEvent means no later native event can
resolve the entry either — that entry's promise then never settles,
leaking the pending send forever. Confirmed by a hanging test that
reproduced intermittently (~20-25% of runs) when a peer transport's
dispose() raced an in-flight reliable-ordered sendFrame() on the same
transport; not reproducible in isolation outside the full node:test file
(over 100 standalone cycles clean), which points at the race window being
narrow enough that only real inter-test timing variance opens it.
sendOnStream() now checks `this.disposed` before registering, closing the
gap: a send that loses the race rejects immediately instead of
registering an entry nothing will ever settle.
Also gives that regression test its own `{ timeout: 10_000 }` so a
recurrence fails that one test loudly instead of hanging the whole file
for the harness's 60s+ default; 60 consecutive full-file runs were clean
with both changes in place (25 with the fix alone still showed the hang
occasionally, so the fix closes a real but not the only contributing
factor — recorded here rather than claimed as a full proof).
New coverage for the Cut 5 provider's runtime-release-on-every-path
requirement:
- a reliable-ordered send failure evicts the peer without broadcast()
throwing to the caller;
- the runtime reference is released when the listener itself fails to
open (a real PKCS12 credential, but a port already bound by another
listener, so listen()'s own `catch { await runtime.release(); throw
error; }` actually runs) — split out from a same-named test that had
only exercised an earlier failure point (extractLeafCertificateDer(),
before CultMeshQuicNativeRuntime.open() is ever called), which is kept
as its own, more honestly named test;
- dispose() with a pending send in flight releases the runtime reference
(the test whose flakiness led to the fix above).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Ports Soul's uaf.js probe (scenario a: bare open/release cycles,
including a fresh open overlapping the previous runtime's close) into a
committed test. UV_THREADPOOL_SIZE and MALLOC_PERTURB_ must be set before
Node starts, so the test spawns a child `node` process running
test/support/release-uaf-probe.js (plain JS, not compiled, requiring
dist/ directly like the package's own compiled tests) once per pool size
(1, 2, 4) and asserts each exits 0.
Trimmed from the probe's original 300 cycles to 20: at pool size 1 with
the crypto.pbkdf2 saturation load, 300 cycles took over two minutes on
this workstation, and the ordering bug this test defends does not need
that many iterations to reach the race — 20 keeps the whole test under
about 25 seconds while still exercising many open/release cycles with a
fresh open racing a still-closing runtime.
Verified by hand: reverting release()'s order in
realtime-quic-native.ts (runtimeCloseFn before awaiting pumpLoop, the
pre-fix order) did not make this test fail on Windows — MALLOC_PERTURB_
is a glibc feature the Windows CRT ignores, so corruption here is silent
rather than a deterministic crash. The revert-and-confirm check that
actually demonstrates the bug runs on Yggdrasil's Linux container, where
glibc honors MALLOC_PERTURB_; see the verify report for that run's
output. The fix was restored immediately either way, confirmed by a clean
git diff before this commit.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Two scenarios the batch specifically calls for: a throwing handler plus a
garbage frame (the native fault path, through faultConnection()'s now
try/catch-guarded onFault) and a throwing handler plus a peer shutdown
(the transport's own cleanup(), now releasing before running handlers).
Both assert the second onDisposed handler still runs despite the first
throwing, and the peer-shutdown case also asserts
CultMeshQuicNativeRuntime.refCount returns to baseline.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
M4b: the mirror image of the existing M4 test — a latest-only stream
carrying a frame whose own delivery byte says reliable-ordered must fault
only that connection, not the pump.
M10: an equal-generation frame is dropped by the receive-side filter.
Sequenced so it is observable (Soul's note): the first frame is consumed
before the duplicate generation arrives, so the coalescer no longer holds
a pending value the duplicate could silently overwrite with itself.
Without this sequencing, M10's `<=0` -> `<0` mutation is invisible: the
coalescer's overwrite semantics make an unconsumed duplicate look
identical to a dropped one.
F7: split into two tests, because a genuine end-to-end
SEND_CANCELED-through-sendFrame() scenario proved unreliable on this
host. The native-layer test (realtime-quic-native.test.ts) is
deterministic: it round-trips two 64-bit codes above 2^53 through
CONNECTION_SHUTDOWN, exercising the exact decode line F7 mutates
(`code: toBigInt64(decoded.code)`, shared by every native event type,
including STREAM_SEND_COMPLETE). The consumer-level attempt — prime a
reliable stream, start a second send, shut the connection down from the
peer mid-flight — is recorded as a comment: MsQuic here appears to
locally complete a `StreamSend` (as an uncanceled SEND_COMPLETE) faster
than any JS-observable window can land a peer shutdown ahead of it,
regardless of payload size (tried up to 8 MiB) or an explicit delay
before the shutdown call. Recorded as not yet reached at the consumer
layer, not as unreachable.
F6kinds: the existing "stream shutdown prunes streamKinds" test only
proved a second frame still worked afterward, which passes whether or not
the map is actually pruned (a stream id is never reused within one
connection, so a leaked entry has no other behavioural symptom — exactly
why the mutation survived). Strengthened to assert
CultMeshQuicRealtimeTransport.trackedStreamCount (added in the previous
commit) directly.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Covers the session-manager fixes in the previous commit:
- dispose() aborts a dial in flight instead of letting the connector's own
timeout run out before the transport is discarded (asserted by
wall-clock bound, not just by eventual rejection).
- disconnect() during a dial does not let that dial cache a session once
it lands; the manager stays usable afterward.
- F5abort: a losing race candidate is told to abort promptly, proven by
recording when spyConnector's own abort listener actually fires rather
than only asserting eventual disposal (which the mutation's removed
controller.abort() call does not affect, since the cleanup loop that
disposes settled losers is untouched by that mutation).
- F5loser: a losing candidate that connects anyway, past the point abort
would have helped (spyConnector gains an ignoreAbort option to model
this), is disposed rather than left connected.
Replaces the old "fix 4: a connect() that completes after dispose()"
test, which pinned the pre-fix behaviour of waiting out the connector's
full delay before discarding the late transport; that delay is exactly
what dispose() aborting the dial now removes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Fix 1 (consumer half): CultMeshQuicRealtimeTransport.cleanup() ran
onDisposed handlers before releasing the runtime reference, unguarded. A
throwing handler skipped every handler queued after it (the throw exits
the for-of) and skipped the release entirely: on the peer-shutdown path
the refcount stuck and the process never exited; on the fault path the
throw escaped cleanup() into dispatch()'s catch in the native runtime,
which is itself now guarded (previous commit), but cleanup()'s own
contract should not depend on that. release() now runs first, and each
handler runs in its own try/catch.
Low fixes:
- receiveFrame(signal) with an already-aborted signal now rejects
immediately instead of registering an "abort" listener that already
fired and will never fire again, which hung the call forever.
- CultMeshQuicRealtimeConnector.connect() merged its four failure paths
(already-aborted, handshake timeout, later abort, handshake-time error)
into one fail() closure. The handshake-time error path was the one that
mattered: it dropped the runtime reference and rejected but never called
connectionShutdown on the native connection, leaking it at the native
level on every rejected handshake past the certificate stage.
- CultMeshQuicRealtimeSessionManager.dispose() now aborts every dial still
in flight instead of only marking sessions disposed and hoping a live
dial notices later; disconnect() aborts the one in-flight dial for its
target so a race with an in-progress connect() cannot cache the session
disconnect() meant to discard. Threaded through dialAsync/
raceFirstSuccessAsync/connectCandidateAsync as an AbortSignal that
combines with the per-race internal controller.
Subtraction (~-70 net in this file despite the fixes above and new
trackedStreamCount/dialAborts surface):
- SimpleGate replaced by a plain promise-chain field on the transport
(reliableSendTail); the class only ever wrapped one method.
- The two dead streamKinds.set() writes on outbound (self-opened) streams
removed: the map is only ever read for streams that deliver an inbound
STREAM_FRAME event, which a stream this side opened never does.
- The write-only authorityRoute field on CultMeshRealtimeCandidate
deleted; nothing read it.
- The one-use TimeoutError class deleted in favour of Error.
- routeCertificateView moved to cultnet-ts (previous commit); the local
copy deleted.
Also exposes CultMeshQuicRealtimeTransport.trackedStreamCount (streamKinds
map size) for the F6kinds regression test: nothing else could observe
whether STREAM_SHUTDOWN actually prunes the map, since a stream id is
never reused within one connection.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Fix 1 (native half): faultConnection() called the caller-supplied onFault
handler unguarded. A throwing handler escaped faultConnection(), then
dispatch(), then the pump's async loop as an unhandled rejection nobody
awaits, crashing the process on a single bad connection. Wrap the call and
report instead of propagating, matching the try/catch dispatch() already
has around the listener call that reaches faultConnection() in the first
place.
Subtraction: fold the private NativeBindings interface and the
bindingsCache wrapper object into the runtime class's own methods, which
already typed the same twelve calls one door down. The wrapper only
restated that typing a second time. loadBindings() now returns the raw
koffi handles (untyped, koffi has no useful types) and the class methods
do the alloc/decode work directly, the way half of them already did for
their out-parameters. Net -101 lines despite the added refCount getter and
a trimmed module doc comment (no longer narrating the koffi probe process
twice).
Also exposes CultMeshQuicNativeRuntime.refCount as a public static getter
(renaming the backing field to refCountValue): the batch's regression
tests assert it returns to baseline after a fault, a peer shutdown, and
release(); nothing else needed to reach into a private static before.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Both packages mapped an advertised route's wire-shaped certificate
(providerKeyId/providerPublicKeyX/providerPublicKeyY/...) to
CultMeshAuthorityRouteCertificate independently. Move the mapping into
cultnet-ts's cultmesh-authority module, next to verifyAuthorityRoute, and
have both consumers call it instead of duplicating the object literal.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Found by a second real failure, this time on Windows: `dotnet test --logger
"console;verbosity=detailed"` prints a passed test's TestContext.Out lines
twice (live-streamed, then again under "Standard Output Messages:"), and
matching CULTMESH_GOLDEN_SEQUENCE and CULTMESH_GOLDEN_HEX with two
independent regexes let the harness pair a sequence from one copy with a hex
from the other, producing a spurious mismatch even though the C# and
TypeScript sides agreed. One regex now requires them adjacent, exactly as
the C# test's two back-to-back WriteLine calls print them. All four QUIC
interop lanes now pass on Windows (real dotnet test run, native connector
lane included).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Found by a real failure on the Linux verification container: latest-only
coalesces on both the provider's outbox and the managed connector's own
receive-side inbox, so a single ReceiveAsync() call is not guaranteed to
land on the first frame broadcast — on a slower host it landed on sequence 3.
The C# test now also prints CULTMESH_GOLDEN_SEQUENCE, and the harness looks
up the TypeScript peer's logged encoding of that same sequence instead of
hardcoding sent === 1.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Verified live on Windows: a passed NUnit test under `dotnet test --logger
"console;verbosity=detailed"` prints "Test Run Successful." and "Passed: N",
never the literal "Passed!" the assertion was checking for. Both the native
and managed connector lanes now match the real strings; both were confirmed
to pass for real against a live TypeScript provider (lane 1: 137ms, lane 2:
145ms with the golden-bytes hex line matching the TypeScript peer's own
encoding byte for byte).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
TypeScript provider vs C# native connector (the Unity path, gated to
Windows since the native connector under test is); TypeScript provider vs
C# managed connector, with a golden-bytes check comparing the C# test's
re-encoded frame against the TypeScript peer's own encoding of the same
frame; latest-only under a stalled consumer (two dials against one 200-frame
provider, one paused mid-stream via the peer script's new --pause-after),
asserting the provider's broadcasts stay bounded regardless of the pause and
both consumers converge on sequence 200.
Both dotnet-test lanes run `dotnet test --filter FullyQualifiedName~<test>
--logger "console;verbosity=detailed"`: detailed verbosity is what surfaces
a passed test's own TestContext.Out lines, which the golden-bytes lane reads
back out of stdout.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
dial gains --expect-sequence (stop once a frame with that sequence arrives,
rather than counting frames received — the number of frames a stalled
latest-only reader sees is not the interesting value, what it converges on
is) and --pause-after/--pause-ms, which simulate the stalled-consumer test
hook at the only level this subprocess can reach: not calling receiveFrame
for a while. The shared native pump keeps dispatching and coalescing into
this transport's own inbox regardless of whether the app reads, so not
reading is externally equivalent to not polling.
serve now also logs each broadcast frame's own TypeScript encoding as hex, so
the harness can compare it against the same frame's C#-side re-encoding
without importing the codec itself.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
NativeConnectorReceivesFromExternalManagedProvider -> NativeConnectorReceivesFromExternalProvider:
CULTMESH_NATIVE_EXTERNAL_ENDPOINT's peer is no longer necessarily the C#
managed provider now that Cut 5 adds a TypeScript one on the same lane; the
env var name stays, since it names the endpoint, not the peer.
Add ManagedConnectorReceivesFromExternalProvider, gated by the same env var:
dials an external provider with the plain managed connector and no
ValidateProviderCertificate (so acceptance runs on the endpoint's own
cert-sha256 pin alone, like an ordinary consumer), and prints the received
frame's CultMeshRealtimeWireProtocol re-encoding as hex for the harness's
golden-bytes comparison against the TypeScript peer's own encoding.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
cultmesh-quic-peer.ts serve opens a CultMeshQuicRealtimeProvider on the same
test fixture credential the unit tests use, prints {status, endpoint} in the
shape spawnServeProcess already expects, waits for a first peer before
broadcasting (a provider's broadcast to zero peers silently sends nothing,
same reasoning as the C# quic-realtime-serve peer's
WaitForFirstConnectionAsync), then broadcasts --frames frames at --delivery
with --interval-ms between them, stopping early on stdin EOF like the C#
peer's WatchStdinCloseAsync.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Reliable-ordered ordering in both directions, latest-only coalescing (a
synchronous back-to-back publish is deterministically coalesced, not by
network timing luck), a stalled peer's outbox holding at most one pending
frame per key across 200 broadcasts, a non-blocking broadcast, the uppercase
cert-sha256 pin, unreliable failing closed, and eviction on peer shutdown.
Runs against the real native bridge and Cut 4's real consumer connector, no
mocks; skips when CULTMESH_QUIC_NATIVE_DIR is unset, matching every other
QUIC test file.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CultMeshQuicRealtimeProvider (packages/cultmesh-ts/src/realtime-quic.ts) is
the StreamPixels role: listen() opens a native listener, computes an
uppercase cert-sha256 advertisedEndpoint (the Unity connector requires the
pin), and accepts peers into a Set of the same CultMeshQuicRealtimeTransport
Cut 4 already built for consumers. broadcast() mirrors
CultMeshQuicRealtimeServer.BroadcastAsync: reliable-ordered awaits every
peer's send_complete via transport.sendFrame directly (Task.WhenAll,
evict-on-failure); latest-only publishes into a new per-peer
CultMeshQuicRealtimeProviderOutbox and returns without waiting on any peer's
I/O. The outbox pumps sendFrame's existing fresh-stream-per-frame path one
frame at a time, so a stalled peer holds at most one pending frame per
(channel, body) key.
Extracted CultMeshRealtimeLatestOnlyCoalescer from the inbound path's
inline Map+FIFO bookkeeping (unchanged behaviour) so the new outbox reuses
the same coalescing idiom instead of a second inline copy.
Accepting a connection buffers native events behind a synchronous listener
registration (the pump dispatches strictly sequentially, so registering
before any await is what makes the async runtime-reference acquisition that
follows safe) and replays them once the peer's CultMeshQuicRealtimeTransport
exists; a handshake timer evicts a peer that never completes CONNECTED.
Wired into the CultMesh facade as createQuicRealtimeProvider.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The Soul gate refused Cut 4. A malformed peer frame killed the process,
release() was a use-after-free, a peer shutdown leaked, the session
manager never went offline, the race waited on the slowest candidate,
growth was unbounded, and eight spec rules had no test.
The interop failure turned out to be test infrastructure, not the
network: the harness gave the peer /dev/null for stdin, the peer
broadcast before any client connected, and .NET could not find
libmsquic.
Fix batch 1 passes 125/125, and the C#-provider-to-TS-consumer lane
passes. The cut grew instead of shrinking. A confirming Soul pass and
Cut 5 are running in parallel.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The PKCS12/DER pair had no record of how it was made or what it is. Documents
the CN, the openssl recipe that reproduces an equivalent pair, and that
nothing depends on today's exact bytes past what the tests read back out.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A non-literal first stdio element loses spawn()'s typed-tuple overload
entirely (verified on Yggdrasil: tsc failed on the two other, unrelated
serve helpers' stdout/stderr going nullable), not just for stdin's varying
element. Asserted at the one call site instead.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Finding 8(c): the dev image builds the Node bridge's own libmsquic copy into
a build-local directory (scripts/build-quic-native.sh), which is never on the
loader's search path. .NET's System.Net.Quic resolves "libmsquic" through
that path independently and failed with PlatformNotSupportedException before
this. The smallest honest fix: dpkg -i the same pinned 2.5.9 deb (same URL,
same SHA-256 as the build script) system-wide and run ldconfig, so both the
Node bridge's own copy and .NET's independent resolution find a libmsquic.
Changes the image; the new tag is cultlib-quic-native-dev:75cb89e73850
(first 12 hex of this file's sha256), built fresh on Yggdrasil.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Finding 8(a): every spawnServeProcess call used stdio ["ignore", ...], which
hands the C# quic-realtime-serve peer an already-closed stdin. Its own
WatchStdinCloseAsync (Program.cs:1261-1269) treats that as EOF and tears the
serve loop down at once, before it can send anything. spawnServeProcess gains
an opt-in `stdin: "pipe"` (default stays "ignore" for every other serve mode,
which never reads stdin), used only by the "CultMesh QUIC realtime" test; its
teardown ends that pipe explicitly before killing the process.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Finding 8(b): the serve mode broadcast immediately after printing its ready
line, before the harness's dial could reach it, so frames went to an empty
peer set and were silently dropped. WaitForFirstConnectionAsync polls
ConnectionCount, bounded at 30s so a peer that never gets dialed still exits
instead of hanging the interop lane forever.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Each rule below fails under its own mutation of the code fixed in the previous
two commits; verified by hand and restored, per the fix-batch brief, not
committed as a mutation harness (the operator retired that pattern).
- realtime-quic-native.test.ts: release() refuses a call with no outstanding
reference (fix 2).
- realtime-quic-consumer.test.ts: a stream-kind/delivery mismatch faults only
its connection while a second, independent connection through the same
pump keeps working (fix 1 / M4); a peer-initiated shutdown rejects a
pending receive and cleans up exactly once, so a further dispose is a
no-op and a fresh receive rejects instead of hanging (fix 3); a slow
latest-only reader accumulates at most one pending frame per key, and an
equal generation is dropped outright (fix 6, M10); a real connected
transport's isVerifiedFor rejects a generation other than the one it
connected with, and the session manager disposes a real (not stubbed)
transport whose generation no longer matches a freshly verified route
(M3); the advertised certificate pin matches case-insensitively (M2).
- realtime-quic-session-manager.test.ts (new): eviction of a dead cached
session and redial on the next connect, single-flight sharing one dial
across concurrent callers, refusing a transport that lands after dispose()
(all fix 4); the race returning on first success without waiting out a
slow candidate (fix 5); the maxRacedCandidates cap (M6); the protocolIds
and authorityRuntimeId route filters (M7, M8); the session key including
authorityRuntimeId (M9). Trust verification is skipped via
mode: "local-development" over loopback endpoints, since these tests are
about the session manager's own binding/racing/caching rules, not route
trust (already pinned by realtime-quic-consumer.test.ts's trust
negatives).
Subtraction: `nativeBridgeAvailable` was defined twice, disagreeing on
whether to check the dependency library too; both test files now import one
definition from test/support/native-bridge.ts (the stricter of the two).
Gap, reported rather than papered over: the stream-shutdown pruning half of
fix 6 (streamKinds) has no direct test, because the map is private and
nothing here exposes it; the committed test only shows a second frame still
arrives after the pruned stream's shutdown, which does not by itself prove
the entry was removed.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul's fix batch, findings 3, 4, 5 and 6:
- A peer-initiated CONNECTION_SHUTDOWN only set `disposed = true`, so
`dispose()`'s early-return skipped `release()` and `offConnectionEvent`: the
runtime reference leaked forever and a pending `receiveFrame` hung forever.
`disposed` and "already cleaned up" are now the same flag, run through one
`cleanup()` reached from `dispose()`, a peer shutdown, and the new `fault()`
(wired from fix 1's `onFault`). Every path releases the runtime reference
exactly once and rejects every pending and future receive/send with the real
reason instead of a generic "disposed" message.
- `CultMeshQuicRealtimeSessionManager.connect()` cached a transport forever,
even a dead one, and raced candidates with `Promise.allSettled`, so one dead
route delayed every connect by the full handshake timeout. It now: evicts a
session once its transport reports itself disposed (`onDisposed`, mirroring
`CultMeshSessions.InvalidateRealtimeSession`); shares one in-flight dial
across concurrent callers for the same target (mirrors C#'s `Lazy`); refuses
a transport that lands after `dispose()` instead of caching it; and races
each tier for the first verified success (mirrors `Task.WhenAny`), aborting
and disposing the losers via an `AbortSignal` threaded into
`connector.connect`, which now also honours a signal already aborted at
entry.
- `streamKinds` never pruned a finished stream (`STREAM_SHUTDOWN` fell into
`default`); it now removes the entry. The receive side queued every
latest-only frame in an unbounded array; it now mirrors the C# reference's
`CultMeshRealtimeInbox`: at most one pending frame per `(channel, body)`,
coalesced on arrival, while reliable-ordered frames still queue directly
(unbounded, matching the C# reference's own unbounded channel).
Subtraction: `trimmedOrEmpty` duplicated `cultnet-ts`'s exported `trimCSharp`;
callers now use the shared one.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul's fix batch, findings 1, 2 and 7:
- The pump's dispatch had no fault boundary: a malformed frame, a stream-kind
mismatch, or a throwing caller callback (certificate validator, waiter) threw
out of `dispatch` and killed the pump, taking the whole process with it.
`dispatch` now catches per-connection, shuts that connection down at the
native level, removes its listener, and hands the error to an optional
`onFault` callback registered alongside the listener. Every other connection,
and the pump itself, keep running.
- `release()` called `runtimeClose` before awaiting the pump's exit, so a
queued-but-not-yet-started `nextEvent.async` call could begin after close and
run against freed memory. It now marks closed, awaits the pump loop (it wakes
within its 250 ms timeout), and only then closes. It also refuses a release
with no outstanding reference instead of silently flooring at zero.
- koffi decodes a 64-bit `uint64_t`/`void *` as a plain `number` when it fits a
safe integer, not `bigint` (probed on the pinned 3.3.0 build against the real
bridge). Every declared type here promises `bigint`; `toBigInt64` is now the
single boundary conversion, applied to every 64-bit decode, so the mismatch
that made `event.code === BigInt(CULTMESH_QUIC_SEND_CANCELED)` never true is
fixed at its source instead of patched at that one comparison.
Subtraction: deleted the unused `lastStatus` export (wired but never called),
and collapsed the stacked eslint-disable-next-line pair over one line into one.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The Windows release configuration passes against the pinned OpenSSL
MsQuic. The DLL hash matches the pin, so the wrong-DLL load did not
happen again.
F5 was an operator fork. The operator allowed load on Starfire with
warning. The first Windows run under 16 burners failed 5 of 5. The
cause was the closerace fixture: its poll timeout (5 s) was shorter
than its own settle budget (30 s). The bridge was not at fault. After
the fixture fix at 44ab376, closerace passes 5 of 5 under load on both
Windows and Linux.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Under CPU starvation (16 burners on 8 logical CPUs), all 5 rounds failed
at iteration 0 or 1 with "256 of 256 pollers returned before the close
began, so they were never in the race". The fixture contradicted itself:
the asserts-on settle waits up to 30s (plus a 50ms remainder) for every
poller to be counted inside the library, but each poller's own
next_event wait used a 5s timeout, so the first pollers timed out for
real while the settle was still waiting on the last ones. The race never
ran; the fixture just gave up on itself.
kPollTimeoutMs is now derived from the settle budget it must outlast in
every build configuration (asserts-on's counted deadline plus its fixed
remainder, and the no-seam #else's longer fixed window), named side by
side so they cannot drift apart again. A long poll timeout costs nothing
on a passing run, since the close is what actually ends each wait.
Also give the fixture's own failure to settle a distinct report. If
counted_deadline expires before every poller is counted, CloseRaceOnce
now says "fixture could not settle: N of M pollers counted inside after
30000 ms" instead of proceeding to a close that blames the pollers for a
race the fixture never set up.
holdclose, latecall and holdtimeout use the same counted-deadline idiom
but are not the same defect: their pollers are parked behind
cultmesh_quic_debug_hold_calls, so they cannot time out for real while
the settle waits, and were left alone.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Program.cs is a top-level-statements file: every type declaration (the
existing ServeConfig, DialConfig and friends) lives after every top-level
statement and local function, all clustered at the end. The new record landed
between two local functions instead, which is CS8803: "top-level statements
must precede namespace and type declarations." Moved beside DialConfig, where
its neighbors already live.
Found by a direct `dotnet build` run on Yggdrasil after the interop test's own
error message was truncated by Node's console printing before reaching the
actual compiler diagnostic.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The interop lane's dotnet build FailFasts on startup in this minimal Debian
image: no ICU package, and .NET's globalization stack aborts without one
rather than degrading. The interop peer already parses every argument through
CultureInfo.InvariantCulture, so invariant mode costs it nothing; it is also
Microsoft's own documented fix for a minimal container, cheaper than pinning
and installing an ICU package by name.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Self killed a verify run after it sat idle for 25+ minutes past its last
printed result and root-caused it: SIGUSR2 diagnostics on the stuck process
showed an empty JS stack, every libuv handle unreferenced, and MsQuic's own
worker threads parked in epoll_wait — a native runtime that was opened but
never closed, so its registration and threads never tore down and the process
never exited on its own.
The cause was mundane and compounds badly: `RawQuicListener.open()` in
realtime-quic-consumer.test.ts opened a native runtime, then read a test
fixture whose path was wrong (see below) and threw — with no try/catch, so
the runtime it had just opened was never released. That test-file pattern ran
on seven separate tests in one process, each leaking its own reference to the
one process-wide shared runtime; by the time the file's tests finished, the
refcount had never reached zero and the live registration held the process
open indefinitely. `RawQuicListener.open()` now releases the runtime on any
setup failure before rethrowing.
Two more real bugs the same run's fixture-path failures exposed:
- Both new test files resolved fixtures at `join(__dirname, "fixtures", ...)`.
`__dirname` at runtime is `dist-test/test`, and binary fixtures are never
compiled or copied there — only `.ts` sources under `test/` and `src/` are.
Both now read fixtures from their real location, two directories up.
- The negative-grep test looked for a literal `.async(` call. The bugfix in
the previous commit references `nextEventFn.async` as a value handed to
`promisify`, never written with a trailing `(` at the reference site, so the
grep found nothing to check and failed by construction. It now matches
`.async` as a property reference, not only a call.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A `node -e` probe against the real bridge (loaded through koffi 3.3.0 on
Yggdrasil) showed the `_Out_` prototype marker does not auto-allocate storage
or reshape the return value: `openFn(null, null)` returned the plain scalar
`-1`, because the null "out" pointer is itself a bad call under the header's
own contract, and koffi never populated anything because there was nothing to
write into. `openFn(null, koffi.alloc("void *", 1))` returns the plain
`int32_t` status and the pointer decodes correctly with
`koffi.decode(buf, "void *")`, as a `bigint`.
Every out parameter is now `koffi.alloc(type, 1)` before the call and
`koffi.decode` after, inside `loadBindings`; the rest of the file is
unaffected, since it already treated ids as `bigint`. The runtime handle gets
its own `CultMeshQuicNativeHandle` alias (`bigint`) so its type is named
rather than smuggled through as `unknown`.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The wrapper cast the raw promisified koffi call to the same 4-argument public
NativeBindings#nextEventAsync type it was implementing, then called it with
six arguments (the two _Out_ placeholders included) — a type checked itself
into a contradiction tsc caught on a clean build. The raw call gets its own
type matching what koffi actually expects; the public wrapper still exposes
four.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- koffi's `_Out_` sugar still counts toward the call's declared arity in this
version (an arity check rejected the shorter call); pass `null` at each such
slot, since the value comes back through the named field on the result
object regardless of what was passed.
- The negative-grep test read `dist-test/src/...`, which does not exist:
`__dirname` at runtime is `dist-test/test`, and only compiled output is
mirrored there, so the original source is two levels up in `src/`, not one.
- Two generated Odin test roots shared one hardcoded key id
("odin-root-1"), so trust negative (c) had the trust policy's root looked
up successfully under the signer's own id and failed on signature mismatch
instead of "not trusted" — proving the wrong rule. Each generated root now
gets a distinct key id.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
koffi's own .d.ts resolves as an ECMAScript module under Node16 resolution,
which a CommonJS `typeof import("koffi")` reference and a `koffi.promisify`
property access (which does not exist) both choke on. Type the loaded module
as `any` and promisify `nextEvent.async` with `node:util` directly instead of
guessing at a koffi-native promise helper.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Cut 4's own npm tests and the C#-to-TS interop lane need to run in the same
container that builds the native bridge, and the image had neither runtime.
Smallest change: install Node 24 from NodeSource and the .NET 10 SDK from
dotnet-install.sh, both pinned by channel/major version the way the rest of
this image pins its toolchain by digest.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
tests/GameCult.Networking.InteropPeer gains mode quic-realtime-serve
(--port, --frames, --delivery, --interval-ms): creates a self-signed
certificate the same way CultMeshQuicRealtimeTransportTests.cs does, starts
CultMeshQuicRealtimeServer, and prints one ready line carrying the advertised
cultmesh-state+quic:// endpoint with its cert-sha256 pin, then broadcasts N
frames with increasing sequence, exiting after the loop or on stdin close.
GameCult.Networking.InteropPeer.csproj references GameCult.Mesh.Quic.
Discrepancy, kept: the cut map's worked line for the ready signal is
`{ endpoint: "..." }` alone; this repo's spawnServeProcess harness
(cultnet-interop.test.ts:2491) waits for a line whose parsed `status` is
"ready", the same convention every other serve mode here already uses. The
printed line is `{ status: "ready", endpoint: "..." }` — the map's shape plus
the field its own named mechanism requires — rather than inventing a second
readiness convention for one peer.
packages/cultmesh-ts/test/interop/cultmesh-quic-peer.ts: the TypeScript half,
`dial --endpoint --expect N`, connecting with CultMeshQuicRealtimeConnector
and printing the received frames as one JSON line (fields plus payload hex).
It reaches the connector through this package's own src/, never importing
cultmesh-ts itself, keeping the dependency direction the map states
(cultmesh-ts depends on cultnet-ts, never the reverse).
packages/cultnet-ts/test/interop/cultnet-interop.test.ts gains
"CultMesh QUIC realtime: C# managed provider and TypeScript consumer",
spawning the C# peer with spawnServeProcess and the TypeScript peer with
runJsonCommand, both established patterns in this file. Skips when
CULTMESH_QUIC_NATIVE_DIR is unset, the same convention the unit tests use.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
packages/cultmesh-ts/test/realtime-quic-native.test.ts: the FFI smoke against
CultMeshQuicNativeRuntime (open/close, a listener with a real PKCS12
credential, a full loopback connection delivering one raw frame end to end),
plus the negative grep that .async( appears only on nextEvent and no
koffi.register exists.
packages/cultmesh-ts/test/realtime-quic-consumer.test.ts: the TypeScript
consumer against a raw listener opened through the native runtime (connect
with the correct pin; refuse a wrong pin; refuse a missing pin with no
validator; a caller validator observes the target and the DER bytes; a late
older generation is dropped), mirroring
CultMeshNativeQuicRealtimeTransportTests.cs and
CultMeshQuicRealtimeTransportTests.cs. Then the five trust negatives through
CultMeshQuicRealtimeSessionManager with a static lookup source and routes
signed with node:crypto (P-256, IEEE P1363 via dsaEncoding): (a) expired
certificate, (b) unsigned remote route, (c) untrusted Odin root, (d) a route
that verifies but pins the wrong TLS certificate, refused by the connector
before isVerifiedFor is ever asked, and (e) a stale/replayed generation,
proven with a stub connector whose transport fails isVerifiedFor, asserting
the session manager disposes it and rejects the connect. There is no
session_open nonce on the QUIC plane (route generation plus the TLS handshake
is its freshness), so (e) is this shape rather than a literal nonce replay,
as the cut map records.
test/fixtures/quic-test.p12 and quic-test-cert.der: a self-signed ECDSA P-256
PKCS12 test credential (OpenSSL, empty password, 1088 bytes — the same shape
as the native bridge's own embedded test PKCS12) and its DER certificate, used
to open real raw listeners and compute the expected cert-sha256 pin at test
time rather than hand-transcribing it.
Discrepancy kept, reported: the C# reference's receive-side inbox also
coalesces a backlog of not-yet-drained latest-only frames of the same key to
just the newest; this transport applies only the generation-drop filter the
cut map names (":538-560") and queues every frame that passes it. The interop
lane's frames are far enough apart that this never matters in practice, but a
consumer that fell behind would see more history here than the C# reference
delivers.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
packages/cultmesh-ts/src/realtime-quic.ts adds, over the native binding:
- parseQuicRealtimeEndpoint, mirroring
CultMeshQuicRealtimeTransportConnector.TryParseEndpoint (:98-109).
- CultMeshQuicRealtimeTransport: sendFrame/receiveFrame/dispose/isVerifiedFor,
the stream-kind-vs-delivery consistency check (:519-524) and the
per-(channel,body) generation filter that drops older (producerEpoch,
sequence) (:538-560), both pinned against CultMeshQuicRealtimeTransport.cs.
Reliable sends reuse one persistent outbound stream behind a gate;
latest-only opens a fresh stream per send; unreliable fails closed. The
bridge owns wire framing both ways, so this file never length-prefixes
anything by hand.
- CultMeshQuicRealtimeConnector: the certificate decision on event 3 (a
caller validator, else SHA-256-of-DER against the endpoint's cert-sha256,
else reject), mirroring MatchesAdvertisedCertificatePin (:111-119).
- CultMeshQuicRealtimeSessionManager and CultMeshStaticRealtimeLookupSource:
binds candidates from authorityRoutes carrying cultmesh.realtime_state.v1,
verifies each with cultnet-ts's verifyAuthorityRoute before any connector is
asked, orders by connector then route priority, races at most
maxRacedCandidates per tier, requires isVerifiedFor after connect and
disposes a transport that fails it — the realtime path of
CultMeshSessions.cs's ConnectRealtimeAsync (:833-896), including its
session key verseId + "\u001f" + authorityRuntimeId (:46).
No Odin WebSocket rendezvous lives here (Q7, ruled A): only the
ICultMeshRealtimeLookupSource port and a static source.
index.ts re-exports the module and gains three CultMesh statics beside the
RUDP ones: parseQuicRealtimeEndpoint, createQuicRealtimeConnector,
createQuicRealtimeSessionManager.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Adds packages/cultmesh-ts/src/realtime-quic-native.ts: a lazy koffi binding
over the twelve v2 queue exports plus cultmesh_quic_last_status
(native/GameCult.Mesh.Quic.Native/include/cultmesh_quic_native.h). koffi is
required only inside loadBindings(), so a consumer that never touches QUIC
loads no native module. CultMeshQuicNativeRuntime is a process-wide,
reference-counted singleton running one event pump (a loop awaiting
nextEvent.async with a 250ms timeout, no koffi callbacks) that dispatches
decoded events to registered listener/connection handlers by id.
Also: add the koffi 3.3.0 dependency to cultmesh-ts's package.json, and switch
its tsconfig from classic Node resolution to Node16 so `cultnet-ts/contracts`
(a package.json `exports` subpath) actually resolves — a gap Cut 1 recorded
and left for this cut ("until Cut 4 changes it").
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Four Soul gates. The first three refused it, each on something real: the row
identity ruling implemented as two different rules that answered opposite ways
on the shape it was written for; both runtimes faulting on an attacker-supplied
cursor through mirror-image overflows; a parity promise neither built nor
listed as outstanding; and a guard deleted on my ruling that took a legal
document shape into an untyped exception.
What lands is one rule for reference resolution in both runtimes, derived from
the wire rather than from preference, with the shared-key vector that never
existed. Malformed cursors reach the wire typed. Every comparator component
that decides output dies under mutation in both runtimes, and the one that
could not be distinguished by any input was deleted rather than recorded as
equivalent. The snapshot_response_raw.v1 envelope and its nested record types
are byte-identical across runtimes at populated and absent optionals, asserted
independently on each side.
The claim is exactly that, and no larger: byte parity for those types and
semantic parity over the named field subset. Whole-message assembly through a
single rows-to-bytes path is its own cut.
Follow-ups stay open and recorded: the C# Stryker survivor list untriaged,
contracts.rs and schema_discovery.rs unmeasured, the --since crash unrooted,
and three Caching tests that need a non-root container to prove and must not
be made to pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Soul found this while confirming batch 8. The Survivor-5 header in the same
file was updated when the component was deleted and this one was not, so it
described a comparator that no longer exists.
Self edited this directly rather than dispatching Hands for a single stale
sentence. Naming that because Self editing code is a boundary worth saying
out loud when it is crossed, even for a comment.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Both runtimes now assert all four cells of the wire vector, and the new C#
constant fails when a single byte is flipped, so it is not decoration.
R-BA came back as a deletion with a better argument than my ruling carried.
Both resolvers are pure functions of the citer's schema, the role and the
record key against a fixed candidate index, so once those tie in the
comparator the resolved row including its schema is forced identical. That is
structural rather than semantic: it does not rest on schema-authoring
discipline, which is what my version rested on and what has misled this cut
twice.
All four pre-merge items are closed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Soul could not construct a selection where To.SchemaId decides the
EdgesFor/page_edges order once From, Role and To.RecordKey already
tie, but flagged that as a claim, not a proof - the same shape of
argument that was wrong twice before on this comparator (R-AT).
Traced instead of asserted: TryResolveReferenceTarget (C#) and
resolve_reference_target (Rust) resolve a reference purely from the
citer's schema, the role, and the record key, against the evaluation's
own fixed candidate index (byKey/by_key) - no edge- or caller-specific
input steers which candidate wins. Once From and Role tie (pinning the
citer's schema and the declared target type/leaves) and To.RecordKey
also ties, the resolved row - schema included - is forced identical by
construction, in both runtimes, every time. No mutation can distinguish
it because no input can, so it was deleted rather than recorded as
"equivalent": a component nothing can distinguish is a line every
future reader must reason about for no behavioural gain.
Deletes the To.Descriptor.SchemaId/to.schema_id() ThenBy/then_with from
CultNetSelectionEvaluator.EdgesFor and selection.rs's page_edges sort,
and updates the comments (in-code and the Survivor 5 test block) that
described the old five-key chain.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The 246-byte all-absent Edge vector (SM-10) existed only on the Rust
side (selection_wire_parity.rs's CSHARP_EDGE_HEX). The all-absent
header gets incidental cover because it is nested inside the 376-byte
envelope test; the edge has no such cover, so a C# edge encoder that
resumed omitting a null payload/payloadEncoding would still pass every
other test here - exactly the asymmetry R-AP existed to kill.
Adds EdgeAllAbsentMatchesTheConstantCommittedIntoTheRustTest, asserting
the same constant against MessagePackSerializer.Serialize. Flipping one
hex digit of RustEdgeAllAbsentHex fails the new test only, confirming
it is not vacuous.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul found the false equivalence argument still standing at its original site
after I deleted the copy batch 6 had restated. Both read as live instructions
and were obeyed for two passes: nobody chased those mutants because the map
told them not to.
Struck rather than edited, so the record shows what was believed and for how
long. The closing line of R-AC survives on its own merits: a survivor
correctly identified does beat a test written to move a number, which is
precisely why identifying one takes a construction attempt rather than an
argument.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every comparator component that decides output now dies under mutation in both
runtimes, and the new fixtures are killers rather than coincidences. The
headline claim is two-sided: each runtime serializes and asserts
independently, and every C# vector fails when a single hex digit of its
expected constant is flipped.
Soul found one real hole. The all-absent Edge vector exists only in Rust,
because the all-absent header is covered incidentally by the envelope and the
edge is not. A C# encoder that resumed omitting a null payload would pass
everything. Five lines closes it, so it closes before the merge rather than
after: shipping over a known asymmetry with a note promising to return is the
thing this cut exists to prevent.
Two misses are mine. R-AT asserted to.schemaId dies alone, which Hands never
claimed and which is not true; it is now construct a killer or delete the
component, because a line no input can distinguish still has to be reasoned
about. And I deleted the false equivalence argument where batch 6 restated it
while leaving the original standing as a live instruction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
All three surviving comparator components now die in both runtimes, the claim
finally has the home the map said it had, and the whole-message vector is
green at 376 bytes. The thin evidence is no longer thin: a populated tags
array and a populated payload both come out byte-identical, which was the
named risk and did not materialise.
Hands also found that Soul's counterexample, taken verbatim, does not kill the
mutation it was built for. It survives in both runtimes because select's push
order happens to match alphabetical sort for that naming, and the two runtimes
push in opposite orders. Not an unreachable branch: a coincidence, the third
of these in three days, and this one was caught while executing rather than by
a later pass.
My byte counts in the landing note were wrong, 86 and 99 against the real 228
and 246. The committed constants matched the real bytes all along; only the
prose beside them was wrong, which is the same shape as the two corrections
before it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Soul had this green in the rig at 376 bytes and called it "blocked on
unbuilt page-assembly ownership" - the wrong thing: only the two
*evaluators* producing the same page from the same rows needed that move
(a separate claim, R-AR); the two *encoders* producing identical bytes for
the same page value never did. Land the whole
cultnet.snapshot_response_raw.v1 envelope, byte-identical between Rust's
encode_cultnet_message_to_vec and C#'s CultNetSnapshotResponseRawV1Message,
on both runtimes.
The existing header/edge vectors also proved nothing about tags
(string[] vs Vec<String>) or payload (byte[] vs serde_bytes): both were
absent at every value pinned. Add a header with tags populated and an
edge with a payload populated, captured the same way against a real
MessagePack C# 3.1.7 run on Yggdrasil. Neither diverged - the two
mappings agree byte-for-byte.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Section 9 of docs/cultnet-selection-cut.md pointed here for the claim;
this file never mentioned typed selection at all. Add the row, in the
words from fix batch 7's R-AY ruling.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A shape that used to register and work - a data member aliased Owner
beside a reference member named Owner - is now refused by D10b (R-AL).
Soul's merge-gate third pass flagged this as a breaking change with no
note; the error message and its owning comment now say so.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The zebra/alpha pairing that kills Rust's drop-both-to-components mutant
does not kill it in C#: EvaluateAll concatenates citesEdges before
citedEdges (the opposite of cultnet_rs's single cited-then-cites Vec), so
that pairing's push order already matched the correct sort order here too
- just for the opposite reason. Swap to alpha (self) / zebra (target),
the pairing that actually disagrees with this runtime's own push order.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The same live mutant Soul found in Rust (dropping to.schemaId/to.recordKey
from the edge comparator) survives in C# too - both runtimes share the
comparator's shape, not a rule proving it. Land the C# mirror of Soul's
self-citing-peer counterexample from the rig, renamed so the self edge's
push order and the correct sort order actually disagree (the rig's own
"hub"/"other" naming left the drop-both mutation surviving by coincidence
of push order matching alphabetical order - see the Rust commit on this
branch for the construction).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul's merge-gate third pass (SM-7) found three live mutants: dropping
to.schemaId/to.recordKey from the edge comparator (both runtimes) and
dropping recordKey from Rust's row comparator (named a pass ago, never
touched). Hands' earlier argument that no selection ties from+role while
leaving to free to vary was false - a selection carrying both cites and
cited mixes their edges into one bucket, and a self-citing peer produces
exactly that tie. Take the counterexample from Soul's rig rather than
rebuilding it, and add the missing row-comparator recordKey tiebreak test.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul built the counterexample. The claim that no construction gives two edges
with tied From and Role and differing To reasoned about each hop separately,
and a selection carrying both hops mixes their edges into one bucket. A row
that self-references through the same role it uses to cite a peer produces
exactly that pair. It runs in both runtimes. Three comparator survivors the
last gate named are still alive, and batch 6 added no Rust row-order test at
all.
Two of my own words were wrong and both are corrected here. Nested wire-type
byte parity overstated evidence gathered at a single all-optionals-absent
value. And calling whole-message parity blocked on an ownership move ran two
independent things together: the encoders agreeing on bytes needs no ownership
move, and Soul wrote it this afternoon at 376 bytes identical. Calling it
blocked made the cheapest available evidence look expensive, which is how it
slipped six passes while sitting there for the taking.
The claim also had no home. The file section 9 names as carrying it contains
no mention of typed selection and never did.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The brief named ygg-verify.sh without its path where every earlier brief gave
it in full. Hands could not find the script, verified the whole batch on the
workstation instead, said so plainly rather than claiming the bar was met, and
left the branch unpushed so nothing looked merge-ready. This batch still owes
its Yggdrasil run.
The regression is backed out and the real fix went to the owner: registration
now refuses a data member whose bare name collides with a reference member's
alias, which is the collision D10 never checked. Writing R-AM's test found a
second defect, the backstop logging DynamicInvoke's wrapper instead of the
fault it was meant to surface.
R-AP's encoding turned out to be C#'s existing behaviour, with the whole gap
on the Rust side omitting keys for absent optionals. Two nested types are now
byte-identical against real C# hex. The half that would move page assembly
into the evaluator was not built, and Hands stopped and explained why rather
than reporting the ruling closed. That half is its own cut, because it changes
two public signatures, every caller and the harness.
So the cut's honest claim is nested wire-type byte parity plus semantic parity
over the named subset. Whether that is enough to merge is the operator's.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Decided the payload shape: a map with camelCase keys, fields in declared
order, every optional field always present and nil when absent - never
omitted. C#'s [MessagePackObject]/[Key("name")] map formatter already does
this by default (confirmed byte-for-byte below); Rust's `to_vec_named`
encoding for these four types used `skip_serializing_if = "Option::is_none"`,
which omits the key instead - the gap the map cited ("C# writes nine fields
with nils, Rust omits absent ones").
Removes skip_serializing_if from Edge, RawDocumentHeader,
SelectionDocumentRecord and SelectionPage, keeping #[serde(default)] for
decode-side robustness. Selection (the request type) is untouched - out of
this ruling's scope, and the map already confirmed the request side agrees
between runtimes.
Adds tests/selection_wire_parity.rs: RawDocumentHeader and Edge, each
carrying several absent optionals, encoded via to_vec_named and compared
byte-for-byte against hex captured from a real MessagePack C# 3.1.7 run
(same pattern as error_message.rs's csharp_hex vectors) - both now agree
exactly.
What is NOT done, and why: R-AP also called for "Select gains projection,
so the thing the vectors test is the thing that decides." In C#, projection
(header vs document, ToRawRecord/ToEdge/FromRecord) is owned entirely by
CultNetDocumentRegistry.SelectPage/SelectAll (GameCult.Networking), not by
CultNetSelectionEvaluator.Select/Page - the evaluator returns raw Rows/
EdgeMatches and knows nothing about a document's payload serializer, which
only the registry's bindings own. Moving page assembly into the evaluator
without threading a caller-supplied serializer (the same pattern ordinalOf/
shardIdOf already use) would give a schema-generic evaluator per-binding
wire knowledge it should not own; threading a serializer callback in is
buildable but changes SelectPage/SelectAll's own signature and every
caller, plus the parity-vector harness's construction - a materially larger
change than this pass's remaining scope. No vector yet exercises the real
CultNetSnapshotResponseRawV1Message/select_page output byte-for-byte; only
the two nested wire types in isolation. Flagged for the operator per the
brief's own stop condition.
R-AQ, updated: type-level byte parity for the nested wire shapes (Edge,
RawDocumentHeader) is now real and vector-tested. Whole-message byte parity
for cultnet.snapshot_response_raw.v1 is still not - the smaller claim (this
cut's page bytes agree only where R-AH's fixed-shape rule reaches, and
projection ownership is still split between the evaluator and the registry)
is the true one until "Select gains projection" actually lands.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Adds citer-both-in-target to the shared fixture, the first row in file (and
row-order) to cite "dup-key" through Design - both leaf_a and leaf_b at that
key are in Design's declared target, unlike citer_narrow's narrow_ref above
it (only ever one in-target candidate). Adds the matching vector case to
both runtimes' case lists and regenerates both written vector files
(CULTNET_WRITE_VECTORS=1), so cross-runtime agreement on this shape is
pinned by the parity check, not just asserted per-runtime.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-AN: the existing edge-order test ties From and To across its two edges,
so it only pins from.record_key as sole discriminator - dropping role
entirely still survives (Soul's SM-4). Two new tests isolate the other two
components of EdgesFor's comparator: role (same citer, same target, two
roles - Citation.role: None matches any reference) and from.schema_id (two
different-schema citers sharing one record key and role). Both confirmed to
kill their own dropped-clause mutation by hand (reverted after).
R-AO: shared_key_with_two_in_target_candidates_resolves_to_whichever_the_-
caller_listed_first pins "first row in the caller's order wins" - R-AF's
existing vector only covered one-in/one-out. Confirmed to kill a
last-wins mutation by hand (reverted after). Also extends the shared vector
case list with shared_key_design_resolves_the_first_in_target_candidate_-
in_row_order (paired with the fixture/vector commit).
To.SchemaId/To.Key were re-checked per Self's ruling: within any one
page_position bucket, `cites` always ties every edge's To to the query's
own single target, and `cited` always ties it to the anchor row's own
identity - the only way for To to vary at a tied page_position is a mixed
cites+cited hop on one row, which always ties or untangles via From first.
No construction was found where To is the sole discriminator; the earlier
"equivalent within one hop direction" claim holds. No code change, no
artificial test.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- R-AL: DataMemberNameCollidingWithAReferenceMembersAliasIsRefusedAtRegistration
pins D10b's registration refusal. ReferenceMembersExcludesADataMember...
pins the restored evaluator guard directly (hand-built descriptor via
reflection, since D10b now closes the registry door this shape used to
slip through).
- R-AM: CultNetDatabaseServer_SnapshotV1Handler_CatchesAnUntypedFault... and
RudpCultNetSchemaServer_DispatchBackstop_LogsEveryUntypedHandlerFault pin
the handler's catch and the backstop's logging, each against a real
untyped fault.
- R-AO: Evaluator_CitedWithTwoInTargetCandidates_ResolvesToWhicheverThe-
CallerListedFirst pins "first row in the caller's order wins" when two
in-target rows share a key, in both row orders.
Every test in this commit was checked against a hand-reverted mutation of
its own production fix and confirmed to fail before being confirmed to pass
against the real fix.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
HandleSnapshotRequestV1Async only caught the three typed selection
exceptions, unlike its siblings HandleShardLogRequestAsync/HandlePutAsync,
which catch Exception and answer a CultNetErrorMessage. An untyped fault
(e.g. a null Selection) escaped into the RUDP dispatch backstop, which
swallowed it with no log - the same backstop that hid SM-6's
ArgumentNullException regression for a whole batch.
- HandleSnapshotRequestV1Async now catches Exception like its siblings and
answers a CultNetErrorMessage.
- RudpCultNetSchemaServer gets a Logger (ILogger/NullLogger, same pattern as
Server/Client), and DispatchAsync's backstop always logs the fault it
swallows, unwrapping DynamicInvoke's TargetInvocationException wrapper so
the logged message is the handler's real exception, not the generic
invocation-wrapper text.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CultNetSelectionEvaluator.ReferenceMembers dropped the `!member.IsReference`
guard on the premise that a data member's role string could never coincide
with a reference member's alias - false. CultCache's D10 only rejects a
collision between two *aliased* members; an unaliased data member's bare
name was never checked against another member's alias. A data member named
`Owner` plus a reference member aliased `Owner` registered fine, then faulted
every hop through it: ReferenceMembers yielded the data member, ReferencesOf
matched it to the reference member's edges by role string, and
`member.TargetType!` was null.
Two independent fixes, both regression-tested:
- Restore the guard, delete the false comment. Pinned directly by hand-
building the colliding descriptor via reflection (bypassing the registry,
since D10b below now refuses that shape at the only production door).
- D10b: registration now refuses a plain member whose bare name collides
with a reference member's alias in the reference-role namespace
(IndexAlias ?? MemberName), not just alias-vs-alias.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-AK was mine. I called a guard duplicate authority and ordered it deleted;
Hands verified, got it wrong, and left a comment claiming an alias uniqueness
property that the source contradicts. A legal document shape now takes every
hop path into an untyped exception, the peer gets no typed refusal, and the
batch's own new dispatch backstop swallows it without a log. Restoring the
guard leaves the suite green, which is the other half of the problem.
That is the second redundancy ruling of mine to be wrong. The first was caught
by Hands; this one was not. A deletion argued from a property has to be probed
at the shape that would break it.
R-AF is genuinely settled and Soul confirmed it harder than the probe did: the
reference type has nowhere to put a schema id, so no formatter or version can
introduce one. Both runtimes now agree on every shape that split them,
including the role-filter reordering no ruling covered.
R-AH has slipped five passes and cannot be built as I ruled it, because the
ruling never fixed map encoding, field order, or whether an absent optional is
nil or omitted, and because C# has no page type at the layer the vectors
test. This batch decides all four. Until it lands, the map now claims semantic
parity over a named field subset rather than byte parity, because byte parity
is not true today and the smaller honest claim is worth more.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
canonical_number went from 29 missed to 4, all four hand-traced as equivalent
rather than assumed. is_canonical, the Q-J door whose every branch mutation
was surviving, had never been called directly by any test; it is now.
The finding that outlives the batch is infrastructural. The same commit gave
different missed and timeout sets across three -j4 runs, while every -j1 run
reproduced the same result, most likely the shared cargo target directory
under concurrent mutant builds. That makes every concurrent mutation figure in
this campaign provisional, Stryker's --concurrency 4 runs included. Killed
mutants stay killed; survivor lists from concurrent runs cannot be trusted as
complete, so the rulings they informed rest on the hand-confirmed evidence
beside them rather than on the counts.
A third test in two days passed by coincidence: render_f32's sign-bit mutant
survived because 1.0 and 2.0 happen to have bit 0 matching their sign bit.
contracts.rs and schema_discovery.rs remain unmeasured. A partial run showed
real misses across the validate family before it was killed; that signal is
recorded so the next pass starts from it. Queued rather than run, because the
merge gate is a Soul pass on batch 5, not a sweep of an adjacent file.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Same gap as render_f32's sign-bit test (previous commit): render_f64's
sign extraction has the identical shape and the identical coincidence
(-1.0_f64/-2.0_f64's bit 0 happens to be 0, matching their real sign
bit's positive-looking value under a wrong-direction shift).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Every existing render_f32 test (including the two added in the previous
commit) used a positive value, so a mutant that shifts the sign-bit read
the wrong direction (bits << 31 instead of bits >> 31, reading bit 0
instead of bit 31) went unnoticed: 1.0_f32 and 2.0_f32 both happen to
have bit 0 equal to 0, same as their real (positive) sign bit, so the
mutant computed the same "not negative" answer by coincidence.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The first canonical_number mutation pass (previous commit) closed every
survivor except four in render_f32 (biased_exponent == 0 flipped to !=,
the subnormal exponent's -149 losing its sign, and the normal path's
implicit-bit | flipped to ^/&). The existing round-trip tests parse the
rendered string back through f32::parse and compare bits, which can
still agree with a wrong bit-extraction when the wrong value happens to
parse back to the bits it was carelessly derived from; none of them
assert on the rendered string's shape or magnitude directly.
Add a subnormal-shape test (2^-149 is "0." plus 149 fraction digits, not
an astronomical integer and not one exponent off) and a normal-path test
where the fraction bits are identical to zero's (1.0 and 2.0) so only
the implicit leading bit the | folds in carries the value.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
cargo-mutants over selection.rs found every branch mutation in
is_canonical surviving (the Q-J door was reachable only through the
handful of spellings validation_refuses_non_canonical_number_spellings
happened to cover), plus decompose/pad_right/canonicalize_decimal_digits/
big_mul_small/big_to_most_significant_first each replaceable with a
trivial stub while the suite stayed green.
Add tests that call is_canonical, decompose, pad_right, compare,
canonicalize_decimal_digits, big_mul_small and big_to_most_significant_first
directly (the nested test module already reaches private helpers via
super::), pinning exact output at the inputs that exercise each branch,
comparison and carry step named in the mutant list rather than only the
few spellings the field-validation door happened to already cover.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-AF was decided by dumping real bytes rather than by preferring an
implementation. A reference serializes as a bare msgpack string with no schema
id anywhere, so C#'s rule was right and Rust adopted it. The shared-key vector
that never existed now covers both directions, including that row order does
not change the answer.
Two findings came out of the work itself. Rust resolved and target-checked
every declared reference before filtering by the requested role, where C#
filters first: a live divergence no ruling had named. And the existing
row-tiebreak test was degenerate, passing with the tiebreak reverted and
swapped, purely because of which real schema hash happened to sort first. It
now derives its keys from the ids discovered at runtime.
Both cursor overflows are fixed, each runtime fed the other's proven input,
with a backstop catch so nothing untyped reaches the poll loop.
Two things are owed rather than done. The diff-scoped Stryker run failed on a
tooling bug that was worked around instead of root-caused, so 47 untriaged
survivors are inherited. The dispatch backstop has no integration test.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
R-AG. ReadLengthPrefixedFields tested pos + length > bytes.Length in int
math; a length prefix near int.MaxValue overflowed the unchecked add,
slipped past the guard, and Encoding.UTF8.GetString threw
ArgumentOutOfRangeException instead of this method returning null the
way every other malformed shape already does. Promotes to long before
the add, which cannot itself overflow (both operands stay far under
long's range), so the guard sees the true sum.
CultNetRudpSchemaServer.DispatchAsync gains a catch around the handler
invocation: HandleSnapshotRequestV1Async already answers
cursor_invalid/selection_invalid/reference_outside_target for its three
typed exceptions, but nothing stopped an untyped one from an
unanticipated bug reaching DynamicInvoke's caller and stalling the poll
loop for every other peer. Answers nothing on the wire (an exception
this generic carries no typed refusal to report) and drops the one
malformed dispatch, the same as an unregistered handler already does.
Hostile-cursor tests (CultNetSelectionFixBatchTests.cs) feed
CultNetSelectionCursor.Parse a length prefix near int.MaxValue (C#'s own
proven overflow), near usize::MAX (Rust's proven overflow, fed here too
per the ruling), wider than any integer type, and negative-looking -
each asserted to throw the typed CultNetSelectionCursorException.
R-AK. Deletes ReferenceMembers's `!member.IsReference` guard.
CultCache.ReferencesOf (CultCache.cs:262-278) already searches only its
own reference members for a role match and returns empty for anything
else; index aliases are unique per descriptor, so a data member's own
role string can never coincide with a reference member's. Verified
before deleting: traced ReferencesOf's role lookup directly rather than
trusting the ruling's premise. A second copy of the same exclusion here
was duplicate authority, not a second decision.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-AF: FixtureRow's references (and every fixture/hand-built row that
constructs them) carry a bare target key now, matching Row::references's
new contract. Adds citer_narrow() (the shared fixture's real schema id,
distinct from this file's own local NARROW_CITER) to the fixture
machinery, and two direct behavioural tests
(shared_key_resolves_to_the_row_inside_the_declared_target_in_either_row_order,
shared_key_refuses_when_the_only_candidate_is_outside_the_declared_target)
pinning both shapes from Soul's table at the select() level, plus a
shared-fixture vector case for the accept shape and an
evaluation_refusal_cases entry (expected_evaluation_refusal, read by the
matching C# addition) for the refuse shape.
Rewrites cites_refuses_a_row_carrying_an_unrelated_edge_outside_its_declared_target
(renamed cites_refuses_an_out_of_target_edge_under_the_same_queried_role):
its original shape queried one role while expecting refusal from an
edge under an unrelated role, a stronger guarantee than C#'s own R-T(F15)
test gives (same role, two elements) - found because it broke once
matches_citation's role-filter order was fixed to match C#. The new
version mirrors C#'s actual test shape.
R-AG: hostile cursor tests feeding Cursor::parse a length prefix near
usize::MAX (Rust's own proven overflow), near i32::MAX (C#'s proven
overflow, fed here too per the ruling - neither side's tests are
evidence about the other), a prefix wider than any integer type, and a
negative-looking prefix - all under std::panic::catch_unwind, asserting
a typed CursorInvalid and never a panic.
R-AJ: edge_order_is_from_key_before_role_when_both_vary_at_the_same_anchor
(a both-hops selection with a hub/other citer pair whose keys and roles
are deliberately anti-correlated, so a Role-first mutant and the real
rule disagree) and
row_tiebreak_orders_by_schema_id_before_record_key_when_both_vary_at_a_tied_ordinal
(the same anti-correlation trick applied to order_rows, computed against
the real leaf_a/leaf_b ids at runtime).
FixtureRow gains #[derive(Debug)] - needed for an assert message on the
new evaluation-refusal-case writer, harmless elsewhere.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-AF: registers ParityCiterNarrow (citer_narrow's C# type, [CultIndex
"narrow_ref"] so its role matches the fixture/Rust spelling), adds
Keys-scoped vectors pinning both shapes from Soul's table - a shared key
where the only candidate is out of target refuses evaluation-time
(expectedEvaluationRefusal, a narrow addition to the vector schema
distinct from the door-only expectedRefusalField - not R-Z/R-AH's
page-bytes harness), and one where an in-target row shares a key with an
out-of-target row and must still resolve.
Landing these exposed a real, pre-existing cross-runtime scope
difference in R-W unrelated to R-AF: Rust checked every declared
reference on a row regardless of the citation's requested role;
C#'s ReferenceMembers(descriptor, citation.Role) narrows by role first.
Fixed on the Rust side (previous commit); hop_by_role_cites_exact_id and
cites_target_by_name_alias keep an explicit Keys scope here anyway as
belt-and-suspenders.
Also corrects S-12's last stale comment: cites_target_by_name_alias's
"Confirmed cross-runtime defect" note was already false - the alias port
fixed it - and still called it confirmed.
R-AJ: CultNetSelectionSurvivorTests.cs's row-tiebreak fixture
(Evaluator_BreaksAnOrdinalTieBySchemaIdThenRecordKeyInBothDirections) was
degenerate - probed by hand-reverting the source and swapping the
tiebreak's two ThenBy calls, it still passed, because whether the old
3-row fixture discriminated the swap depended on which of SelLeafA's/
SelLeafB's real schema ids happened to sort first, and on this repo's
ids it didn't. Rewritten to assign row keys relative to the schema ids
discovered at runtime (low-sorting schema gets the high-sorting key and
vice versa) so the two candidate orderings are always the exact reverse
of each other; reprobed the same mutant afterward and it now fails.
Adds a direct EdgesFor test (constructing FullEvaluation/EdgeMatch by
hand) pinning "From.Key before Role" in the five-key edge tiebreak - the
two existing EdgesFor tests each tie one of those two components, so
neither could tell a Role-first mutant from the rule.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The fixture had zero duplicate record keys across its ten rows and no
ordinal tied across two different schemas - both shapes this fix batch's
rulings needed and neither runtime's parity vectors ever covered.
R-AF adds a "citer_narrow" schema (real SHA-256 id, computed the same way
CultDocumentRegistry does - a probe run against a type matching the
eventual C# type's exact namespace/nesting confirmed it) whose one
reference targets leaf_a only, plus rows for both shapes from Soul's
table: a lone out-of-target row at a key (refuses), and two rows sharing
a key where only one is in-target (accepts, resolving the in-target one
regardless of row order).
Also drops "targetSchema" from every reference entry: it was never
anything but documentation (SelectionParityVectorTests.BuildDocument and
cultnet-rs's build_row both already read only targetKey) and keeping a
field the wire does not actually have would contradict the rule R-AF just
established.
R-AJ adds a tied-ordinal pair across leaf_a and leaf_b, with keys chosen
opposite to the real schema ids (probed) on purpose, so a schemaId-first
order and a recordKey-first order are forced to disagree regardless of
which real id happens to sort first - see the row-tiebreak test this
enables.
selection-vectors.cs-written.json and selection-vectors.rs-written.json
are regenerated from the new fixture (CULTNET_WRITE_VECTORS=1 on both
sides) and cross-checked in both directions locally.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-AF. Probed the real wire bytes a stored CultRecordRef<T> produces
(GameCult.Caching.MessagePack.CultRecordRefFormatter<T>.Serialize writes
only value.Key.Value - a probe against a real document confirmed the
serialized "Design" field is the bare msgpack string "a-eq", nothing
else). The wire carries no schema for a reference; C#'s rule was right
and Rust's exact-(schema,key) lookup assumed a schema id the edge never
actually carries.
Row::references() now returns (role, target_record_key, payload) - a
bare key, matching C#'s ReferencesOf. matches_citation/build_incoming_index
resolve through the new resolve_reference_target, which mirrors C#'s
TryResolveReferenceTarget: search every row sharing the key (by_key is
now keyed by record_key alone, grouping rows across schemas) for the one
whose schema falls inside the declared leaf set, refusing deterministically
(by schema id, code-point order) when none do. Folds D9's leaf-set check
into resolution itself rather than a second ensure_within_declared_target
pass - deleted, since it also dropped the old bypass-on-empty-leaves
behaviour C# never had (empty leaves always refuses when a candidate
exists).
Also moved the citation's own role filter before resolution in
matches_citation (mirroring C#'s ReferenceMembers(descriptor,
citation.Role), which narrows to that role's members before R-W's
per-edge check ever runs) - found while landing R-AF's vectors: Rust
was checking every declared reference a row carries regardless of the
requested role, a stronger and different guarantee than the C#
reference actually gives.
R-AG. read_length_prefixed_fields used unchecked usize addition
(pos + len) to bound a length prefix parsed from the wire; a hostile
prefix near usize::MAX overflowed and panicked. Uses checked_add now,
returning None (which Cursor::parse already turns into a typed
CursorInvalid) the same way every other malformed shape here already
does.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
S-1 is not closed. R-V became two different rules: C# resolves an edge's
target by record key plus the declared leaf set, Rust by an exact schema-and-
key lookup. They give opposite answers on the exact shared-key shape the
ruling was written for, and no vector covers it because the fixture has no
duplicate keys at all. That is why it survived four passes.
Both runtimes fault on an attacker-supplied cursor, on mirror-image inputs.
C# overflows an int and lets an untyped exception reach the server's poll
loop; Rust overflows a usize and panics. Each holds exactly the hole the other
closed, which is what happens when a rule is hardened twice independently and
neither run is fed garbage.
R-Z was neither done nor reported as outstanding. I recorded batch 4 as
closing its rulings without checking that one, so the tracking failure is
mine. It also cannot be built as written: the two runtimes have no common
wire type for a page, so the comparison needs a shape decision first. I have
ruled one and flagged it for review rather than letting the vectors harden
around an unexamined choice.
Two further findings worth the pass. The Rust vocabulary has 59 live mutants
nobody has looked at, and about 35 of them sit on canonical_number, whose door
survives every branch mutation: the byte-for-byte parity rule is pinned by
nothing in Rust while the C# side was blocked at 66% over the same class. And
Soul settled the out-of-bounds read I had ruled most urgent: it is not a
defect, the mutant was a coverage gap.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
R-AB worked and cost two fixes: the named include_str! escape, then a second
one in the fixture loader that kept the crate unmutable even after the first.
cultnet-rs can now be mutated at all for the first time. Scoped to the diff:
23 tested, 23 caught, nothing missed.
The price is that twenty-two schemas now exist twice, guarded by a drift test
rather than by structure. It drifted within hours: the rebase onto the C#
side's R-X changed a schema and the vendored copy did not see it. The drift
test caught it. The cleaner shape, a build script staging into OUT_DIR, fails
for exactly the reason the include_str! did, because cargo-mutants copies the
package tree. Publishing the contracts as their own crate would fix it and is
out of this cut. So the duplicate is accepted on the record, in the follow-ups
rather than the win column.
S-2 turned out to be C#-only. Rust already checked the declared target before
filtering by key, so the divergence was one-sided.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Rebasing onto the C# side's landed R-X work (fix batch 4) picked up
changes to contracts/cultnet/cultnet.error.schema.json that predate the
vendored copy this branch added. The drift test added for R-AB caught
it immediately: re-copied, no other vendored schema or the vendored
fixture differed.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
actually runs
The schema_discovery.rs include_str! escape was not the only one:
tests/selection.rs's fixture() read
../../contracts/cultnet/interop/selection-vectors.fixture.json, the
same three-levels-up path shape, and 37 of 45 tests panicked on it
under cargo-mutants' isolated copy - confirmed on Yggdrasil after the
schema_discovery.rs fix alone still landed "cargo test failed in an
unmutated tree, so no mutants were tested".
selection-vectors.fixture.json is a checked-in input (not a generated
artifact), so it gets the same treatment as the 22 schemas: a
byte-identical vendored copy under
packages/cultnet-rs/contracts/cultnet/interop/, a drift test
(fixture_contract_matches_the_canonical_source, same skip-when-absent
shape as schema_discovery_contracts.rs), and the loader reads the
vendored copy unconditionally.
selection-vectors.cs-written.json is different in kind: it is the C#
reference's own generated output, produced only by a real cross-runtime
checkout running SelectionParityVectorTests with
CULTNET_WRITE_VECTORS=1. There is nothing to vendor - an isolated
Rust-only copy can never satisfy that test's premise regardless of what
this crate does with its own files. That one test now skips (with an
explanation) when the canonical contracts/ tree is entirely absent,
rather than panicking and failing the whole suite over a cross-runtime
artifact the isolated copy cannot produce by construction.
Verified locally by hiding contracts/ entirely and rerunning the full
suite (217 tests across every binary): unchanged pass count, same as
with it present.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
share one alias rule
selection-vectors.fixture.json's "//aliases" comment still called the
two alias forms a cross-runtime defect ("no candidate string that
resolves through both"). That was true before the Rust alias port; it
is not true now. Both runtimes resolve a schema alias the same one way
(strip a trailing .vN, compare the remainder against the descriptor's
declared SchemaName): nameAlias resolves through both, and hashAlias -
kept in the fixture on purpose - is refused as unmatched by both, since
neither runtime's production call sites produce or accept it.
Prose-only change to this shared fixture file, as ruled: no vector
values, schema ids, or row data touched.
Also recorded for Self's map: cultnet-rs's alias port (R-E) does not
carry SchemaVersion or CompatibleSchemaIds - no Rust call site needs
either, so schema_alias::matches folds them away and agrees with the C#
rule only on the subset it actually uses (candidate/id/name comparison,
not version or compatibility-list matching).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
schema_discovery.rs embedded the 22 contract schemas via
include_str!("../../../contracts/cultnet/*.schema.json"), three levels
above src/ and outside packages/cultnet-rs entirely. There is no
workspace Cargo.toml tying this crate to its siblings, so cargo-mutants
treats this package's own directory as the whole tree it copies - every
one of those paths escaped it, and the unmutated baseline build failed
on all 22 with "No such file or directory". There was no mutation
evidence for the Rust half of this cut at all.
Fix: vendor byte-identical copies of the 22 schemas under
packages/cultnet-rs/contracts/cultnet/, git-tracked so they travel with
the package, and repoint every include_str! at the vendored copy. A
build script staging files into OUT_DIR was the other option named for
this rule, but it does not actually solve the problem: the script would
still have to read from outside the package to find its source, and
would fail the same way inside cargo-mutants' copy. Vendoring is
"another route that keeps every source path inside the package."
Vendoring duplicates files this repo treats as canonical
(contracts/cultnet/*.schema.json, shared across every runtime per
docs/cultnet-selection-cut.md section 1), so a new test,
schema_discovery_contracts.rs, compares every vendored file against its
repo-root original byte for byte and fails loudly on drift. It skips
(not fails) when the canonical contracts/ tree is not present beside
the package - the isolated-copy case R-AB exists to survive - so the
check cannot reintroduce the same escape it guards against.
cargo-mutants run on the crate, at least covering this cut's diff:
report to follow with every survivor triaged.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Cursor::compute_digest hashed the selection alone (cursor/limit cleared),
so a caller holding one valid cursor could splice its genuine digest onto
a body naming a different asOf/ordinal/schemaId/recordKey and reuse it -
the digest never bound the cursor to the position it names.
compute_digest now takes (asOf, ordinal, schemaId, recordKey, selection,
key) and folds the body into the same length-prefixed canonical string
before hashing; mint and verify_digest pass their own body through.
Tests: an in-module unit test (cursor_tests, next to Cursor, where the
private digest field and the length-prefix helpers are reachable) splices
a genuine digest onto a rewritten ordinal and a rewritten record key and
confirms verification fails on both; an integration test confirms two
cursors minted for different positions under the same selection and key
never share a digest.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
edge is anchored by its own direction
R-V (S-1): select()'s row index and the cited incoming index were both
keyed by record key alone, so two schemas sharing one key collided -
which row a `cites` edge resolved to, and whether a `cited` selection
counted a row as cited, depended on row order. Both indexes are now
keyed by (schemaId, recordKey), matching CultCache's per-schema key
uniqueness.
R-W (S-3), Rust half: a selection carrying both `cites` and `cited`
computed one `cited_direction` flag and applied it to every edge in the
batch, so the `cites` edges vanished whenever `cited` was also present.
Each `EdgeMatch` now carries its own `EdgeAnchor` (Citer/Citee) set at
the point it is produced, and the page filter/sort anchor each edge by
its own hop instead of one shared direction.
Tests: two schemas at one record key, in both row orders, agree on
which row a citation resolves to and which row counts as cited; a
selection with both hops returns edges from both directions.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Score 66% to 77.62%, every named survivor group killed, no production code
touched. The score is the least interesting part.
Two of the mutants Soul called live are equivalent: within one hop direction
the anchor and role already determine the target, so the pair cannot diverge.
They are recorded as equivalent so nobody pays for them a third time. And a
sixth mutant nobody had named survived in the same method on R-B's primary
ordering key, which means the rule R-B exists to protect had its first
component unguarded and the triage missed it.
The digest needed a golden value rather than a comparison. A same-versus-
different test cannot catch a mutant that swaps which case gets which flag,
because the two digests still differ, just mislabelled. The golden was
reconstructed from the method's documented contract rather than its own
helpers, which is what makes it evidence.
Three suspected real defects were flagged rather than quietly fixed, and get
behaviour tests first. The reachable one is an out-of-bounds read from an
attacker-supplied cursor.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The list still said Windows owed its release configuration hours after the
Windows leg landed and closed it. A map that disagrees with itself is worse
than one that is merely behind, because both halves look current.
What is actually outstanding is narrower: the burner stress, which is the
operator's fork, and whether earlier Windows numbers were measured against
the pinned MsQuic at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The first Stryker pass (72.18%, up from 66%) showed the named dozen mostly
dead but surfaced two related gaps the "differs" style of assertion cannot
reach on its own:
- ComputeDigest's field-number and cites.role presence flags each had an
equality-mutation survivor (field.Number != null / selection.Cites.Role
!= null) that a same-vs-different digest comparison cannot catch: the
mutant swaps which of two cases gets which flag value, so both cases
still digest differently, just under swapped labels. A handful of plain
AppendString/sb.Append deletions survived alongside it for the same
reason - removing content that a same-vs-different pair didn't happen
to isolate. One independently reconstructed golden HMAC value, built
from the method's own doc comments rather than its private helpers,
pins the whole wire shape at once: any single append or flag polarity
wrong breaks the exact match.
- EdgesFor's OrderBy(pageRowIndex) - the primary "page-row order" key R-B
names before "(from, role, to)" - had no test forcing page position and
from/role/to order to disagree, so it survived alongside (and separate
from) the five ThenBy mutants already named.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The bridge imports msquic.dll by name and nothing ever copied the pinned
library beside the test binary. Linux failed loudly on this. Windows did not
fail at all: DLL search fell through to PATH and loaded an unrelated msquic
from the Windows Performance Toolkit, a tenth the size of the pinned build,
and every scenario passed against it with no sign anything was wrong. The
green result was what hid the defect.
Fixed by mirroring the Linux POST_BUILD copy on the Windows side. Every result
recorded here is from a rerun afterwards, and F1 is now closed on both
platforms.
The open question is how far back this reaches. Every earlier Windows number
in this campaign was measured on this machine, where the copy has never
existed. Those figures are unconfirmed until someone establishes whether the
deleted harness placed the library itself.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Running QUIC Cut 3's Windows leg exposed the same class of defect Linux
had at 5533c34: CULTMESH_QUIC_BUILD_TESTS has never had its own packaging
step on Windows, and build-quic-native.ps1's copy of msquic.dll only
lands in the final packaged output directory, never beside the test
binary in the CMake build tree.
Unlike the Linux failure, this one did not fail loudly. gamecult_mesh_quic_native.dll
imports msquic.dll by name, and Windows DLL search falls through to PATH
when the importing binary's own directory doesn't have it. This
workstation has an unrelated msquic.dll on PATH from the Windows
Performance Toolkit (529,480 bytes, unversioned) that is not the pinned
Microsoft.Native.Quic.MsQuic.OpenSSL 2.5.9 build (4,181,856 bytes). Every
asserts-build scenario ran clean against that wrong binary before this
fix, with no error to say so.
Mirrors the Linux POST_BUILD fix: CMake already receives MSQUIC_ROOT, so
the test target now copies msquic.dll from MSQUIC_ROOT/bin/x64 beside
itself instead of leaving the loader to guess.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul's triage on 52082ec named ~12 live survivors across
CultNetSelection.cs/CultNetSelectionEvaluator.cs, each sitting on a rule
this cut exists to enforce, plus 75 uncovered mutants behind them. Each
new test drives the public selection/evaluator surface with a fixture
built to make the rule's failure observable in the answer, not in the
code's spelling - several needed new fixtures (a mixed numeric/string
alias, an index alias differing from its member name, two reference
members resolving to one target) because their absence is why the
mutant survived in the first place.
Two of the five EdgesFor order-flip mutants (To.SchemaId/To.Key) are not
addressed: within one hop direction an edge's (From, Role) already
determines its To deterministically, so no legal selection can tie
anchor+From+Role while To differs. Recorded as a likely equivalence,
left for Stryker's own verdict rather than forcing a contrived test.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
They shipped without ever being added to the public surface list.
Scalar-only is noted explicitly: every other scalar in math.cs has
float2/3/4 overloads, erf/erfinv do not, and that stays a separate
scope decision since the only consumer (Aetheria fire control) is
scalar.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The round-trip test (ErfinvInvertsErf) was the only defense of
erfinv's accuracy, and it is blind by construction: it can only prove
erf and erfinv agree with each other, not that either is correct. A
hand mutation of a single tail coefficient degraded erfinv's worst
error 300-400x (5.0e-7 to 2.0e-4) and every committed test still
passed. Added ErfinvMatchesReferenceValues, pinned against reference
values from bisecting the platform's double-precision erf (Python's
math.erf, glibc-backed) rather than this package's own erf, out to
|y| = 0.9999, at the same 2e-6 absolute bar erf's own reference test
carries. Expected values are computed for the actual float32-rounded
input, not the decimal literal, because erfinv's derivative diverges
near +-1 and that rounding alone moves the true answer by more than
the error bar if ignored.
Verified against three hand mutations, each restored byte-exactly
afterward:
- erfinv tail leading coefficient 2.83297682f -> 2.83317682f: killed
(4 assertions failed).
- erfinv tail second coefficient 1.00167406f -> 1.00187406f: killed
(4 assertions failed).
- erf's `p` coefficient 0.3275911f -> 0.3275930f: survives. Soul
measured this one 2.3x worse but still under the 2e-6 bar the spec
pins; this is not a fixture gap, it is the mutant staying inside the
accuracy envelope the tests are supposed to allow. Recorded here
rather than tightening the bar past what the spec asked for.
Added ErfinvAtDomainEdgesAndBeyond, pinning the domain and edge
behavior from the matching source comment: erfinv(1) = +infinity,
erfinv(-1) = -infinity, and NaN outside [-1, 1] including +-infinity
and NaN input.
Deleted ErfinvIsMonotonicallyIncreasing: it asserted strict
monotonicity, which erfinv does not have. Scanning every representable
float32 in [0, 0.999999] finds 9662 backward steps, first at
y = 0.00022214651, worst drop 5.960464e-8 (one ULP at that scale) --
invisible only because the old fixture stepped by 0.001. erf has the
same shape: 100 backward steps over [0, 4.5], worst drop 1.192093e-7.
Replaced it with two tests: ErfinvIsMonotonicWithinFloatingPointTolerance,
pinning monotonicity to within 1e-6 (comfortably above the measured
noise floor), and ErfCellMassesStayNonNegativeAcrossRealisticGeometries,
the consumer-level property that actually matters for fire control:
differences of adjacent erf values (cell probability mass) stay
non-negative across 16/64/256/1024/4096 cells over +-4 sigma, plus a
tight 1 sigma spread and a very tight 0.25 sigma spread.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
math.erfinv(1.0f) returned -infinity and erfinv(-1.0f) returned
+infinity: at value = +-1, w = -log((1-value)(1+value)) = -log(0) =
+infinity, the tail branch's leading coefficient is negative, and
`return p * value` delivered the inverted infinity. erf saturates to
exactly 1.0f at 5.543 sigma, so any interval edge beyond that reaches
this bug in ordinary use.
Guard both edges explicitly before the polynomial runs: erfinv(1) is
now +infinity, erfinv(-1) is -infinity, and anything strictly outside
[-1, 1] (including +-infinity) is NaN, which is the correct answer and
was already the fallthrough behavior. Documented the domain and the
edge behavior in a comment, matching the house style `sign` already
uses at math.cs:84. The approximation itself is unchanged.
Also recorded the true measured accuracy beside erf and erfinv: worst
absolute error 6.621e-7 and 5.066e-7 respectively, both inside the
2e-6 bar this package's tests pin.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CultMath.Tests is on xunit.v3. Stryker's default vstest runner cannot
capture coverage against xunit.v3's test host and hangs retrying test
sessions indefinitely (observed: 2096 mutants queued, zero completed after
~9 minutes of retries). Stryker's Microsoft.Testing.Platform runner
(-t mtp, still preview) works correctly against xunit.v3 and completed the
same 1351-mutant run in ~4.5 minutes. Pin it in config so a plain
`dotnet stryker` works without a CLI flag.
Add docs/mutation-testing.md: how to run it, the --since form for a single
cut, the triage rules, a known coverage-capture caveat in the mtp runner
(confirmed by hand-mutation against static readonly fields), and the
packages still needing their own config.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two independent breaks, not one. The loader failure is fixed in CMake rather
than in prose, using an argument the build already required. The second break
fires earlier: relative cache paths resolve against the CMakeLists.txt, not
the shell, so the build died on a missing header before the loader was ever
reached.
That example was published in the same commit that deleted the working path,
and nobody ran it. Running the documentation is the check that would have
caught it, and running it is what caught it now.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three commits on cultnet/selection-cut1 close R-V, R-W, R-X, R-Y, S-4, S-8,
S-9, S-10 and S-11. R-V is proven in both row orders, which is the point: the
old behaviour depended on which row was enumerated first.
Making the validator refuse keywords it does not implement immediately found
twelve uses across three committed schemas that nothing had ever validated.
Those three are formally unvalidatable now. Nothing broke because nothing ran
them, which is the finding rather than a reassurance.
Hands named two gaps instead of papering over them: S-11 is fixed by
inspection with no dedicated test, and the three schemas need a decision.
Caching's three failures remain unproven rather than passing, since the
stopgap's container runs as root and they expect a write to fail.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The README's Linux test build has never worked: both the bridge and the
test executable carry INSTALL_RPATH "$ORIGIN" and expect libmsquic.so.2
beside them, but only scripts/build-quic-native.sh's packaged output
directory ever got that copy. A test build produced a binary that died
at once with "cannot open shared object file". Fix it where the
knowledge already lives instead of adding a README workaround: CMake
already receives MSQUIC_LIB_DIR, so the test target now copies
libmsquic.so.2 beside itself in a POST_BUILD step.
While verifying the README's own worked example end to end, found a
second, independent break in the same docker block: MSQUIC_INCLUDE_DIR
and MSQUIC_LIB_DIR were passed as paths relative to the repository
root, but CMake resolves a relative cache-variable path against the
CMakeLists.txt that consumes it, not the invocation directory, so the
compile failed on a missing msquic.h before it ever reached the
loader. Made both paths absolute with $(pwd) and documented why.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A Hands agent working the selection cut reported that the map had no fix
batch 4 section. It did not, on the branch it could see: the live copy was on
my own working branch, main was a day behind, and the selection work branch
carried a copy from 2026-09-17. It worked from its brief instead and said so,
which is the only reason this surfaced rather than producing a cut built on
five-day-old design.
That is my failure, not the agent's. Self commits the maps and I had been
committing them wherever I happened to be standing. A map is only current if
it is current on the branch the reader has.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CultNetErrorMessage is a MessagePack-CSharp map-mode object, so every
declared key is written on the wire, nil included - the same fact R-S
already established for the raw document header. cultnet.error.schema.json
was missing routingHint entirely and typed code/details as always-present
non-null values; Soul's P7 found real bytes failing it on exactly this.
routingHint, code and details are now `required`, with nullable types
matching CultNetErrorMessage/CultNetErrorDetails. A new test decodes real
bytes for all four refusal codes (selection_invalid, cursor_stale,
cursor_invalid, reference_outside_target) against the schema.
MiniJsonSchemaValidator:
- implements minimum, maximum, maxLength and maxItems, which were declared
in committed schemas (fieldPredicate's number pattern, selection.limit,
...) but silently never checked;
- anchors `pattern` to a full match (Index==0 && Length==value.Length)
instead of Regex.IsMatch's "search anywhere", so .NET's `$` (which,
unlike ECMA-262's, matches immediately before one trailing newline even
without RegexOptions.Multiline) can no longer let e.g. "5\n" pass a
schema author wrote as "exactly one digit";
- refuses a schema node that uses a keyword this validator gives no
constraint semantics to, instead of silently ignoring it. Running every
committed contracts/cultnet/*.schema.json schema through this found 12
such uses, none reached by any existing test: `allOf`/`if`/`then`/`else`
in cultnet.database-change-raw.schema.json, and `uniqueItems` (six
places) plus one `anyOf` in the cultmesh.verse-catalog-*.schema.json
pair. contentEncoding, $schema/$id/$defs/title/description/default and
the x-cult-runtime-type vendor extension are treated as metadata with no
constraint semantics to implement, not refused.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
S-9: CurrentAsOf() returned the maximum shard-log sequence across every
shard the database holds, but that sequence is a per-shard counter -
mixing shards' counters together could report a page as exact "as of" a
sequence a different, untouched shard advanced to while the page's own
shard sat still. CultNetDatabase gains CurrentAsOf(shardId), the watermark
of one shard's own log, and CultNetDocumentRegistry.SelectPage now scopes
asOf to the single shard EnsureSingleShard already resolves the matched
rows to, falling back to the caller-supplied value only when the selection
matches no rows.
S-11: CultNetDatabaseServer.HandleSubscribeV1Async refused a hop-bearing
selection before running the door (Validate), so a selection invalid for
an unrelated reason (an empty keys list, an undeclared index, ...) got a
hop refusal that had nothing to do with why it was actually rejected. The
door now runs first, unconditionally. The hop refusal also always named
the field "cites" even when the selection carried Cited instead; it now
names whichever hop is actually set.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
All ten scenarios pass under asserts and all four required under release, on
Yggdrasil, in the committed debian:13 image. Worst overshoot anywhere was 4 ms
against a 400 ms tolerance. F1 is closed on Linux: the build that ships now
runs the timeout rule.
Running the README's own worked example is what found the second thing. It
cannot work: the test binary carries an $ORIGIN rpath and nothing puts
libmsquic beside it for a test build. That example was written in the same
commit that deleted the harness, so the working path was removed and the
broken one published together.
Two things stay open. Windows still owes its release configuration for the
same reason Linux did. And F5's stress wants sixteen CPU burners on the
platform where burners are now forbidden, which is the operator's call rather
than mine to route around.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
R-V (S-1): byKey and BuildIncomingIndex keyed the hop's row resolution by a
bare record key. Two schemas sharing one key collapsed into a single
dictionary slot (last writer wins), so which row a reference/citation
resolved to - and therefore whether a selection matched or refused
reference_outside_target - depended on row insertion order. byKey is now a
bare-key -> row-list index, and a shared TryResolveReferenceTarget searches
each edge's declared leaf set to resolve the one row that actually belongs
to it, deterministically regardless of order (Soul's P2, tested both ways).
EdgesFor's page-row anchor index is keyed the same way for the same reason.
R-W (S-2): EnsureWithinDeclaredTarget's equivalent check (now folded into
TryResolveReferenceTarget) now runs for every edge a reference carries,
before a citation's own key/schema filters that edge out - not only for the
edge the citation happens to ask about. A many-reference edge into a schema
outside its declared target refuses the selection even when the caller's
key filter would otherwise skip past it without ever inspecting it (Soul's
P3).
S-8: the dead `targetType ?? typeof(object)` fallback is deleted. D11
already refuses every reference shape whose target the cache cannot infer
before it can reach the evaluator, so the branch had no reachable fixture.
R-Y: the cursor's HMAC now covers its own body (asOf, ordinal, schemaId,
recordKey) as well as the selection. Before this, a forged cursor could
reuse a legitimately minted digest while rewriting the ordinal or record
key underneath it, since nothing about the body was ever authenticated
(Soul's P4).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The pass ran against 52082ec. Its nine hand probes, including the Rust
both-hops probe that proved S-3 on a matched pair, sit unpushed on soul-probe
in a scratch clone. Probes are throwaway by doctrine, so the pointer says both
where they are and that the findings recorded in this map are the durable
copy. Fix batch 4's tests have to reproduce the behaviour independently
anyway.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
My previous commit rewrote the batch-8 rulings as if the batch were still
owed. It was not: it landed at 5646ff7 and the harness went at 796433f, both
on hands/cultmath-erf. The map was behind the code, not the other way round.
What that changes: the release configuration and the seven mutation entries
landed inside the harness and left with it two commits later, so the release
build is owed again as a plain build. The scenario runner and every seam
survived untouched and now run as direct arguments to the binary.
Cut 3's only outstanding work is the Linux leg.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four of the six rulings for the eighth fix batch named harness entries. The
operator retired the hand-written mutation harness the same day, with no
fallback, so those clauses are dead text. What each ruling protected is not:
the release configuration, the seam wrapping the whole duration expression,
pollhammer against a runtime holding live objects, the zero and negative
timeout bounds, and closerace surviving starvation all stand as behaviour.
The bridge is C++ and no ecosystem mutation tool is wired here, so it gets no
committed mutation evidence at all. Its rules are pinned by scenarios at the
layer that decides them; a mutant is something Soul tries by hand and throws
away.
F5's 16 burners now run on Yggdrasil. Those burners are what froze the
operator's workstation.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The triage landed on the fifth attempt. Most of the 85 survivors are message
strings; about a dozen are real, and they sit on the rules the cut exists to
enforce: the alias-numeric refusal degrades to "only if every schema agrees",
a half-blank cites.target stops being refused, the index-alias role naming is
untested at all four sites, the row tiebreak and EdgesFor's ordering are
pinned by nothing, and ComputeDigest's hop branches have no coverage at all.
The equivalent mutants are named so nobody chases them twice.
R-AA's triage half is done; what it now owes is a behavioural test per live
survivor.
Soul also withdrew its own claim that C#'s cursor digest does not sort its
lists. It does, at CultNetSelectionEvaluator.cs:637.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Scope mutation testing to CultMath.Tests only, matching Aetheria's
stryker-config.json shape: Standard level, progress/html/json reporters,
break threshold 0 (score is not a gate). Mutate only CultMath's own
sources, excluding the generated Swizzles.g.cs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Pin dotnet-stryker 5.0.0 as a repo-local tool, matching Aetheria's Cut 11
adoption (F:\Projects\Aetheria dotnet-tools.json) and the current latest
release on nuget.org.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Cut 1 may not merge. Row identity becomes (schemaId, recordKey), the hop
behaves identically in both runtimes, the error message obeys its own schema,
and cultnet-rs becomes mutable by cargo-mutants.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Covers the pinned accuracy bar against tabulated erf values (0, +-0.5,
+-1, +-2, +-3, within 2e-6), odd symmetry, the +-1 limits, erfinv's
round trip through erf over y in [-0.999, 0.999], and erfinv monotonicity.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Fire control's Gaussian scatter needs erf for a cell's probability mass
and erfinv to draw the lateral offset, both called per shot. Abramowitz-
Stegun 7.1.26 for erf (stated max abs error 1.5e-7), Giles 2010 (GPU
Computing Gems) for erfinv. Neither is an HLSL intrinsic, so there is no
shader-parity obligation.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
fieldPredicate's oneOf branches use `not: {required: [...]}` to make
values/number exclusive; the validator had no "not" handling at all, so it
silently ignored the keyword and FieldPredicate_ValuesAndNumberTogether_FailsItsSchema
found nothing wrong. A schema matches "not" when validating against the
subschema produces zero errors, in which case the instance must not match it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
New contracts/cultnet/cultnet.raw-document-header.schema.json for
CultNetRawDocumentHeader - it has no payload/payloadEncoding members at all,
so headers must not $ref the record schema (which required both). The
response page's own schema now lists every property CultNetSnapshotResponseRawV1Message
declares as `required` (present on the wire, nil included) rather than only
the four that happened to be non-nullable - MessagePack-CSharp's map-mode
formatter writes every declared key regardless of null, so `headers`/`documents`
are always both present, one an array and the other an explicit null by
projection. The old `oneOf: [{required:[headers]},{required:[documents]}]`
matched both branches on every real page (both keys present) and so never
validated anything; it is now `oneOf` over each key's type (array vs null),
which is mutually exclusive. cultnet.raw-document-record.schema.json gained
schemaName/schemaVersion/schemaContentHash (present on every real record, not
previously in the schema at all - any real page would have failed
additionalProperties:false on them). fieldPredicate's two oneOf branches each
gained `not: {required: [the other]}`, so `values` and `number` cannot both
be present (previously only `required` was checked, which does not exclude
the sibling key).
Added MiniJsonSchemaValidator, a deliberately small draft-2020-12 subset
validator (type/required/additionalProperties/properties/items/const/enum/
pattern/minLength/minItems/oneOf/$ref, same-document and cross-document) -
no schema-validation dependency existed in this repo, and this only needs to
cover what contracts/cultnet/*.schema.json actually use. CultNetSchemaContractTests
decodes real wire bytes from a live CultNetDatabaseServer (header projection,
document projection, and a hop with edges) through MessagePackSerializer.ConvertToJson
and checks them against the committed schema files, plus three
fieldPredicate-shape tests pinning the values/number exclusivity.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
AppendString (CultNetSelectionEvaluator.cs) wrote value.Length, .NET's UTF-16
code-unit count, but Cursor.Parse's ReadLengthPrefixedFields slices that many
bytes out of the UTF-8-encoded body. An astral character (a surrogate pair:
2 UTF-16 units, 4 UTF-8 bytes) made the two counts disagree, so a cursor
minted for a row keyed with one (Evaluator_CursorPositionUsesCodePointOrderMatchingTheRowTiebreak's
astral key) failed to parse back - caught by that existing test on Yggdrasil,
not a new one. The prefix is now Encoding.UTF8.GetByteCount(value); the
digest's canonical string never round-trips through the byte-sliced reader
(it is hashed whole), so this only had to be right for the cursor body, but
one counting rule for both is what keeps AppendString actually unambiguous.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CultNetDatabase already had a public ResolveShard(schemaId, key) wrapping
ResolveShardInternal; the previous commit added a second one with the same
signature. CultNetDatabaseServer.CreateSelectionResponse's shardIdOf callback
uses the pre-existing one.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-O: CultNetSelectionCursorKey is a per-process (per-CultNetDatabase-instance) 32-byte
random key. CultNetSelectionCursor.ComputeDigest is now an HMAC-SHA256 under that key
instead of an unkeyed SHA-256, so a cursor cannot be forged without the key and does not
verify across a restart (a fresh CultNetDatabase mints a fresh key). The cursor body
(asOf/ordinal/schemaId/recordKey/digest) is length-prefixed (<byte length>:<value>, no
delimiter) instead of joined with a bare "\x01"-separated string, so a record key
carrying any character, U+241F included, round-trips. CultNetSelectionEvaluator.Select/
Page/ComputeDigest take an optional CultNetSelectionCursorKey, defaulting to a
process-wide key generated once so existing callers that never wire up a database still
get a keyed digest. CultNetDatabase gets a CursorKey property (random unless
CultNetDatabaseOptions.CursorKey is supplied) and passes it through
CultNetDatabaseServer.CreateSelectionResponse.
Rust already landed R-O (5aac120) under its own algorithm, deliberately not byte-matching
the reference: its own doc comment says a cursor is never handed across runtimes, so only
collision-freedom and keying are load-bearing, not the wire bytes. That is a real
divergence from this pass's instruction to match Rust's wire encoding exactly - reported
in full in the batch report.
R-Q: the two replica-apply paths (ApplyCommittedPutAsync, ApplyCommittedDelete) recorded
the mutation-log entry but never set _lastWriteSequence, so a row's order/cursor position
looked like "never written" until it was written again locally. Both now set it from the
entry's own sequence, matching AppendMutationLogEntry's local-commit path.
InitializeLogSequencesFromStore rebuilt only _nextLogSequences at startup; it now also
rebuilds _lastWriteSequence by replaying the durable log's Put/Delete entries in sequence
order. CultNetDatabase.ResolveShard is a new public wrapper over the existing shard
resolution; CultNetDocumentRegistry.SelectPage/CreateSelectionResponse take an optional
shardIdOf callback and refuse (reusing CultNetSelectionInvalidException/selection_invalid,
R-N) when a selection's matched rows resolve to more than one shard's log, because asOf is
a single watermark and is not exact for a shard-spanning page.
CultNetDatabaseServer.CreateSelectionResponse wires shardIdOf/cursorKey through from the
database.
Also fixes a defect the Rust Hands found while matching wire bytes:
CultNetErrorMessage.ForReferenceOutsideTarget double-prefixed "reference_outside_target: "
- the exception's own message already carries it. Now prefixed once, matching Rust's
ForReferenceOutsideTarget.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CultNetMessage::Error gains typed code (CultNetErrorCode) and details
(CultNetErrorDetails) fields mirroring the landed C# reference
(CultNetErrorMessage/CultNetErrorDetails, CultNetSchemaMessages.cs). The
message now encodes/decodes through the raw cultnet.schema.v0 path
(encode_raw_cultnet_schema_message/parse_raw_cultnet_schema_message)
instead of the generic serde_json::Value round trip every other
CultNetMessage variant uses, because that path's BTreeMap-backed
serde_json::Value would sort keys alphabetically and break the C#
reference's fixed schemaVersion/error/routingHint/code/details map
order - every optional key is written nil, never omitted, matching
MessagePack C#'s always-five-key formatter.
Adds the SelectionRefusal -> CultNetMessage mapping the fix-batch-3
comment in selection.rs called out as pending on the C# shape landing,
covering all four codes (selection_invalid, cursor_stale, cursor_invalid,
reference_outside_target).
cultmesh-rs's one CultNetMessage::Error match gets an added `..` so it
still compiles against the wider variant; it does not read code/details.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Four CultNetErrorMessage wire round-trip tests (decode the bytes a peer actually
receives) proving code+details for selection_invalid, cursor_stale, cursor_invalid,
reference_outside_target. A subscription-server round trip proving snapshot and
first live change are authorized by the same wire id (R-P). A subscription-server
round trip proving a removal under a field-predicate selection is delivered with
no exception (R-T(a)). An evaluator test proving an implicitly-typed many
reference still refuses an out-of-target edge (R-T(b)).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-N: cultnet.error.v0 gains additive code/details fields (contracts/cultnet).
CultNetErrorMessage grows ForSelectionInvalid/ForCursor/ForReferenceOutsideTarget
factories; both C# servers (CultNetDatabaseServer for snapshot+live,
CultNetDatabaseSubscriptionServer) use them at every existing catch site, plus
two sites that previously let the exception escape uncaught instead of reaching
the peer (CultNetDatabaseServer's v1 subscribe snapshot branch,
CultNetDatabaseSubscriptionServer's Subscribe around CreateProjectedSnapshot).
CultNetSelectionCursorException now carries AsOf/Current for cursor_stale.
R-P: CultNetDatabaseSubscriptionServer.CreateMatchedRecord (the live fast path)
authorized against descriptor.SchemaId; the snapshot path (CreateProjectedSnapshot)
already authorized against the wire id. Both now call the same
CultNetDocumentRegistry.WireSchemaId (made internal, was private) so one id
reaches the authorizer on the snapshot, live and reconcile paths alike.
R-T(a): CultNetSelectionEvaluator.MatchesSchemaKeysFields no longer reads a
Removed change's (always-null) document against Fields - it matches on
keys/schemas alone and treats a field predicate as unknown instead of throwing
TryGetIndexValue(null).
R-T(b): EnsureWithinDeclaredTarget no longer skips the out-of-target check when
TargetType is null; it checks against every registered leaf (unconstrained is a
real, checked target set, not a bypass). CultCache's ResolveReferenceTarget also
now infers the target from a many reference's CultRecordRef<T> element when no
explicit [CultReference(typeof(...))] is declared, which was the one remaining
source of a null TargetType reaching that check on a real, testable path.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Self's sequencing note (2026-09-22): the C# R-N pass (code/details on
cultnet.error.v0) is running separately in C:\ws3 and will report the
exact field names and code strings once it lands. Do not guess them.
Reverts the earlier Error{code,details} fields on CultNetMessage, the
SelectionRefusal::wire_code/wire_details/to_wire_message mapping, and
their five tests. R-O, R-T, R-U's hand-probes and R-R are unaffected -
none of them depend on the wire error shape. When the C# R-N commit
lands, pull it, read its exact field names and codes, and implement the
Rust mapping against those instead of a guess.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
docs/cultnet-selection-cut.md, Self's rulings for fix batch 3 (2026-09-22),
R-R: "the Rust v0 lowering must match the C# pre-cut v0 behaviour" -
recordKeys: [] answers empty, and the requested key order is kept. The
parallel C# pass landed the same correction on the reference server at
5e33d81/1afbaed/f8ba2d0 (pulled via rebase); this is the Rust side of it.
lower_v0_list (packages/cultnet-rs/src/snapshot_query.rs) used to collapse
a present-but-empty-after-trimming schemas/recordKeys list down to `None`
("no filter"), turning `recordKeys: []` into "every row" instead of the
pre-cut server's actual answer of nothing. It now returns `Some(vec![])`
for that case; the shared evaluator's matches_schema_keys_fields already
treats a present-but-empty list as "match nothing" (R-E), so no further
door/short-circuit logic was needed here - unlike the C# reference,
serve_read_only_raw_snapshot never calls validate, so there is no v1 door
for this shape to hit (the R-R fixup at 1afbaed doesn't apply to Rust).
serve_read_only_raw_snapshot also now restores the caller's requested
record-key order on the response (a stable sort by position in
record_keys), matching the pre-cut server's HashSet<string>
insertion-order iteration instead of leaking plain snapshot-source order.
Rewrote the stale test that pinned the old (wrong) "empty/blank list is no
filter" behaviour into
serve_read_only_raw_snapshot_omitted_v0_schema_list_is_no_filter_present_empty_or_blank_answers_empty,
and added serve_read_only_raw_snapshot_keeps_the_requested_record_key_order.
Hand-verified both by reverting each fix locally and confirming its own
test fails.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The source and test changes for the previous commit's message - split out
because the harness-deletion staged from an earlier `git rm` committed
alone before these paths were added. Same ruling, same scope: R-N's wire
code/details, R-O's keyed length-prefixed cursor, R-T's every-edge cites
check, and their behavioural tests (including the hand-verified mutants:
reverting each of the three kills its own new test).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
docs/cultnet-selection-cut.md, Self's rulings for fix batch 3 (2026-09-22),
the Rust-owned share:
- R-N: cultnet.error.v0 gains additive code/details fields. SelectionRefusal
now maps to a wire code (selection_invalid, cursor_stale, cursor_invalid,
reference_outside_target) and a flat string-map details payload via
to_wire_message(); the legacy array wire (gamecult.networking.v0) has no
slot for either and never carries them, matching v0 peers ignoring them.
- R-O: the snapshot cursor is keyed. CursorKey is a random 32-byte
per-process key; Cursor::mint/compute_digest now take it and use
HMAC-SHA256 instead of plain SHA-256, so a cursor forged without the key,
or minted under a different process's key, is refused cursor_invalid.
Cursor::parse stays public and decodes as_of without the key (Huginn's
dependency); only verify_digest (called from select()) needs it. The
cursor body itself is now length-prefixed fields, not a delimiter-joined
string, so a record key carrying any character - including the pre-fix
C# delimiter U+241F - round-trips.
- R-T: matches_citation (the `cites` path) now checks every one of a
citer's edges against its declared target before applying the
citation's own role/key filters, closing F15: a corrupt edge used to
hide behind an unrelated filter and never get validated.
- The retired Rust mutation harness entries file
(scripts/mutate-cultnet-selection-rust.entries.ps1) is deleted; tests
pin the behaviour directly.
Regenerated contracts/cultnet/interop/selection-vectors.rs-written.json
from the shared fixture; it is byte-identical to what was committed, so
R-T's fix does not change any existing fixture vector's outcome. The Rust
reader still evaluates every vector in the committed cs-written.json
(15/15) unchanged - that file predates the parallel C# Hands' R-N/R-S
work and was not touched here.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Ran CULTNET_WRITE_VECTORS=1 dotnet test tests/GameCult.Networking.Tests -c
Release --filter FullyQualifiedName~WriteSelectionVectors on Yggdrasil
against b1932d9 and committed its output verbatim. 17 vectors total (up from
15): the 4 door-refusal vectors now include
refuses_unmatched_cites_target_schema and
refuses_cites_target_by_hash_shaped_alias.
Per the brief, this is the fix-batch-3 report's parity-vector table baseline.
The C# reader (SelectionVectorsWrittenByRustDecodeAndEvaluateIdenticallyIn
TheReference) now passes all 16 of rs-written.json's vectors, up from 14/16 -
rs-written.json already carried both refusal vectors; only this file and the
door check that makes it agree were missing on the C# side. rs-written.json
itself was not touched (Rust-owned).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Item 1 of the fix-batch-3 brief: check whether CultNetSelection.Validate
refuses a cites.target.schemaId that CultNetSchemaAliasMatching cannot
resolve. It already does (CultNetSelection.cs:389-394), landed by an earlier
Hands pass on this branch (the "C# matrix Hands", after the parity table in
the spec was captured) - so no production code changed here.
What was still missing: SelectionParityVectorTests.RefusalCases() had a
comment explaining why refuses_unmatched_cites_target_schema stayed
"Rust-only", dated before that door check landed. The comment was stale and
the vector was never added. Added it, plus
refuses_cites_target_by_hash_shaped_alias (Self's ruling 2026-09-22: "the C#
reference's alias rule is the rule" - C# only resolves a name-form alias, so
a hash-shaped cites.target.schemaId is exactly the same unresolved-target
case and must refuse at the same field). RefusalCases() now takes the
fixture's schemas dictionary to reach schemas["leaf_a"].HashAlias.
Verified on Yggdrasil: dotnet test tests/GameCult.Networking.Tests -c Release
--filter FullyQualifiedName~SelectionParityVectorTests (the existing
cross-runtime reader against the committed rs-written.json still passes, so
this change does not depend on anything the Rust Hands owns). Regenerated
contracts/cultnet/interop/selection-vectors.cs-written.json with
CULTNET_WRITE_VECTORS=1: both new refusal vectors are present and both
refuse at "cites.target.schemaId" as expected before being committed as
vectors (WriteSelectionVectors's own Assert.Throws gate).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Discrepancy between spec and Body, reported per Hands' brief ("make the
smallest change that keeps the spec's intent, and report the discrepancy").
R-R says "recordKeys: [] answers empty" for v0 lowering. That holds for the
raw v0 wire protocol server (CultNetDocumentRegistry.CreateRawSnapshotResponse,
CultNetDatabase.CreateShardSnapshotResponse - fixed two commits back). It does
not hold for CultMesh's typed convenience API: R-F, landed earlier in this
cut, made CultMesh.EnsureV0Compatible run CultNetSelectionValidation.
ValidateShape client-side before a request is ever sent, specifically so that
"Keys=[]/[\"\"] is refused rather than silently lowered to 'every key'"
(CultMeshSnapshots.cs:836-841). Once R-R made ResolveDefaultSelection's
Clean(recordKeys) produce Keys=[] instead of Keys=null for an explicit empty
list, every FetchDocumentsAsync(recordKeys: []) call now trips that existing
R-F door and throws, instead of reaching the wire at all.
Kept both rulings as each one's own owner decided them: the raw wire protocol
answers empty (R-R); CultMesh's typed wrapper refuses before sending (R-F).
Rewrote FetchDocumentsAsync_WithExplicitEmptyRecordKeys_AnswersEmpty, added
two commits ago on the wrong assumption that CultMesh's typed API also
answers empty, to FetchDocumentsAsync_WithExplicitEmptyRecordKeys_
IsRefusedClientSide, asserting the actual (and, on inspection, correct) R-F
behaviour instead.
Verified on Yggdrasil: dotnet test tests/GameCult.Mesh.Tests -c Release.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The first attempt at R-R built a CultNetSelection with an empty (non-null)
Keys/Schemas array straight from the lowered v0 filter and handed it to the
shared evaluator. CultNetSelectionValidation.ValidateShape refuses exactly
that shape ("selection.keys is present and empty"), because for the v1 typed
door an empty list is a caller mistake. v0 and v1 read the same emptiness
differently on purpose (v0: answer nothing; v1: refuse), so v0 must decide
its answer before the selection ever reaches Validate.
CultNetDocumentRegistry.CreateRawSnapshotResponse and
CultNetDatabase.CreateShardSnapshotResponse now short-circuit to an empty
document set when the lowered schemas or keys list is present-and-empty,
without constructing a CultNetSelection or calling into the evaluator at all.
Caught by the test added in the previous commit
(CultNetDocumentRegistry_RawSnapshot_ExplicitEmptyRecordKeys_AnswersEmpty),
which failed with CultNetSelectionInvalidException before this fix. Verified
on Yggdrasil: dotnet test tests/GameCult.Networking.Tests -c Release, all
green (217 passed, 2 skipped by design).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CultNetV0SelectionLowering.Lower collapsed a non-null-but-empty filter list to
null ("no filter"). That corrupted two pre-cut v0 behaviours the same
function feeds:
- CultNetDatabase.CreateShardSnapshotResponse and
CultNetDocumentRegistry.CreateRawSnapshotResponse (b3d9cf7) both test
`filter?.RecordKeys != null`, not a length, so `recordKeys: []` always
answered empty. Lowering the empty list to null made it answer every row
instead. This corrects the ruling from the commit-2 fix batch that lowered
an empty v0 list to "no filter".
- CultMesh.ResolveDefaultSelection used the same collapse to decide whether an
explicit empty recordKeys should still fall back to the document type's own
schema default; under the corrected Lower it must not.
Lower(null) still means "no filter"; Lower(non-null) is always a real
(possibly empty) array. The evaluator already treats a non-null empty Keys
array as "match nothing" (CultNetSelectionEvaluator.MatchesSchemaKeysFields);
only the v0 lowering was erasing that signal before selections reached it.
Also restores the pre-cut requested-key order for
CultNetDocumentRegistry.CreateRawSnapshotResponse: pre-cut, requested record
keys were iterated in the order the caller sent them (a HashSet<string> built
from the array, which enumerates in insertion order); the new shared
evaluator's constant ordinal fell back to schema-and-key sort order instead.
ordinalOf now ranks rows by their position in the requested key list when one
was given.
Rewrote CultMeshStreamingTests.FetchDocumentsAsync_WithExplicitEmptyRecordKeys
(renamed ...AnswersEmpty) to assert the corrected behaviour instead of the
ruling it used to pin. Added
CultNetDocumentRegistry_RawSnapshot_ExplicitEmptyRecordKeys_AnswersEmpty,
CultNetDocumentRegistry_RawSnapshot_KeepsRequestedRecordKeyOrder and
CultNetV0SelectionLowering_Lower_DistinguishesNullFromExplicitEmpty to
NetworkingTests.cs.
Not done in this commit: CultNetDatabase.CreateShardSnapshotResponse (the
shard-replication snapshot path) still orders by the evaluator's default
schema-and-key tiebreak. Its own pre-cut order was raw cache iteration order,
which the new evaluator has no equivalent for and shard replication does not
appear to depend on row order; flagged as a discrepancy rather than changed
without a test that would pin the "right" order. The side-by-side b3d9cf7 vs
HEAD comparison run the spec asks for was not executed in this pass; the
pre-cut behaviour above comes from reading CultNetDatabase.cs and
CultNetDocumentRegistry.cs at b3d9cf7 directly.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Operator ruling, 2026-09-22: "Better to have nothing than a harness that
punishes refactoring." Removes scripts/mutate-dotnet.ps1 and every
scripts/*.entries.ps1 mutation-entry file for CultNet selection (networking,
mesh, caching, fix-batch-2). Also drops the two live code comments in
tests/GameCult.Mesh.Tests/CultMeshStreamingTests.cs that cited entry ids
(MESH-KEY-Loosening, MESH-DEFAULT-Loosening), rewording them to describe the
rule instead of the retired harness. Mutation testing now uses Stryker.NET on
the cut's diff, run by Soul.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Better to have nothing than a harness that punishes refactoring
(operator ruling, 2026-09-22). Remove scripts/mutate-cultmesh.mjs
(1451 lines) and its Node dependency from the Linux QUIC dev image.
The behavioural coverage stays: native/GameCult.Mesh.Quic.Native's
scenario runner and its seams (waitseam, pollhammer, holdtimeout,
zerotimeout, closerace, and the rest) are untouched. The README and
the dev Dockerfile now document building with
-DCULTMESH_QUIC_BUILD_TESTS=ON -DCULTMESH_QUIC_DEBUG_ASSERTS=ON and
running each scenario as a direct argument to the built binary,
instead of through the harness.
docs/cultnet-selection-cut.md and docs/typescript-quic-realtime-cut.md
still name the harness; they are Self's campaign history and stay as
written.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Repairs two stale mutation entries in
scripts/mutate-cultnet-selection-caching.entries.ps1
(CACHE-QJ-ExponentExpansion-Revert, CACHE-QJ-NaNInfinity-Revert): both
anchored on ExpandScientificNotation/the pre-R-D switch arm, which the
first fix-batch pass's R-D commit (7903853) deleted outright in favour of
RenderExactFloat/RenderExactDouble. Replaces the exponent-expansion entry
with CACHE-RD-ExactExpansion-Revert (reverts to .NET's shortest
round-trip ToString(), the exact defect R-D fixes) and
CACHE-RD-FractionSplit-Loosening (an off-by-one digit-split bug that
still looks "exact"), and re-anchors the NaN/Infinity entry on the
current RenderExactDouble call.
- Repairs one stale mutation entry in
scripts/mutate-cultnet-selection-mesh.entries.ps1 (MESH-FOREIGN-Revert):
anchored on the pre-R-L inline TryReadSchemaVersion call, which R-L's
commit (c01bc0a) replaced with CultNetDocumentRegistry.PayloadMatchesSchema.
Re-anchors it and adds MESH-RF-Door-Revert (R-F: EnsureV0Compatible runs
CultNetSelectionValidation.ValidateShape before its own v0 transport
refusals), with a new tests/GameCult.Mesh.Tests/CultMeshSnapshotOptionsTests.cs
proving FetchSnapshotAsync refuses Keys=[]/[""] before ever touching a
client.
- All entries verified: Caching 11/11 killed, Mesh 7/7 killed (pending final
confirmation on this run), restores byte-exact.
R-L (docs/cultnet-selection-cut.md): the Python refusal of vN>0 already has
a real regression test (packages/cultnet-py/tests/test_cultnet.py:1958-1978,
unsupported_schema_version) but no mutation entry - mutate-dotnet.ps1 is a
dotnet/NUnit harness only and this repo has no Python mutation harness. R-L's
C#-adjacent piece (TryReadSchemaVersion going internal behind
CultNetDocumentRegistry.PayloadMatchesSchema) is covered by
MESH-FOREIGN-Revert above.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Adds mutation entries for R-A, R-B, R-C, R-E, R-F, R-G, R-H, R-J
(scripts/mutate-cultnet-selection-networking.entries.ps1), each a revert
and a loosening where the rule supports one, targeting
CultNetSelectionEvaluator.cs, CultNetSelection.cs, CultNetDatabaseServer.cs
and CultNetDatabaseSubscriptionServer.cs.
- Adds S9 (docs/cultnet-selection-cut.md section 10): a new RUDP round-trip
test proving a hop-bearing v1 subscription reconciles the whole matched set
when the citer (not the page row itself) changes, plus its own mutation
entry (NET-S9-HopFastPath-Revert) reverting D6's Watch routing.
- Adds tests/GameCult.Networking.Tests/CultNetSelectionFixBatchTests.cs:
regression coverage for R-B (edge anchor/order), R-C (code-point order in
both the row tiebreak and the cursor position compare), R-E (alias
matcher; door refusal), R-F (Select validates unconditionally), R-H
(digest collision), R-J (numeric any_of canonical rendering) - none of
these rules had a test that could fail independently of the fix-batch
regression tests already in place.
- Fixes a real gap R-E named but the first fix-batch pass never wrote:
CultNetSelection.Validate now refuses a cites.target.schemaId that
matches no declared schema, at the door, instead of MatchesCitation
silently treating an unresolved target as "not found" and answering an
empty page. Resolved through CultNetSchemaAliasMatching, the reference
rule, against every declared schema - not only the selection's own
Schemas-filtered reachable set, which a second entry
(NET-RE-DoorTargetReachableOnly-Loosening) proves: a cites target is a
separate structural reference, not itself constrained by Schemas.
Coordinator-assigned item; the shared parity fixture's
refuses_unmatched_cites_target_schema vector (Rust side, another Hands)
is not yet in selection-vectors.rs-written.json, so
SelectionParityVectorTests.SelectionVectorsWrittenByRustDecodeAndEvaluateIdenticallyInTheReference
does not yet exercise it - it will once that vector lands, unmodified
by this commit.
- NetworkingTests.cs gains CultNetDatabaseServer_AnswersV1SnapshotRequestRespectingItsFields
(R-A's loosening) and DatabaseSubscriptionServer_SubscribesV1AndReconcilesAHopBearingSelectionWhenTheCiterChanges
(S9 / R-A's subscribe-listener revert).
Discrepancy: R-A's snapshot-request v1 listener registration
(CultNetDatabaseServer's _server.OnCultNet(_snapshotHandlerV1)) has no
mutation entry. CultNetDatabaseServer pairs only with Server (the
LiteNetLib/TCP-secured production construction site), and no existing test
in this suite - before or after this pass - drives Server's real
client/listener dispatch for CultNetDatabaseServer; every test calls
CreateSelectionResponse/CreateSnapshotResponse directly. That revert is
recorded as not yet reached in the entries file, not swapped for an easier
mutant.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Self's ruling of 2026-09-22 named six findings against Soul's eighth pass.
This lands the code side of all six:
- The wait-record seam (CULTMESH_QUIC_DEBUG_RECORD_WAIT) now wraps the whole
duration handed to `wait_for`, not the raw `timeout_ms` before it is
converted to milliseconds, so arithmetic applied after that conversion can
no longer live outside what the seam records. Its "unset" sentinel moves to
a separate `debug_wait_recorded` flag so a recorded -1 could never be
confused with nothing recorded.
- `pollhammer` also runs against a runtime holding a live connection to an
unroutable address (RFC 5737 TEST-NET-1), a best-effort stream on it, and a
real (never-connected) listener, closing the gap where the wait's predicate
was only ever exercised on an empty runtime.
- A new `zerotimeout` scenario pins that `timeout_ms <= 0` returns within
50 ms, probing 0, -1 and INT32_MIN.
- `closerace`'s settle no longer sleeps a guessed duration: where the
development seam exists it waits for `debug_peak_calls` to reach the poller
count (reset before any poller thread exists, so nothing can zero a count a
poller already raised) before a short fixed margin closes the last gap
between being counted and being inside the condition wait.
- `kGenerousLateToleranceMs` narrows from 800 ms to 400 ms so the wall-clock
scenarios can catch arithmetic applied outside the seam's own argument,
which `waitseam`'s equality check structurally cannot see.
scripts/mutate-cultmesh.mjs gains a plain `release` configuration on both
platforms (CULTMESH_QUIC_DEBUG_ASSERTS=OFF, CMAKE_BUILD_TYPE=Release) running
`polltimeout`, `pollbusy`, `pollhammer` and `zerotimeout`, and seven new
mutation entries: R1 (the release macro divides the duration by 10), W1-W3
(600 ms added, a 500 ms floor, and a 1000/909 scale, each applied outside the
seam's own argument), P1 (the idle predicate also ends on a live connection),
and Z1-Z2 (a 500 ms sleep and a 300 ms spin on a non-positive timeout). Two
stale comments are corrected: the scenario list above `assertScenarios` now
names `pollhammer` among the clock-sensitive entries, and a historical "1300"
probe value that no longer matches the test file is removed. The header's
section 7 intro now says the seam serves both the quiesce and the wait
recording, not quiesce alone.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The alias-matching call sites changed shape (schema_alias::matches/
matches_any now take (candidate, schema_id, schema_name)), so the entries
that edit them needed their Old/New anchors and Test names updated:
RS-E-Alias-Revert (now targets the ported name-comparison line, not the
deleted hash-stripping compare), RS-E-CitesTarget-Revert, and
RS-E-EmptySchemasNeverMatchesEverything-Revert.
New entries for this cut's rule, per spec:
- RS-E-Alias-RevertToHashStripping: the revert - go back to hash stripping
(compare schema_id against schema_id's own inferred name, the deleted
matcher's behaviour).
- RS-E-Alias-CaseInsensitive-Loosening: the loosening - compare
case-insensitively instead of Ordinal.
- RS-E-Alias-AcceptAnyVersionSuffix-Loosening: a second loosening - accept
any ".vN" suffix without comparing the name.
- RS-E-InferSchemaName-StripAnyLastDotSuffix-Loosening: a function-of-input
mutant - infer_schema_name strips any suffix after the last dot, not only
a ".v<digits>" marker.
Also fixed the invocation: File/Command values are relative to the CultLib
repo root and carry --manifest-path packages/cultnet-rs/Cargo.toml (there is
no repo-root Cargo.toml/workspace), matching -Repo C:\wal and -Target
'packages/cultnet-rs/src/selection.rs,packages/cultnet-rs/src/snapshot_query.rs'.
Verified: M0 green (4 commands, 36+81+7+1 tests), all 40 entries killed by
their own test, every restore byte-exact (git diff clean against the
committed source after the run).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Refusals become wire messages, cursors are keyed, one id reaches the
authorizer, and v0 lowering preserves v0's pre-cut answers exactly,
correcting the earlier ruling that lowered an empty list to no filter.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Self's ruling, 2026-09-22 ("the C# reference's alias rule is the rule"): an
alias is a schema *name* with a trailing .vN suffix, resolved against the
descriptor's declared name - never against a hash id. C#'s production call
sites (CultNetSelection.cs:438, CultNetSelectionEvaluator.cs:88,340) all use
the descriptor overload of CultNetSchemaAliasMatching.Matches, never the bare
id-vs-id overload.
- Row gains schema_name() (defaults to schema_id() for rows with no separate
name, e.g. raw wire records); RowSet gains schema_name(schema_id) ->
Option<String>. Both stand in for CultDocumentDescriptor.SchemaName, which
this crate has no type for (D7).
- schema_alias::matches/matches_any now take (candidate, schema_id,
schema_name) and resolve a candidate's inferred name against the declared
name, exactly as the C# descriptor overload does - not against schema_id's
own inferred name, which is what the deleted hash-stripping matcher did.
- Every call site (reachable_schemas, the cites.target door check,
matches_schema_keys_fields's D6 fast path, matches_citation) now resolves
and passes a schema's name instead of comparing two schema-id strings.
- Deleted: the hash-stripping schema_alias::matches(candidate, schema_id)
two-string overload. A hash-shaped alias ("<hash>.v1") is not a wire form
either runtime's production rule ever resolves.
Behaviour verified against a standalone run of the exact C# algorithm
(CultNetSchemaAliasMatching, CultNetDatabase.cs:47-97) covering exact-id
match, bare-name match, .vN alias resolution, MatchesAny's empty-matches-all
convention, ordinal case sensitivity, ".v" with no trailing digits, and
multiple ".v" markers (last one wins) - every case is now a Rust unit test
in schema_alias::tests.
Test fixture (tests/selection.rs) updated: the fixture's hashAlias entries
are now a shared negative check (neither runtime resolves them), and the
nameAlias entries now resolve in Rust too, fixing the confirmed cross-runtime
defect (schemas_by_name_alias). Regenerated
contracts/cultnet/interop/selection-vectors.rs-written.json (15 -> 16
vectors): added refuses_cites_target_by_hash_shaped_alias, renamed
hop_by_role_cites_hash_alias to hop_by_role_cites_name_alias.
cultnet-rs: cargo test 223/223. C# reader (rs-written.json): 14/16 - the two
failures are both the pending unmatched-cites.target door refusal
(CultNetSelection.cs:361-391 has no such check yet), not a Rust defect; the
other Hands' C# door fix covers both.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Companion to the C# side's shared-fixture commit. Replaces the
invented fake sha256-shaped LEAF_A/LEAF_B/CITER/LEAF_A_ALIAS consts
with lazily-loaded values read from
contracts/cultnet/interop/selection-vectors.fixture.json (a
std::sync::OnceLock-backed loader; FixtureRowSet's declared_indexes/
declared_roles/is_numeric/target_leaves become runtime string
comparisons instead of const match patterns). NARROW_CITER stays a
local, unshared id - it never appears in a cross-runtime vector.
base_rows() keeps S2-S23's original 7-row set (now sourced from the
fixture's first 7 rows, unchanged in value or order, so every test
that enumerates it - orders_by_ordinal_then_identity_and_reverses,
pages_exactly_once_and_refuses_a_stale_or_mismatched_cursor,
matched_is_the_total_count_not_the_page_count, ... - still holds
unchanged). A new all_fixture_rows() carries all ten rows (the
fix-batch float-tie and astral/BMP rows too) for the two parity-vector
functions only.
write_selection_vectors_for_the_reference gains float_tie_exact_ge,
astral_key_code_point_order, and the fixture's hash alias
(leaf_a_hash_alias(), a synthetic ".v1" this crate's own schema_alias
resolves) and name alias (leaf_a_name_alias(), "leaf_a.v9" - the form
only CultNetSchemaAliasMatching's descriptor overload resolves)
probes, mirroring the C# side. selection_vectors_written_by_the_
reference_decode_and_evaluate_identically_in_rust no longer panics on
the first disagreement (evaluate_vector now returns a Result instead
of .expect()'ing); it collects every vector's mismatch and fails once
with the full list, matching the C# reader's same change.
Regenerates selection-vectors.rs-written.json from the shared fixture.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Reconciles the two Hands' independently-regenerated parity vectors
(Soul's whole-cut pass, F3): the C# fixture used schema *names*
("leaf_a") as literal ids, the Rust fixture used invented fake
sha256-shaped ids - every vector carrying an id failed when read by
the other runtime.
Adds contracts/cultnet/interop/selection-vectors.fixture.json: the one
row set both writers now build from, keyed by each schema's real
SHA-256 content-hash id exactly as CultDocumentRegistry computes it
(dumped via a temporary DumpFixtureSchemaIds run). Carries the rows
either runtime needs - astral/BMP-private-use keys, a float tie,
cites/cited, a cites target reached by alias - plus two DIFFERENT
alias forms, because the two runtimes' alias matchers key off
different descriptor attributes and no single candidate resolves
through both (see the fixture's "//aliases" note and the report).
Rewrites SelectionParityVectorTests.cs to load the fixture instead of
hand-building rows (BuildFixture/BuildExtendedFixture deleted), assert
the fixture's declared schema ids still match what the registry
computes (fails loudly on drift instead of silently comparing wrong
ids), and use the real schema id (not SchemaName) as a row's
cross-runtime identity. Also fixes a real bug this file carried since
before the shared fixture: both the writer and the reader compared
`evaluation.Rows.Count` (the page size) against "matched" instead of
`evaluation.TotalMatched` (the total, per R-G) - selections with a
Limit were writing/checking the wrong number. The reader now collects
every vector's mismatch instead of stopping at the first one, so a
real parity defect is reported in full.
Regenerates selection-vectors.cs-written.json from the shared fixture.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Adds a revert (and loosening, where the rule has one) per new rule to
scripts/mutate-cultnet-selection-rust.entries.ps1, and repairs three things
the entries file's own first run against this fix batch found:
- The RS-QJ-ExponentExpansion-Revert entry targeted
expand_scientific_notation, which R-M deleted; replaced by entries
against R-D's real replacement (the mantissa/exponent bignum).
- RS-D-ExponentBranch-Revert's first draft ("New = if false") hung: it
routes every value, including f64::MAX's exponent of ~971, into the
fraction branch, whose "k = (-exponent) as usize" wraps a positive
exponent's negation into a huge usize and spins effectively forever - the
run had to kill that process by PID. "New = if true" is the safe
direction (a negative exponent's "for _ in 0..exponent" is just an empty
range, not a cast) and is pinned against a negative-exponent test
instead.
- RS-V0Lowering-Revert (pre-existing, not from this batch) unexpectedly
survived the first run: matches_schema_keys_fields's move to
schema_alias::matches_any (R-E) picked up that helper's own "empty
candidates matches everything" shortcut, masking the v0-lowering
mutation this entry exists to catch. Fixed in the prior commit
(matches_schema_keys_fields no longer uses matches_any's shortcut); a new
entry, RS-E-EmptySchemasNeverMatchesEverything-Revert, pins the fix
directly rather than only through v0's own lowering.
All commands are narrowed with "-- --skip
selection_vectors_written_by_the_reference_decode_and_evaluate_identically_in_rust":
that vector-file test is legitimately pending the parallel C# Hands's own
regenerated cs-written.json (this suite does not touch that file), the same
carve-out this file already gave the pre-existing
reactive_document_coalesces_direct_same_schema_alias_member_writes flake.
Full run against this batch's final state: 36/36 entries killed, M0 green
on both commands, no survivors.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- contracts.rs: replaces the ~335-line hand-written require_*/encode_*
codec for SelectionDocumentRecord/RawDocumentHeader/Edge/RecordRef with
two generic helpers (require_typed/encode_typed) that go through
rmp_serde::to_vec_named/from_slice and rmpv::Value's own Serialize -
these four types already derive Serialize/Deserialize with the wire's
camelCase field names. rmpv::ext::to_value/from_value was tried first and
rejected: it serializes a struct positionally (one array slot per field
in declaration order), which breaks the moment a skip_serializing_if
field is actually skipped and is not the named-map encoding this wire
always uses. Proven both ways by
tests/selection.rs's snapshot_response_raw_v1_round_trips_* tests, which
round-trip real headers, documents-with-payload, and edges with and
without a payload through the real cultnet.snapshot_response_raw.v1
message envelope.
- snapshot_query.rs: deletes reject_duplicates and its two call sites -
dead, because lower_v0_list already dedups (via its own HashSet) before
reject_duplicates ever runs, so its duplicate check can never fire.
(RawDocumentHeader::from_document, the Eq derives on the selection wire
types, and two dead branches the mutation harness's own first run proved -
exact_decimal_from_mantissa's redundant "if mantissa == 0" guard and
big_mul_small's unreachable multi-digit carry loop - are selection.rs
changes already covered by the prior commit in this batch.)
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The Rust half of the Cut 1 fix batch (docs/cultnet-selection-cut.md, "Self's
rulings for the Cut 1 fix batch, 2026-09-22"), against packages/cultnet-rs.
- R-B: hop edges now follow the hop's direction - cites keys an edge off its
From (the citer), cited off its To (the citee), each in page-row order
then (from, role, to) code-point order. Before this, both hop kinds keyed
off From, so a cited selection's edges were silently empty unless the
citer also happened to be on the page (Soul's F2). build_incoming_index
iterates the caller's own row slice, never a HashMap, for its own order.
- R-C: code-point order was already Rust's default str::cmp (UTF-8 byte
order); no change needed there.
- R-D: canonical_number::render_f32/render_f64 render the exact IEEE-754
decimal expansion from mantissa and exponent, via a small fixed-point
bignum (~60 lines), replacing the previous shortest-round-trip-form path
that rounded ties differently from .NET. expand_scientific_notation and
its test are gone (R-M); the Display probe test is kept per the map.
- R-E: schema_alias (ported from CultNetSchemaAliasMatching's string
overload) is the one schema-identity rule for `schemas` and
`cites.target.schemaId`; an unmatched cites target is refused at the
door. Found and fixed along the way: matches_schema_keys_fields must not
use schema_alias::matches_any's own "empty candidates matches everything"
shortcut (correct for the door/reachability call sites it mirrors from
C#) - v0's snapshot server calls matches() directly, without validate(),
so an empty-but-present schemas list must still match nothing there too.
- R-F: select() validates first and returns the typed refusal; matches()'s
hop guard is a real assert!, not a debug_assert! a release build would
compile out.
- R-G/P-1: Evaluation and SelectionPage both carry matched, the total match
count, not the page count.
- R-H: Cursor::compute_digest is length-prefixed for every string and list,
closing the collision Soul found (["a|b"] and ["a","b"] used to digest
the same).
- R-I: select_page projects a selection into the wire's SelectionPage -
headers or documents under Selection::projection, matched, and edges; a
header projection carries no payload, in edges too (S20).
Test fixture: real SHA-256-shaped schema ids (sha256:<hex>.v1) replace the
old fixture's schema names doubling as ids, which had silently masked the
R-E gap (vectors passed only because "leaf_a" happened to already be both
the alias and the id).
Two documented discrepancies against the map's own worked numbers, both
verified independently outside this crate (Rust bit decomposition plus a
Node/BigInt cross-check): an f32 of 3e20 is 300000006012263202816, not the
map's 300000002010536247296; 1e-7 as f64 is its own 79-digit exact
expansion, not "0.0000001" (a stale shortest-form assertion this fix batch
did not originally touch).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Regenerated contracts/cultnet/interop/selection-vectors.cs-written.json
under R-B/R-C/R-D/R-E/R-I.
SelectionParityVectorTests.cs: the six original cases' fixture
(BuildFixture) is left untouched, because SelectionVectorsWrittenByRustDecodeAndEvaluateIdenticallyInTheReference
reads the already-committed selection-vectors.rs-written.json against
it, and a new row reachable by an unfiltered or loosely-filtered case
(descending_with_limit, gt_numeric_boundary) would silently change that
case's expected page. The fix-batch rows and cases live in a new
BuildExtendedFixture/Cases set, additive over BuildFixture, used only
by WriteSelectionVectors:
- float_tie_exact_ge: a row at 2233759.25f, on the round-to-even tie
R-D's exact BigInteger rendering fixes (its canonical decimal is
"2233759.25" regardless of which shortest form a runtime's ToString
would have picked).
- cites_target_by_alias: cites.target.schemaId given as "leaf_a"
(ParityLeafA's SchemaName) against a-eq's real content-hash SchemaId,
exercising R-E's alias-matched cites (the pre-fix code required an
exact SchemaId match and this always missed).
- astral_key_code_point_order: two rows sharing one ordinal, keyed by
a BMP private-use character (U+E000) and an astral character
(U+1F602, surrogate pair 😂), pinning R-C's code-point
tiebreak order.
- header_projection/document_projection: the same hop-bearing match set
under each projection.
Known-expected failure: SelectionVectorsWrittenByRustDecodeAndEvaluateIdenticallyInTheReference
fails against the currently-committed selection-vectors.rs-written.json
(incoming_exists_true's edge count) because that file predates R-B and
was not touched here - the parallel Rust Hands owns it. Re-run once it
regenerates.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- cultnet.selection.schema.json: schemas/keys gain minItems: 1 (present
and empty is refused at the door, never "every schema/key" - omit the
field instead); limit documents that it is clamped 1..=200 by the
answering server rather than refused outside that range.
- cultnet_py/interop_peer.py: handle_server_message refused only an
exact ".v1" suffix, so a v2+ selection message fell through to a
silent empty-list answer instead of a loud refusal. is_unsupported_cultnet_version
matches any cultnet.*.vN with N > 0 instead. test_cultnet.py's R-2
test now covers v2 and v10, not only v1.
- mutate-dotnet.ps1: an entry's File (from -Entries, often written with
"/") is looked up against the $targets dictionary keyed by -Target's
own spelling; a "\\" vs "/" mismatch didn't throw at the lookup -
$targets[$file] silently returned $null - so the crash surfaced one
call later as a null-argument error inside a .NET path API instead of
as "unknown target". Both -Target and every File read from an entry
now normalise through Normalize-TargetPath before indexing $targets.
Discrepancy: the ruling names "both R-2 sites" for the Python fix; only
one site with this exact narrow-match bug exists in packages/cultnet-py
and packages/cultmesh-py (cultmesh_py/server.py's handle_message already
refuses every unrecognised schemaVersion, v1+ included, through its own
default branch). Fixed the one that exists; did not invent a second.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- R-I: new CultNetSelectionPageTests.cs kills Soul's two named S20
mutants - dropping `wantDocument &&` in ToEdge (a header-projection
edge would carry document payload bytes anyway) and always emitting
Edges instead of `selection.HasHop ? ... : null` (a hop-less
selection's page would carry an empty array instead of no Edges
member) - plus a matched-vs-page-length test for R-G. Fixing
ToEdge's mutant coverage surfaced a real bug the new tests caught:
it serialized an edge payload through CultDocumentMessagePackSerialization,
which requires the payload's type to be a registered CultDocument
and throws for a plain declared value type (float, in the fixture);
it now serializes directly through MessagePackSerializer.
- R-D: CultNetCanonicalNumberTests gains "5\n"/"5\r\n" cases - .NET's
$ matches immediately before a single trailing newline even without
RegexOptions.Multiline, so the door's pattern is re-anchored with \A/\z
and would have accepted "5\n" as canonical without this. Caching's
TryGetIndexNumberRewritesADoublesExponentNotationToPositionalDigits
is updated for 1e-7's actual exact decimal expansion (the R-D commit's
own effect): the old assertion pinned "0.0000001", the shortest
round-trip form a pre-R-D render would have produced.
- R-K: every serverThread.Join() in NetworkingTests.cs that sat outside
a try/finally now sits inside one, so a failed assertion cannot leave
the poll thread running past the point where `using` disposes the
socket it polls (an ObjectDisposedException on that raw background
Thread was a host crash, not a test failure). The bodyDemand.HasConsumers
assertions wait on the condition instead of assuming
SubscribeAsync's completion and PublishDemand happened in an order
a network round trip does not guarantee.
- Reflection fix: DatabaseSubscription_FiltersCanonicalWireChangesByRequestedSchemaAlias
built a v0-shaped CultNetDatabaseSubscribeMessage to invoke
CreateMatchedRecord by reflection; that method now takes the shared
SubscriptionRequest (R-A), so the test builds one through the same
reflection path instead.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Wires v1 into both C# servers (R-A): CultNetDatabaseServer registers
CultNetSnapshotRequestV1Message/CultNetDatabaseSubscribeV1Message
listeners and Server.cs's pre-verification allowlist gains them;
CultNetDatabaseSubscriptionServer's subscribe path is generalised to a
shared SubscriptionRequest so v0 (lowered) and v1 selections run the
same Subscribe/Watch/Reconcile core, with D6's rule wired in: a
hop-bearing selection runs the full reconcile diff on every change
instead of the single-row fast path. CultNetDatabase.CurrentAsOf gives
both servers a real snapshot watermark.
Fixes the confirmed parity and correctness defects from Soul's
whole-cut pass:
- R-B: cited/cites edges are anchored to the row the direction actually
reports against (the citer under cites, the cited row under cited),
not filtered by "From is on the page" regardless of direction; order
is page-row order then (from, role, to) in code-point order.
- R-C: a new CultNetCodePointComparer replaces every StringComparer.Ordinal/
string.CompareOrdinal in the row tiebreak, cursor position and edge
order, so an astral key sorts the same way here as it does under
Rust's UTF-8 byte comparison.
- R-D: CultCache renders a float/double's exact decimal expansion from
its IEEE-754 bits via BigInteger (RenderExactFloat/RenderExactDouble),
never .NET's shortest round-trip form, which rounded a tie the
opposite way from Rust at some values. The now-dead ExpandScientificNotation
is deleted.
- R-E: MatchesCitation compares a cites target through
CultNetSchemaAliasMatching instead of an exact SchemaId string
compare; ToEdge and ToRawRecord now emit the same wire schema id
(WireSchemaId) for the same row, where ToEdge previously used the
descriptor's own id and ToRawRecord the binding's override.
- R-F: CultNetSelectionEvaluator.Select validates first, unconditionally
- the validate:false/true flag on SelectPage/CreateRawSnapshotResponse
is gone, so no public evaluation path can skip the door. Mesh's
EnsureV0Compatible runs the new CultNetSelectionValidation.ValidateShape
(the declaration-independent half of the door) before its own v0
transport refusals, so Keys=[]/[""] is refused there too.
- R-G: Evaluation.TotalMatched is the selection's whole matching count;
SelectPage's Matched field now reports it instead of the page's row
count. CreateRawSnapshotResponse and CreateShardSnapshotResponse
(v0/shard full walks) call the new CultNetDocumentRegistry.SelectAll,
which evaluates once via CultNetSelectionEvaluator.EvaluateAll and
projects every matched row, instead of looping SelectPage per 200-row
page and re-filtering/re-sorting the whole row set each time.
- R-H: the cursor digest is length-prefixed for every string and list
(AppendString/AppendList) instead of delimiter-joined, closing the
collision ["a|b"] == ["a","b"] and ["a,b"] == ["a","b"].
- R-J: any_of on a numeric alias compares TryGetIndexNumber's canonical
rendering, not the string getter's culture-dependent ToString().
- R-M: WithBindingSchemaAlias is deleted; both servers' single-change
fast path calls CultNetDocumentRegistry.ExpandSchemaBindingAliases
instead, the one binding-alias reconciler. The duplicated v0/shard
lower-and-page loop is gone with SelectAll above.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The C# reference never serves v1, cited returns the wrong edges in both
runtimes, and the runtimes disagree on floats, schema aliases and ordering.
Floats render as their exact decimal expansion, flagged for the operator.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two things landed on this branch while commit 3 was in flight, both addressed
here rather than left for the vectors to paper over:
**Self's ruling, 2026-09-22 (commit 2 fix batch, 568e8e3/021e3f1): empty
lists, blank entries, and v0's own cleaning.** Rust already refused an empty
schemas/keys list (inherited from before the ruling); it now also matches
the reference's three further decisions exactly, not just in spirit:
- The door's refusal field is the plain list name ("schemas"/"keys"), not an
indexed "schemas[i]" - selection::validate's SelectionInvalid.field now
matches CultNetSelectionInvalidException("schemas", schema, ...) literally,
because the parity vectors compare this string across runtimes.
- snapshot_query.rs's lower_v0_list now deduplicates a mixed list the same
way CultNetV0SelectionLowering.Lower does (Distinct(StringComparer.Ordinal)
- order-preserving, first occurrence wins), not just blank-filters it.
A new test (serve_read_only_raw_snapshot_lowers_an_empty_or_blank_v0_schema_list_to_no_filter,
extended) pins the dedup case directly.
- Two mutation entries (RS-EmptySchemas-Revert, RS-EmptyKeys-Revert) added
for the door's original empty-list refusal, which had tests but no
mutation coverage before this commit.
**S12: the two-way parity vectors**, contracts/cultnet/interop/selection-vectors.{cs,rs}-written.json,
eight vectors each: the six evaluated selections from commit 3 (any_of+ge
conjunction, descending+limit, gt boundary, cited exists true/false, three
rows sharing one index value) plus two door-refusal vectors added for this
ruling - `refuses_empty_schemas_list` (schemas: []) and `refuses_blank_key_entry`
(keys: ["a-eq", " "]) - which decode their selection bytes and assert
`validate`/`Validate` refuses at exactly the named field, never evaluating.
The blank-key vector could not be generated before this commit: the door
rule it exercises did not exist in the reference until 568e8e3 landed.
`cites`-based hop vectors are deliberately absent (documented in both test
files): MatchesCitation compares Citation.Target.SchemaId against the
reference's real SHA-256 schema id exactly, with no SchemaName fallback,
and this fixture pair has no way to agree on that hash across runtimes;
`cited` (role/existence, matched by record key alone) is the hop direction
both sides can cross-check, so incoming_exists_true/incoming_negation carry
that coverage instead. Both reader tests pass:
SelectionVectorsWrittenByRustDecodeAndEvaluateIdenticallyInTheReference (C#)
and selection_vectors_written_by_the_reference_decode_and_evaluate_identically_in_rust
(Rust), each judging the other runtime's file.
**The remaining mutation matrix.** Of the coordinator's named set (S1, S2,
S6, S9-S15, S17, S20, S23): S1 was already covered (RS-EmptySchemas/Keys,
RS-BlankSchema/Key); S2 (any_of matches membership, RS-S2-Revert) and S6
(cites.role filters the hop, RS-S6-Revert) are newly covered - S6 needed a
new fixture (cites_role_excludes_a_second_reference_at_the_same_target)
because base_rows()'s citer-1 only ever carries one reference, so a mutant
dropping the role filter had nothing else to wrongly match and survived
against the original hops_one_edge_by_declared_role test; that finding is
recorded in-line as a comment, not silently swapped for an easier mutant.
S9-S15 (subscription reconcile, v0-lowering-of-subscribe) do not apply:
D7 - no Rust caller subscribes in this cut, so there is no subscription
client to mutate. S17 (inherited alias reaches every leaf) does not apply
either: Rust's toy Row/RowSet has no inheritance to begin with - both
leaves in the fixture declare "mass" independently, which is the Rust
spelling of the same guarantee, already exercised throughout the suite
without a dedicated mutant (there is no shared base-type discovery path to
revert). S20 (edges only for a hopping selection) was investigated and
found structurally vacuous in Rust as built: `edges` is only ever populated
inside the cites/cited branches, so a mutant on the has_hop() gate cannot
produce a different observable result for a non-hopping selection - no
entry added; a real gap would need the evaluator's structure to change
first. S23 (matches every row sharing an index value, never a cache's
unique-index winner) has no revertable bug surface: Rust's Row/RowSet
never had a unique-index map to begin with (D7), so this is pinned by
construction, not by a mutation-testable rule; matches_every_row_sharing_an_index_value
still exists as a positive test naming it.
**A pre-existing, unrelated flaky test, found and worked around, not
fixed.** tests/cultnet.rs's reactive_document_coalesces_direct_same_schema_alias_member_writes
fails consistently when the mutation harness runs the full tests/cultnet.rs
binary (both default-parallel and RUST_TEST_THREADS=1), but passes 5/5 in
isolation and passes in a plain `cargo test -p cultnet-rs` run - state
pollution from another test in that binary under the harness's specific
invocation path, not a race this campaign's code participates in (nothing
here touches reactive_document machinery). RS-V0Lowering-Revert's Command
is narrowed to `cargo test --test cultnet serve_read_only_raw_snapshot`
(filtering both M0 and the entry's own run to this campaign's test) so the
suite does not depend on that unrelated test's health; the flake itself is
reported here as a finding, not patched.
Tests: cultnet-rs 174 passed, 0 failed (default parallel run).
GameCult.Networking.Tests 196 passed, 1 skipped (WriteSelectionVectors,
env-gated), 0 failed.
Mutation harness: 24 entries (was 22 in commit 3; +RS-S2-Revert, +RS-S6-Revert;
RS-BlankSchema/BlankKey and RS-V0Lowering-Revert's anchors and field-name
assertions updated to match the landed fix batch's exact spelling), all
KILLED, M0 green.
Negative grep: `rg -n "untagged" packages/cultnet-rs/src packages/cultcache-rs/src` empty.
Structural delta, this commit only: 6 files changed, +108/-23 (selection.rs
+15/-9 field-name fix; snapshot_query.rs +25/-14 dedup; tests/cultnet.rs +21
dedup test; tests/selection.rs +35/-2 S6 fixture and field-name fixes;
scripts/mutate-cultnet-selection-rust.entries.ps1 +33/-4 two new entries
plus three corrected anchors/commands; SelectionParityVectorTests.cs +2/-1
field name), plus the two committed vector JSON files (~90 lines each, new).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
R-2 was owed by Cut 1 and not yet done (docs/cultnet-selection-cut.md, section
9, Self's rulings 2026-09-22): a v0-only Python peer must refuse a
cultnet.selection.v1 carrier message loudly instead of silently answering
with an empty list.
- packages/cultnet-py/src/cultnet_py/interop_peer.py:
handle_server_message's fallback now recognises any cultnet.*.v1 schema
version and replies cultnet.error.v0 { code: "unsupported_schema_version",
details: { schemaVersion } } via the new unsupported_schema_version_error
helper, matching cultmesh_py's existing error shape. One test
(test_python_interop_peer_refuses_v1_selection_messages) proves a
snapshot_request.v1 and a database_subscribe.v1 message both get the
refusal, not an empty list.
- packages/cultmesh-py/src/cultmesh_py/server.py's delegating path
(_handle_connection_message -> handle_message) already refused every
unsupported schema version generically, v1 included - no code change
needed there. test_cultmesh_local_server_returns_errors_for_bad_requests
gains an explicit assertion for a snapshot_request.v1 message through that
same path, so the coverage is named rather than only implied by the
generic case.
This is the one exception to "do not touch packages/" in this fix batch: only
these two Python packages' interop/server files and their tests.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul found commit 2 didn't pass (docs/cultnet-selection-cut.md, 2026-09-22):
empty/blank schemas-keys still reached the evaluator, the shard snapshot ran
its own loop instead of the evaluator, Mesh's v0 wire silently dropped v1-only
selection terms, the Mesh one-pass read could return the wrong schema's
record at a shared key, two decode rules for a foreign schema id disagreed,
dead identity-comparison code from the _projectRecord removal was still live,
and a disposed-socket race could abort a mutation run instead of recording
NO VERDICT. This is Self's fix batch for all of it.
- CultNetSelection.Validate now refuses a blank/whitespace entry inside
schemas or keys, not just an empty array. CultNetSelectionEvaluator no
longer treats [] as "no filter" (the door already refuses it before
evaluation). CultNetV0SelectionLowering is the one place that lowers an
empty or all-blank v0 list to null; CreateRawSnapshotResponse and
CreateShardSnapshotResponse both route through it, and CultMesh's own
Clean() now delegates to it instead of duplicating the rule.
- CreateShardSnapshotResponse no longer carries a private loop or its own
schema/key matching: it builds a CultNetSelection and answers through
CultNetDocumentRegistry.SelectPage (the same evaluator path as the
non-shard snapshot), with the shard's own membership check as a row
filter. Binding-alias expansion now runs on both paths identically.
- ResolveDefaultSelection decides "no schema filter" from the cleaned
recordKeys, not the raw array length. OverlaySelection stops deciding what
a missing override means: any explicit schemaIds/recordKeys override
replaces the default selection wholesale; no override keeps the default
untouched.
- CultMeshSnapshotRequestOptions.Selection is now checked by
EnsureV0Compatible before every send: fields, cites, cited, limit, cursor,
descending, and any non-default projection are refused typed
(CultNetSelectionInvalidException), never silently dropped from the v0
wire message. Mesh sending v1 is FU-Mesh-v1, not this cut.
- CultMeshHotBodySubscription.Selection returns a defensive copy on every
read so its arrays can't be mutated out from under the stored selection.
- ReadDocumentFromSnapshotResponse now searches candidates at one recordKey
in one order - exact schema, then alias, then a foreign schema id decoded
from the payload - instead of one OR'd condition where a case-insensitive
or structurally-lucky candidate could win first. The foreign-id decode
reads the payload's embedded schema version with the registry's shared
CultNetDocumentRegistry.TryReadSchemaVersion and alias-matches it against
the caller's own descriptor directly, rather than resolving to "whichever
registered type owns this schema string" - that resolution is ambiguous
when two document types alias the same schema id (MeshNoteDocument and
MeshNoteAliasDocument in the existing fixtures), which is what broke on
the first pass; TryResolveDescriptorByPayloadSchema stays for its existing
caller (ResolveDescriptorForRawDocument) with a note on when not to use it.
DecodeSnapshotDocuments and the Mesh copy of the payload-schema decode
(RawSnapshotPayloadMatchesSchema/TryReadSchemaVersion) are deleted in
favour of the shared one.
- CultNetDatabaseSubscriptionServer: the dead RecordKey/SchemaId identity
comparisons in Reconcile and ApplyProjectedChange, and the duplicate-key
throw in CreateProjectedSnapshot, are deleted - they can no longer fire
now that a projection's dictionary key is always the record's own key.
A new test flips authorizeRecord per key while the whole request stays
authorized (so DemandActive never toggles) and proves both halves of the
remaining diff: the removal of a now-unauthorized key and the add loop
picking up a newly-authorized one.
- Every NetworkingTests.cs test that starts an IsBackground poll thread now
joins it before its server disposes (about 20 tests; one already did).
- scripts/mutate-dotnet.ps1: a dotnet-test-host crash writing to stderr no
longer throws the whole run away under PowerShell 5.1's
$ErrorActionPreference = 'Stop' (2>&1 on a native command wraps stderr in
a terminating NativeCommandError); Run-Tests catches it and the entry loop
records NO VERDICT instead of aborting every remaining entry.
- Two probe tests: CultNetDatabase_ShardAndNonShardSnapshot_Agree{On
UnfilteredRequest,OnBindingSchemaIdOverride} prove the shard and non-shard
snapshot paths agree, including shard-membership exclusion of a document
outside the shard's key prefix and a binding schema-id override matching
on both paths.
- Mutation entries updated/added to match: the stale NET-D4-FourthMatcher
entry (its anchor was the private loop this batch deleted, not refactored)
is replaced by NET-D4-ShardMembership-Revert; the Mesh entries file's
header miscount ("two of three" loosening entries when there are only
two) is corrected and both previously-surviving loosening mutants
(MESH-DEFAULT-Loosening, MESH-KEY-Loosening) now have killing fixtures;
scripts/mutate-cultnet-selection-fixbatch2.entries.ps1 is new, covering
Reconcile's remaining two mutants.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
cultmesh_quic_native.cpp's wait_for callsite now reads
std::chrono::milliseconds(CULTMESH_QUIC_DEBUG_RECORD_WAIT(timeout_ms)) instead
of std::chrono::milliseconds(timeout_ms), so every existing entry anchored on
the old spelling is repointed at the new one, with the loosening's derivation
placed inside the macro's argument so the seam's record and the wait's own
duration stay the same evaluation.
Soul's seventh pass named eight derivations that were each the identity at
every wall-clock probe the sixth pass's tolerances could afford: a floor at 50
with no early check, a 16 ms and a 40 ms quantum, a wait that doubles past a
runtime's twelfth poll, a clamp at 7275, an offset of 25, a floor of 40, and a
6% scale. All eight are new entries here, killed by waitseam's exact mismatch
rather than by timing.
The seam itself gets a revert (it never records, so waitseam reads -1
forever) and a loosening (the record moves to an unconditional statement
ahead of the branch, reading the host's raw argument on every call instead of
only what the bridge hands its wait — indistinguishable at every value that
reaches a real wait, and caught only at timeout_ms <= 0, where the bridge's
own guard never waits and the honest seam must stay untouched).
S1's two predicate survivors get names and entries: K1 gives up after 64
wakes, K2 ends the wait on any wake once runtime->error is non-empty. Both
are new pollhammer targets; the existing "no predicate" revert is S1's
"returns on any wake" and needed no new entry, only its anchor fixed. S5 gets
a second hold-guard entry at the stated 1500 ms ceiling, alongside the
existing 1000 ms one.
assertScenarios gains waitseam and pollhammer.
Every added/changed entry was hand-verified against a manual win32-x64 build
before this commit: waitseam killed the seam's revert, its loosening, and a
representative +25 ms offset by exact mismatch; holdtimeout killed the S5
1500 ms guard by a 1.4 s overshoot; pollhammer killed K1, K2 and the
no-predicate revert within tens of milliseconds. Each mutation was applied by
hand, run, and hand-reverted with a diff against the working tree before the
next.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Self's ruling of 2026-09-22 called it after seven Soul passes: a wall-clock
probe against cultmesh_quic_next_event's timeout can never tell the host's
argument from a clamp, floor, round, scale or later-poll fault that happens to
be the identity at the probe's own value, because there is always another
function that matches. Move the observation to where the rule is decided.
Add cultmesh_quic_debug_last_wait_ms / _reset_last_wait_ms, built exactly like
the existing hold seam: gated behind CULTMESH_QUIC_DEBUG_ASSERTS, absent from a
shipped build. CULTMESH_QUIC_DEBUG_RECORD_WAIT wraps the literal timeout_ms
handed to the condition wait as a pass-through, so what is recorded and what is
actually waited on are the same evaluation and cannot diverge except by
changing what the macro wraps.
Two scenarios:
- waitseam asserts the recorded wait equals the host's own argument, exactly,
over 0, 1, 15, 16, 39, 40, 999, 1000, 1001, 7300 and INT32_MAX (the header
states no narrower maximum), and over 40 polls on one runtime past the point
every prior scenario stopped asking. timeout_ms <= 0 never reaches the wait
at all, so its probe checks the honest converse — the seam stays untouched —
which is what catches a seam that records the raw argument on every call
instead of only the value it actually hands its wait.
- pollhammer runs one thread hammering every gate-touching entry point that
cannot itself queue an event, in a yield loop, against another thread's idle
poll, once on a clean runtime and once after a refused listener_open has
left runtime->error non-empty. The idle poll must still stay for its
timeout.
polltimeout, pollbusy and holdtimeout stay as generous-margin proof that the
recorded wait is really waited, not as the precision instrument they used to
have to be: the fastest-of-ten mechanism, the 15 ms probe's tight tolerance,
and the stated-limits table they rested on are gone, because waitseam now
kills every clamp, floor, round, scale, offset and later-poll mutant by
equality regardless of timing. The Windows priority raise these three
scenarios carried is gone with them: a flat 800 ms late tolerance already
covers the worst overshoot the old table recorded (127 ms at normal priority
under load), so raising the process above the load bought nothing the new
margin does not already have. holdtimeout's long probe moves to 1600 ms, above
the stated 1500 ms ceiling on the hold guard, with the hold kept armed to
3000 ms so a guard that parks a naturally-timed-out call anyway overshoots by
well more than the margin.
Verified locally on win32-x64: waitseam and pollhammer pass against the
unmutated bridge; a seam-loosening mutant (recording the raw argument outside
the wait's own branch) fails waitseam's 0 ms probe; an inline +25 ms offset on
the wait's argument fails waitseam by exact mismatch; the S5 hold-guard mutant
(timeout_ms > 1500) fails holdtimeout by 1.4 s overshoot; K1 (++wakes > 64), K2
(!error.empty()) and a no-predicate revert all fail pollhammer's idle poll
within tens of milliseconds. All were hand-reverted and diffed clean against
the working tree before this commit.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
docs/cultnet-selection-cut.md section 6/7: packages/cultnet-rs/src/selection.rs
(new) - the Rust spelling of the vocabulary at wire parity with the corrected
C# reference (e5348d3), matching D7: no cultcache-rs knowledge, the row owner
supplies declared values and edges through two traits.
- Wire types: Selection, FieldPredicate, Citation, Incoming, RecordRef, Edge,
RawDocumentHeader, SelectionDocumentRecord (the v1 full-shape record, distinct
from v0's CultNetRawDocumentRecord, which lacks schemaName/schemaVersion/
schemaContentHash - a pre-existing v0 gap this cut does not touch).
SelectionOperator mirrors CultNetSelectionOperator for evaluator ergonomics;
FieldPredicate.op stays a raw wire string, never a serde-tagged union.
- canonical_number module (Q-J, section 2 "Numbers"): is_canonical (a
hand-rolled digit-grammar parser, no regex dependency added),
canonicalize_decimal_digits, expand_scientific_notation, compare (sign, then
integer length, then digits, then padded fraction - no f64 anywhere), and
render_i64/render_u64/render_f64/render_f32 for a Row implementer's own
rendering. probe_f64_display_never_uses_exponent_notation empirically
verifies the map's claim about Rust's f64 Display (confirmed true) rather
than trusting it.
- Row/RowSet traits (D7): schema_id, record_key, ordinal, values, number
(canonical string, never f64), references for Row; all_schema_ids,
declared_indexes, declared_roles, is_numeric, target_leaves for RowSet.
A blanket `impl<T: Row> Row for &T` lets sorted/filtered slices of
references satisfy generic bounds without an extra copy.
- validate (S1 door refusals), matches (D6 single-row fast path, hop-bearing
selections assert via debug_assert since Rust has no exception type to
mirror C#'s InvalidOperationException), select (order/hop/cursor/paging,
D9's reference_outside_target refusal, S23's per-row match - never through
a unique-index map, because there is none here to begin with).
- Cursor: mint/parse/digest is this runtime's own internal algorithm (a
cursor is opaque and never crosses runtimes - each server answers its own
pages), not a byte-for-byte port of the reference's; what must match is the
*rule* (asOf + last position + a digest bound to the selection), which it
does, pinned by RS-CursorStale-Revert/Loosening and RS-CursorDigest-Revert.
- contracts.rs: three v1 CultNetMessage variants (SnapshotRequestV1,
DatabaseSubscribeV1, SnapshotResponseRawV1) matching the C# messages'
field-for-field shape. SnapshotResponseRawV1 carries binary payloads
(documents[].payload, edges[].payload), so it is routed through the
existing raw JSON-bypass path (parse_raw_cultnet_schema_message/
encode_raw_cultnet_schema_message) alongside v0's DocumentPutRaw/
SnapshotResponseRaw, with its own require_*/encode_* helpers for
SelectionDocumentRecord, RawDocumentHeader, Edge and RecordRef.
SnapshotRequestV1/DatabaseSubscribeV1 carry no binary and use the
ordinary serde_json-intermediate path. No message-layer selection
semantics validation here, matching the reference
(CultNetSchemaMessageSerialization.cs only deserializes; a server calls
selection::validate against its own RowSet before evaluating).
- snapshot_query.rs (D3/D4): serve_read_only_raw_snapshot's two hand-rolled
allowlist filters collapse into one call to selection::matches over a
lowered v0 Selection, through a minimal RawSnapshotRow wrapper whose
ordinal/values/number/references are unreachable (the lowered selection
never sets fields/cites/cited, so matches_schema_keys_fields never calls
them) - documented, not hidden.
Ruling (Self, 2026-09-22, docs/cultnet-selection-cut.md commit 2 fix batch,
mirrored here per the coordinator's mid-task instruction rather than C#'s
pre-fix spelling as it stood at e5348d3):
- validate refuses an empty-but-present schemas/keys list (already had this)
and now also refuses any individual entry that is empty or whitespace
(fields[i]-style names: schemas[i]/keys[i]).
- The evaluator never sees []; matches_schema_keys_fields does not special-case
an empty list as "no filter" (it never did here - that C# fallback,
CultNetSelectionEvaluator.cs:84 pre-fix, was never mirrored into Rust).
- snapshot_query.rs's v0 lowering gets its own cleaning, owned by the
lowering and not the evaluator: lower_v0_list turns an empty or
all-blank v0 schemaIds/recordKeys list into None before it ever reaches
validate or select.
Tests: cultnet-rs 173 passed, 0 failed (was 111 baseline #[test]s pre-cut;
+62 across lib unit tests, tests/selection.rs and tests/cultnet.rs).
cultcache-rs unmodified, 64 passed, 0 failed (untouched by this commit -
confirmed unaffected). GameCult.Networking.Tests (the new
SelectionParityVectorTests.cs file, compiled but not yet fully exercised -
see below) builds clean.
Mutation harness (tools/eureka-mutations.ps1, scripts/mutate-cultnet-
selection-rust.entries.ps1): 22 entries against src/selection.rs and
src/snapshot_query.rs, all KILLED, M0 green. Covers: S1's empty-list and
new blank-entry refusals (schemas and keys, revert x4), order/descending
(revert), cursor staleness (revert + loosening) and digest (revert), S7's
incoming negation (revert + loosening), S16's le/ge boundaries (revert x2),
S18's reference_outside_target (revert), Q-J's door regex and -0 exclusion
(revert x2), Q-J's comparator integer-length and sign (revert x2), Q-J's
integer-through-float64 renderer, decimal trailing-zero stripping and
exponent expansion (revert x3, the last two pinned via direct
canonical_number unit tests since render_f64 itself cannot reach exponent
notation given Rust's Display contract - verified, not assumed), and the
new v0-lowering rule (revert).
Negative grep: `rg -n "untagged" packages/cultnet-rs/src packages/cultcache-rs/src`
empty.
Not included here, landing as commit 4 once ready: the committed parity
vector files (contracts/cultnet/interop/selection-vectors.{cs,rs}-written.json).
Both directions already work for six evaluated selections and the []
door-refusal case; the blank-key-entry refusal vector cannot be generated
from the C# side yet because the parallel C# fix batch implementing that
door check (docs/cultnet-selection-cut.md, "Self's rulings...for the commit
2 fix batch") has not landed on this branch as of this commit. Rust's own
tests (validation_refuses_a_blank_entry_in_schemas_or_keys,
serve_read_only_raw_snapshot_lowers_an_empty_or_blank_v0_schema_list_to_no_filter)
already pin both new rules independently of C#.
Structural delta: packages/cultnet-rs/src/selection.rs +1407 (new),
contracts.rs +616/-21, snapshot_query.rs +76/-21 (net +55, collapsing the
fifth selector engine into one call), lib.rs +2, tests/cultnet.rs +67 (new
v0-lowering test), tests/selection.rs +900 (new, evaluator tests + vector
machinery), scripts/mutate-cultnet-selection-rust.entries.ps1 +233 (new),
tests/GameCult.Networking.Tests/SelectionParityVectorTests.cs +296 (new).
Against section 14's ~430-line estimate for selection.rs+contracts.rs: the
two together land at ~2000 lines added, well over estimate - the map's own
§14 correction already flagged commits 0/1 running roughly 2x estimate for
the same reason (the hop's incoming index, the cursor, and per-field door
messages cost more than the prose assumed), and the v1 raw-message handling
(SelectionDocumentRecord/RawDocumentHeader/Edge require_*/encode_* helpers,
~400 lines) was not separately budgeted at all.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Seven passes on one rule, each killing the last Soul's functions and each
followed by another that is identity at every probe. Wall-clock observation
from outside cannot converge on an arithmetic rule. A dev-only seam records
the computed wait, and equality kills the whole family.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The reconcile tests lost what they proved, the shard snapshot still loops on
its own, and the public options type silently drops every term v0 cannot
carry. Empty lists are refused at the one door, Mesh refuses what it cannot
send, and the dead identity code is deleted.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Corrects commit 1's Number field: docs/cultnet-selection-cut.md section 2
"Numbers" (added 2026-09-22, superseding the map's own stale float64 text)
specifies canonical decimal strings per the operator's Q-J ruling, and
commits 0/1 shipped double instead - the Rust Hands caught the conflict
before writing a line and stopped; this is the correction the map's
Progress block calls for before Rust starts.
- CultNetFieldPredicate.Number: double? -> string?. New CultNetCanonicalNumber
(CultNetSelection.cs) is the wire's canonical grammar and comparator:
IsCanonical validates ^-?(0|[1-9][0-9]*)(\.[0-9]*[1-9])?$ and excludes "-0"
(the regex alone matches it); Compare orders by sign, then the integer
part's length, then its digits, then the fraction padded on the right to
equal length - a pure string function, no double anywhere on the path.
ValidateField refuses a non-canonical number at the door
(selection_invalid) rather than normalising it.
- CultNetSelectionEvaluator.CompareNumber now takes two canonical strings and
switches on CultNetCanonicalNumber.Compare's sign instead of comparing
doubles directly.
- CultCache.cs: CultDocumentDescriptor.TryGetIndexNumber and
PersistedMember.NumberOf render a member's exact canonical decimal instead
of a double. An integer renders exact (no 2^53 limit); decimal strips
trailing fractional zeros; float/double use .NET's shortest round-trip
ToString(), rewritten out of exponent notation to positional digits
(ExpandScientificNotation) before the shared CanonicalizeDecimalDigits
strips zeros and collapses a negative-zero result to "0"; NaN and infinity
render null (TryGetIndexNumber returns false, matching "matches no
comparison").
- contracts/cultnet/cultnet.selection.schema.json: "number" is now a string
with the canonical pattern (a defence-in-depth (?!-0$) lookahead beyond
what the door enforces in code), not a JSON number.
- Tests: CultNetCanonicalNumberTests (new) pins IsCanonical/Compare directly,
including the door's five named non-canonical spellings, magnitude vs.
lexicographic ordering, sign, integer-part length, and a long past 2^53
against its float64 rounding. CultNetSelectionEvaluatorTests gains
ValidationRefusesNonCanonicalNumberSpellings (+1, 1.0, 01, 1e3, -0) and its
existing Number literals become canonical strings.
CultDocumentSelectionSurfaceTests gains a dedicated FixtureLeafNumbers
fixture and four tests: the 2^53+1 long rendering exactly, a decimal's
trailing zeros stripped, a double's exponent notation rewritten to
positional digits (1e21, 1e-7), and NaN/Infinity answering false.
Mutation entries (scripts/mutate-dotnet.ps1): four new Caching entries
(integer-through-float64, decimal trailing zeros, exponent expansion,
NaN/Infinity) and five new Networking entries (the door's regex, the door's
explicit "-0" exclusion, the comparator's integer-length check, a
lexicographic-whole-string mutant, and the comparator's sign check) - all
KILLED. The two pre-existing S16 (le/ge boundary) entries had to be
re-anchored: their Old text quoted CompareNumber's previous double-based
`left <= right`/`left >= right` spelling, which no longer exists now that
CompareNumber switches on Compare's int result: `cmp <= 0`/`cmp >= 0`.
Verification: dotnet test - GameCult.Networking.Tests 192/192 (was 183 before
these additions), GameCult.Caching.Tests 192/192 (was 188), GameCult.Mesh.Tests
252/253 (one pre-existing unrelated skip, unaffected by this commit - no Mesh
source touched). Caching mutation run: 10/10 KILLED (6 pre-existing + 4 new),
M0 and final both green. Networking mutation run: 15/15 KILLED (10
pre-existing + 5 new, across CultNetSelectionEvaluator.cs, CultNetDatabase.cs
and CultNetSelection.cs), M0 and final both green.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
scripts/mutate-cultnet-selection-mesh.entries.ps1 covers the two rules moved in
commit 2d against tests/GameCult.Mesh.Tests: ResolveDefaultSelection's "no schema
filter when keys are given" default, and ReadDocumentFromSnapshotResponse's
recordKey-anchored exact read (including the foreign-schema-id decode fallback it
still honors).
M0: green, 87 passed (CultMeshStreamingTests). Verdicts:
- MESH-DEFAULT-Revert: KILLED
- MESH-DEFAULT-Loosening: SURVIVED
- MESH-KEY-Revert: KILLED
- MESH-KEY-Loosening: SURVIVED
- MESH-FOREIGN-Revert: KILLED
- Final: green, 87 passed. All restores byte-exact (working tree clean after the run).
Both SURVIVED entries are recorded as fixture gaps, not swapped for easier
mutants: no test in CultMeshStreamingTests.cs passes an explicit empty (non-null)
recordKeys array, and none varies a record key's case, so nothing distinguishes
MESH-DEFAULT-Loosening's `recordKeys != null` from the correct `recordKeys is
{ Count: > 0 }`, or MESH-KEY-Loosening's OrdinalIgnoreCase from the correct
Ordinal comparison. MESH-FOREIGN has a revert entry only; no distinct, still-wrong
"loosening" of "decode successfully as TDocument" was found that isn't already
one of the other two entries.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Collapses the rest of the Mesh selector surface named in docs/cultnet-selection-cut.md
section 4/11 onto CultNetSelection:
- CultMeshSnapshotRequestOptions becomes a record carrying Selection in place of
SchemaIds/RecordKeys; its three hand-written clone functions
(CloneSnapshotRequestOptions x2, CloneSnapshotFacadeRequestOptions) are gone in
favor of `with`.
- CultMeshPeerSnapshotDocumentOptions becomes a record for the same reason;
WithPublicationBindingOptions and ToSnapshotRequestOptions collapse to `with` and
one inlined Selection literal at their single call site.
- CleanSnapshotFilter and the three-times "no schema filter when keys are given"
rule collapse into two owners: CultMesh.ResolveDefaultSelection (the typed-document
default) and CultMesh.OverlaySelection (explicit override over defaults).
- CultMeshHotBodySubscription carries one Selection instead of parallel
RecordKeys/SchemaIds allowlists.
- ReadDocumentFromSnapshotResponse's four-tier fallback collapses to one
recordKey-anchored read: exact schema/alias match, or (foreign/runtime-generated
schema id) a payload that decodes as TDocument. The two tiers that matched by
schema alone or recordKey alone - the ones that could return a record under the
wrong key - are gone; DocumentHandle_ReadsRemotePeerSnapshotsAsTypedDocuments
pins the surviving recordKey-anchored decode fallback.
Two items in section 11's table are not cut here, by finding rather than oversight:
- IsSameCultDocumentSchema (CultMesh.cs) already delegates to the one
CultNetSchemaAliasMatching owner (no second engine); deleting the wrapper would
duplicate its type-to-descriptor resolution at four call sites instead of
collapsing it. Left in place.
- The selector literals in CultMeshClient.cs (574-604, 830-860, 932-962) call
CultNetDatabaseSubscriptionClient.SubscribeAsync, which stays a v0
(schemaIds/recordKeys) API in this cut - that client is not named in section 7's
per-file plan. There is no Selection to move these onto without adding a v1
subscribe path to Networking, which is out of this commit's scope. Left in place.
Tests: GameCult.Mesh.Tests 252/253 (1 pre-existing unrelated skip), 0 failed.
GameCult.Networking.Tests 161/161 excluding one pre-existing flaky test
(DatabaseSubscriptionServer_ReconcilesDeliveredProjectionAndBodyDemandWhenAuthorityChanges,
a background RUDP poll thread racing a disposed socket in code this commit does not
touch; it crashes the test host the same way on this branch before this commit too).
GameCult.Caching.Tests 188/188.
Net line count, src/GameCult.Mesh only: +143/-226, net -83. Section 11 estimated
about -326 for this cut's slice. The gap is mostly the two items above (about -49
of the estimate) plus one already-ruled keep this map's Progress block already
named (the foreign-schema-id decode fallback, -41 in the table, kept per the prior
Hands' ruling and untouched here) and the four-tier fallback surviving as a
two-tier one rather than a zero-fallback exact read (an honest test-enforced
requirement, not a padding choice).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The operator ruled decimal strings on 2026-09-17 and the map recorded it in
section 16, but section 2 and D8 still said float64 and nothing specified the
canonical form or the comparison. Commit 1 built what section 2 said. The Rust
Hands stopped on the conflict before writing a line.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Self's ruling (docs/cultnet-selection-cut.md, 2026-09-22) required a mutation
entry for the fourth schema-matching copy's collapse into
CultNetSchemaAliasMatching, landed in commit 2a. Added
NET-D4-FourthMatcher-Revert, pinned by
CultNetDatabase_Creates_ShardSnapshotResponse_ForCompatibleSchemaAlias: killed.
Every pre-existing entry in this file now carries its own File, so the whole
file runs as one invocation against both CultNetSelectionEvaluator.cs and
CultNetDatabase.cs (it previously assumed a single implicit target and would
throw given two). Ran the combined file: 10/10 entries killed, M0 green, both
targets restored byte-exact (verified by git status showing no diff on
either .cs file after the run).
A loosening mutant for the same site
('(requestedSchemaIds != null && requestedSchemaIds.Length == 0) ||', i.e.
reject only an empty-but-non-null requested-schema list) was run by hand and
SURVIVED: CultNetDatabase_Creates_ShardSnapshotResponse_ForCompatibleSchemaAlias
puts exactly one document in the whole cache, so an over-permissive filter
still returns the same single row. Recorded as a comment in the entries file
per the brief ("report it, do not replace it with an easier mutant") rather
than committed as a fake kill or swapped for an easier mutant. Fixing it needs
a second document of a different schema in the shard's fixture; not added
here.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CultMesh.IsSameCultDocumentSchema (CultMesh.cs) and the decode-time alias check
in CultMeshSnapshots.DecodeSnapshotDocuments both re-implemented schema-identity
comparison independently of CultNetSchemaAliasMatching, the one matcher landed
in commit 1. Both now call CultNetSchemaAliasMatching.Matches; the private
InferSchemaName duplicated at CultMeshSnapshots.cs is deleted (Matches already
applies the same .vN inference internally).
Spec discrepancy, reported per the Hands brief (map contradicted by the Body):
docs/cultnet-selection-cut.md section 4 lists CultMeshSnapshots.cs:1033-1047
(RawSnapshotPayloadMatchesSchema) and :1078-1088 (InferSchemaName) as deletions,
on the stated reasoning that decode no longer needs a client-side re-filter
because the server's v1 answer is exact. Deleting RawSnapshotPayloadMatchesSchema
outright breaks two existing, still-relevant Mesh tests:
SnapshotHelpers_DecodeSameSchemaAliasFromForeignSchemaId and
SnapshotEndpoint_SyncsCompatibleForeignSchemaIdIntoLocalCache
(CultMeshStreamingTests.cs). Both cover a wire schemaId this runtime cannot
resolve by any alias at all (a foreign/runtime-generated id, e.g.
"runtime.generated.mesh-note.ui.42") where the only way to decode is to trust
the payload's own embedded schemaVersion. That is a decode capability, not a
re-filter of an already-exact server answer, and the map's stated reason for
deleting it does not apply to it. Smallest change keeping the map's intent
true: kept RawSnapshotPayloadMatchesSchema and TryReadSchemaVersion, deleted
only the duplicate InferSchemaName inside it and routed the comparison through
CultNetSchemaAliasMatching.Matches.
Test counts, no regressions: Networking 162/162, Mesh 252/253 (the one
pre-existing skip), Caching 188/188.
Negative grep: MatchesRequestedSchema empty. The one remaining
"MatchesSchema(" grep hit is RawSnapshotPayloadMatchesSchema's own name and
declaration, kept for the reason above.
Still outstanding from the map's Mesh collapse (§4/§11, ~365 gross lines):
the five cloned CultMeshSnapshotRequestOptions-carrying option records, the
"no schema filter when keys are given" rule (x3), CleanSnapshotFilter, the
four-tier ReadDocumentFromSnapshotResponse fallback, and every selector
literal in CultMeshClient.cs/CultMeshBodyDemand.cs/CultMesh.cs peer-snapshot
and body-subscribe construction. Those require CultMeshSnapshotRequestOptions
(and the v0 wire builders that read it) to carry a Selection instead of
separate SchemaIds/RecordKeys — a wider, structural change across Mesh's
public surface that is not done in this commit. Recorded as the next slice.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
_projectRecord (CultNetDatabaseSubscriptionServer.cs) reached no production
caller and ran after authorization with no wire consumer. Removed the field,
constructor parameter, and both call sites in Watch() and
CreateProjectedSnapshot(); projection is a value on a v1 selection now, not a
delegate. The two tests that used projectRecord to simulate a record's
identity changing for the reconcile diff
(DatabaseSubscriptionServer_AuthorizesRequestAndFiltersSnapshotAndLiveRecords,
…ReconcilesDeliveredProjectionAndBodyDemandWhenAuthorityChanges) are rewritten
to drive a real rename through the database (delete the old key, put the new
one) instead, so they still prove authorization filtering and the Reconcile()
diff/body-demand behavior.
Self's ruling (docs/cultnet-selection-cut.md, 2026-09-22): the fourth
schema-matching copy at CultNetDatabase.cs, serving CreateShardSnapshotResponse
and SchemaMatchesCatalogFilter (MatchesRequestedSchema + a duplicate
InferSchemaName), collapses into CultNetSchemaAliasMatching.MatchesAny, the one
matcher landed in commit 1.
Networking tests: 162/162 (was 162/162 after commit 1; no count change, two
tests rewritten not added). Negative grep: _projectRecord|projectRecord empty;
MatchesRequestedSchema gone (one remaining substring hit is
RawSnapshotPayloadMatchesSchema in CultMeshSnapshots.cs, the Mesh copy this
commit does not touch — collapses in the next commit).
Mesh collapse (§4/§11 of the map, ~365 gross lines across CultMesh.cs,
CultMeshSnapshots.cs, CultMeshClient.cs, CultMeshBodyDemand.cs) is not in this
commit; it is large enough to build on its own and follows as commit 2b.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
docs/cultnet-selection-cut.md section 2/4/6/7/8: CultNetSelection (schemas,
keys, fields, cites, cited, projection, descending, limit, cursor) with its
door validation (S1), and CultNetSelectionEvaluator: the single-row Matches
fast path (D6) and the full Select (order, hop, cursor, paging). Neither
reflects over a document; both read only through the cache's public surface
(commit 0).
Three v1 messages beside v0 (cultnet.snapshot_request.v1,
cultnet.database_subscribe.v1, cultnet.snapshot_response_raw.v1), their
schema files under contracts/cultnet, registry entries and dispatch arms.
CultNetSchemaAliasMatching goes internal -> public (D4), the one alias
matcher. CultNetDatabase gains LastWriteSequence(schemaId, key), confirmed
at the one append site (AppendMutationLogEntry) that every committed write
takes a shard-log sequence, including a local PutAsync outside any shard.
Three engines deleted, one evaluator wired in their place:
- CultNetDocumentRegistry: CreateRawSnapshotResponse's reflection loop over
cache.AllEntries and MatchesRequestedSchema gone; cache.AllStoredDocuments
replaces the reflection entirely. v0 lowers into a Selection and walks the
evaluator's own pages so a v0 caller still sees its whole matching set
(D3: v0 has no engine of its own). CreateSelectionResponse/SelectPage/
ToRawRecord/CreateSnapshotRequestV1 added.
- CultNetDatabaseServer: CreateChangeMessage/CreateRawDocumentRecord/Matches/
MatchesSchema/InferSchemaName (122 lines) deleted; CreateSubscription lowers
v0 into a Selection and matches through the evaluator.
- CultNetDatabaseSubscriptionServer: CreateAuthorizedChange/CreateChange/
CreateChangeCore/ResolveWireSchemaId/MatchesRequestedSchema/CreateRawRecord
(92 lines) deleted, replaced by CreateMatchedRecord. _projectRecord and
CreateProjectedSnapshot's collapse are commit 2's, per the map's own
deletion line, and are untouched here.
Discrepancy found and fixed (not in the map): CultNetDocumentBinding can
carry a schema id different from its descriptor's own
(CultNetDocumentBinding.ForDocument's schemaId override), which the old
per-engine matchers checked and a descriptor-only evaluator cannot see by
design (it stays decoupled from CultNetDocumentRegistry). Fixed at the two
call sites and in CultNetDocumentRegistry.SelectPage by resolving the
binding's schema id into the effective selection before evaluation
(CultNetSchemaAliasMatching.WithBindingSchemaAlias,
CultNetDocumentRegistry.ExpandSchemaBindingAliases), not in the evaluator.
Found by DatabaseSubscription_FiltersLiveChangesByWireSchemaBinding timing
out; both call sites now pass.
Second discrepancy, reported not fixed: CultNetDatabase.cs:1707 has a fourth,
undocumented copy of schema-matching logic (MatchesRequestedSchema, private,
serving CreateShardSnapshotResponse), which the map's section 1 audit did not
name among "three separate selector engines". Left untouched this commit:
it is out of the map's explicit deletes list and touching it risks shard
snapshot behavior with no budgeted verification.
Verification: dotnet test tests/GameCult.Networking.Tests - 162 passed (150
existing + 12 new), 0 failed; the seven map-named tests all still pass, two
rewritten for the new signatures (CreateChangeMessage now takes
CultNetSelection; CreateChange/CreateAuthorizedChange collapsed into
CreateMatchedRecord, returning CultNetRawDocumentRecord? directly).
tests/GameCult.Mesh.Tests - 252 passed, 1 pre-existing skip, 0 failed (no
Mesh source touched this commit). tests/GameCult.Caching.Tests - unaffected
(188, unchanged from commit 0).
Negative greps: GetProperty|GetField|GetCustomAttribute in
GameCult.Networking - only against the generic `object change` dispatch
(CultNetDatabaseChange<T>) and CultRecordHandle<T> construction, pre-existing
and out of this cut's scope, never against a document's declared values.
_projectRecord survives in CultNetDatabaseSubscriptionServer.cs (expected,
commit 2). CultNetSchemaAliasMatching is the only schema matcher outside the
one reported discrepancy.
scripts/mutate-dotnet.ps1: the caching mutation runner generalized into a
shared dotnet/NUnit harness (Target/TestProject/Entries/ControlFilter),
replacing scripts/mutate-cultnet-selection-caching.ps1 with an entries file
of the same six mutations (re-verified: all six still KILL). It is now
generic to any NUnit project in this repo, not only this campaign's two
targets - a candidate to sit beside tools/eureka-mutations.ps1 as the
dotnet-side counterpart. Nine new networking entries (S3, S5 x2, S7 x2,
S16 x2, S18) all KILLED. Not yet mutation-run: S1, S2, S6, S23 (pinned by
tests, not by a committed mutant); S9-S15 (subscription reconcile lowering,
v0 no-engine-of-its-own); S17, S20 (pinned by tests only). S19/S21/S22 are
cache-layer, covered in commit 0.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Soul's sixth pass (N4): `(woken || timeout_ms > 1000) && HELD` survived both
targets. The two existing hold entries only rearrange the guard's syntax,
and `holdtimeout`'s single 150 ms poll sits below any threshold such a guard
would carry.
`holdtimeout` now runs two idle polls under the armed hold, 150 ms and
1300 ms, and holds for 2200 ms. The new entry is that guard. Compiled in on
both targets, the 1300 ms poll is parked until the release and returns at
about 2200 ms, and every other scenario stays green. The harness kills it
on both targets.
With the two polls running together, the first poll's exit also wakes the
second. That gives the asserts build a second multi-threaded witness for the
predicate: the N1 revert and read-what-woke mutants die here as well as on
`pollbusy`.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Soul's sixth pass (N1): every CallScope destructor wakes every waiter, so
without the predicate a poll leaves whenever another host thread makes any
call. Every scenario had a single-threaded host, and dropping the predicate
survived the whole matrix on both targets.
`pollbusy` is a public-API scenario and needs no development seam. A second
host thread calls `cultmesh_quic_last_error` every 20 ms for twice the
poll's timeout while an idle `poll(1000)` waits. It checks the timing first.
It then checks the other thread's call count, so a poll that passed without
ever being woken reads as the fixture failing.
Three harness entries:
- revert: no predicate, `woken` taken from `cv_status`;
- loosening: the wait reads what woke it and does not wait again;
- loosening: a hand-written loop that restarts the whole timeout on every
wake. Its lateness is a function of the input.
Each was compiled in and run through every scenario on both targets:
- revert and read-what-woke: `pollbusy` returns at 0 ms;
- re-arm: `pollbusy` returns at about 3000 ms.
The earlier scenarios all stay green under all three. The harness kills all
three on both targets.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Soul's sixth pass (N2): with probes at 150, 200 and 5000 ms, `min(t,5000)`,
`min(t,4990)`, `t*11/10` and `t+70` survived both targets, `max(t,220)`
survived linux-x64, and `max(t,250)` and `t+80` died by 22 and 2 ms. "5000 is
above any plausible ceiling" was wrong: 5000 is a round ceiling.
- A 15 ms probe, asked ten times with the fastest checked against 30 ms. It
is the only place a floor or an added constant shows. Scheduler delay only
adds, so noise has to delay all ten attempts to fail the bridge.
- The long probe is 7300 ms, which is not a round number.
- A single poll gets 60 ms plus a sixteenth of the timeout.
- Each timed scenario prints its worst overshoot per probe.
- Three harness entries: a round clamp `min(t,5000)`, a floor
`max(t,100)` and an offset `t+100`.
Stressed with two busy threads per logical CPU (16 on 8), five rounds of
every timed scenario on both targets. At normal priority, Windows put a
single poll up to 127 ms late and the fastest of ten 15 ms attempts up to
87 ms late, and a 45 ms tolerance failed the unmutated bridge in four rounds
of five. The timed scenarios therefore raise their own process to
HIGH_PRIORITY_CLASS on win32-x64, which needs no privilege. With that, the
worst single-poll overshoot was 15 ms on win32-x64 and 41 ms on linux-x64,
at normal priority there. The fastest 15 ms attempt was 0 ms late on both.
The limit in source is measured, not asserted. Each boundary mutant was
compiled in and run on both targets:
- `t+25` survived and `t+40` died;
- `max(t,40)` survived and `max(t,50)` died, on linux-x64 by 5 ms;
- `t*106/100` survived and `t*108/100` died;
- `min(t,7285)` survived and `min(t,7275)` died.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
docs/cultnet-selection-cut.md section 6/7/10: the reflection loop at
CultNetDocumentRegistry.cs:325-342 exists only because nothing outside
GameCult.Caching can read a declared index value, reference, or numeric
member (Members/IndexAccessors are internal, no InternalsVisibleTo to
Networking). This adds the one public door instead: CultDocumentMemberView,
CultDocumentDescriptor.DeclaredMembers/TryGetIndexValue/TryGetIndexNumber/
ReferencesOf, and CultDocumentRegistry.ResolveTargetLeaves (D9, computed live
against the registry, never cached or persisted).
Two registration refusals land beside DiscoverMembers' existing
DuplicateSlotMessage/HiddenMemberMessage: D10 renames the ArgumentException
ToDictionary already threw for two differently-named members sharing one
index alias into a rejection that names the schema, both members and the
alias; D11 refuses a reference member whose CLR shape the cache cannot walk
(many:false needs CultRecordRef<T>, many:true needs
IEnumerable<CultRecordRef<T>> or IDictionary<CultRecordRef<T>, V>) instead of
silently answering it as unreadable later.
GameCult.Networking is untouched; the evaluator that will read this surface
does not exist yet (commit 1).
Verification: dotnet test tests/GameCult.Caching.Tests - 188 passed (176
existing + 12 new), 0 failed. New tests: TryGetIndexValue/TryGetIndexNumber
reading an inherited alias; ReferencesOf over a single/many-list/many-dictionary
reference, including two dictionary entries with different payload bytes
(S19); ResolveTargetLeaves against an abstract target with no [CultDocument]
of its own (D9), including a leaf registered after an earlier descriptor was
built; IsNumeric agreement for a nullable numeric and a decimal member (S22);
the two new registration refusals, dynamically emitted so registry discovery
by attribute scan cannot see the rejected shapes.
scripts/mutate-cultnet-selection-caching.ps1: a dedicated dotnet/NUnit
mutation runner (the shared tools/eureka-mutations.ps1 harness only speaks
cargo; scripts/mutate-cultmesh.mjs is the QUIC campaign's file and this
campaign does not edit it, per the map's section 10). Byte-exact restore
verified by SHA-256, a sidecar for crash recovery, M0 no-op control, one
anchor match required per entry. Six entries (D8/S22, D10, D11 - revert and
loosening each) all KILLED; final restore verified green.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
`*.sh text eol=lf` reaches new clones only. A clone checked out before the
rule keeps scripts/build-quic-native.sh with carriage returns after pulling,
because the blob did not change: git neither rewrites the file nor reports it
modified, and the container still dies on the interpreter `bash\r`.
The README now gives the repair where a newcomer meets the container path,
and .gitattributes points to it. Verified on clones made at 367c9bc and moved
to c25de5a, from Git Bash and from Windows PowerShell 5.1: 109 CR bytes
before, none after, status clean. The line uses `;` so it runs in both.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
1.0.60 removed public API (CultInspectorAssetPath) under a patch bump; its
own changelog says "Breaking" and the version number lied. Operator ruling:
semver is policy, strictly, for every published package family.
docs/semver-policy.md states what counts as breaking per surface (managed
API, Unity inspector attributes, CultCache wire format/store compatibility,
CultNet protocol, native bridge exports), the 0.y.z pre-1.0 convention, tag
naming, and required changelog shape. Records the 1.0.60 scar as history;
does not renumber it.
scripts/check-changelog-semver.mjs is the mechanical check: refuses a
changelog "### Breaking" section under too small a bump, and refuses a
version that skips, reverses, or repeats the previous published tag. Skips
re-litigating a version that is already tagged (rebuilding immutable
history is not a release action). Wired into publish-packages.yml for the
four npm/PyPI packages, and into build-unity-package.ps1 (cultlib),
packages/cultmath/scripts/build-unity-package.ps1 (cultmath), and a new
scripts/verify-caching-unity-release.ps1 (caching.unity has no build script
of its own to hook into).
18 node:test cases cover the required table (breaking+major/minor/patch,
non-breaking, 0.y.z, skip, downgrade) plus CLI-level git-tag resolution.
Hand-mutated each rule (bump classification, pre-1.0 lane requirement,
breaking-heading detection, breaking early-return, already-published skip)
and confirmed each kill, then restored byte-exact. No mutation harness is
committed; CultLib has none today.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Soul S1: the 1.0.60 entry claimed the OpenSSL MsQuic bridge had shipped
since 1.0.59; git says both blobs first appear in this release. S2: the
package build staged package.json and README only, so the produced package
and the tracked template differed by the changelog.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CultAssetGuidKey.cs (and the test file added in 0b88706... no, 6c7cfa7) had a
literal LF working-tree checkout under core.autocrlf=true / text=auto, so a
build run from a stale bypassed-checkout copy of this worktree embedded a
source checksum a fresh clone's CRLF checkout would never reproduce. The
tracked git object was already normalized; only the working-tree bytes in
this worktree were wrong. Restored the CRLF checkout (git add reports no
diff against the index for CultAssetGuidKey.cs itself), then rebuilt with
scripts/build-unity-package.ps1 -UpdateTemplate. GameCult.Caching(.pdb) and
every managed plugin that depends on it (GameCult.Caching.MessagePack,
GameCult.Mesh, GameCult.Mesh.Quic.Native, GameCult.Networking,
GameCult.Networking.WebSockets) picked up new MVIDs/PDB checksums as a
result; the native gamecult_mesh_quic_native.dll/msquic.dll are untouched
since the native build did not run.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
*.meta is gitignored repo-wide; every tracked Unity .meta file, these two
included, is force-added on purpose.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CultLib 1.0.60: CultInspectorAssetPath removed for CultInspectorAssetGuid +
CultAssetGuidKey (breaking for inspector annotations), and the native QUIC
bridge's Schannel to OpenSSL MsQuic 2.5.9 static-link change shipping since
1.0.59. Caching Studio 1.4.0: the GUID picker, sub-asset keys, built-in
refusal, and the stale-value warning/Clear button. build-unity-package.ps1
only stages files it names explicitly, so it neither drops nor overwrites
these; no script change needed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
TrimEnd('0','f') missed Library/unity default resources built-ins whose GUID
has no all-zero or all-f suffix (e.g. 0000000000000000e000000000000000, the
LegacyRuntime font / Cube mesh in 6000.3.24f1). Ask AssetDatabase instead: a
real project asset is always a main or sub-asset under Assets/ or Packages/;
built-ins are neither. Also add a Clear button beside the stale-value warning
so a non-resolving stored key can be cleared without picking a substitute
asset first (drawing still never rewrites a stored value on its own).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Soul flagged (F3) two docs still describing the pre-release native QUIC
bridge: unity/org.gamecult.cultlib/README.md said Schannel, and
native/GameCult.Mesh.Quic.Native/README.md still described the older
40,448-byte committed plugin as current. The shipped bridge (since the
1.0.60/1.4.0 release, commit 0155ba8) is the OpenSSL MsQuic 2.5.9 build,
302,080 bytes, static CRT, no CRT import. Both docs now describe that build
and name what it replaced.
The OpenSSL build's NuGet package carries an Apache-2.0 notice for OpenSSL
itself (Microsoft.Native.Quic.MsQuic.OpenSSL's THIRD-PARTY-NOTICES) that the
committed Third Party Notices folder never carried - the Schannel build had
no such dependency to notice. build-quic-native.ps1 now copies it out
alongside MSQUIC-LICENSE.txt, build-unity-package.ps1 stages it into the
package output and the tracked template, and it is now committed as
OPENSSL-NOTICE.txt.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Ruling S (2026-09-17): a stored [CultInspectorAssetGuid] value is the
Addressables key form itself, the bare GUID for a main asset or
"guid[subAssetName]" for a picked sub-asset (one sprite of a sheet, say) -
the same shape AssetReference keeps and Addressables.LoadAssetAsync<T>
accepts.
CultAssetGuidKey (engine-free, GameCult.Caching) pins the parse/format rule
with unit tests (bare guid, guid[name], names with brackets/spaces,
malformed input), so headless checks can share it without touching Unity.
The Studio drawer's AssetGuid branch now: resolves either key form on read;
refuses to store a built-in resource's GUID (it has no Addressables entry)
and logs why instead of silently writing nothing; shows a warning next to a
stored value that fails to parse or resolve instead of drawing it as an
ordinary empty field, and never rewrites that stale value on its own.
Rebuilt GameCult.Caching/GameCult.Caching.MessagePack plugin DLLs
(scripts/build-unity-package.ps1 -UpdateTemplate) since the new type ships
in GameCult.Caching. Compiled CultCacheStudioDrawers.cs and
CultCacheStudioWindow.cs against Unity 6000.3.24f1's editor reference set
plus these plugins in a scratch csproj (removed after): 0 errors.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Removing the public asset path attribute is breaking. Runs
scripts/build-unity-package.ps1 -UpdateTemplate to refresh the tracked
plugin DLLs/pdbs in unity/org.gamecult.cultlib/Runtime/Plugins, and bumps
unity/org.gamecult.cultlib/package.json to 1.0.60 and
src/GameCult.Unity/Assets/Caching/package.json to 1.4.0 (dependency bumped
to cultlib 1.0.60). Tags are the next, separate step.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CultInspectorAssetPathAttribute stored an engine asset path, which breaks
when the asset moves. CultInspectorAssetGuidAttribute stores the asset's
stable identity (in Unity, the .meta GUID) instead. The Studio drawer now
resolves the stored GUID to a path only to load the preview object, and
writes back AssetPathToGUID on a change. No path form survives for the
next field to pick up.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The last report quoted the restored sha256 as though two runs agreeing on it
meant they had run the same source. They had not: `text=auto` gives a Windows
checkout carriage returns, so the win32-x64 run restored to b163cab7 and the
linux-x64 run to e3c6528a from the same commit. The digest is a restore check —
the file this run started with is the file it ended with — and it is said where
it is printed, so the next report does not borrow it for something it cannot do.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The note beside the close's wake claimed the narrowing to `notify_one` was
equivalent, and gave the weaker half of the reason. The equivalence needs two
premises together: `closing` is set permanently under the gate before the notify
and appears in every waiter's predicate, so after a close begins no poller can
block at all and the set of waiters is fixed; and each exiting call's own scope
destructor takes the gate, decrements and wakes everyone, so one wake starts a
chain that reaches all of them. Neither carries it alone. Both are now written
where the claim is made, so it can be checked rather than believed.
Narrowing both wakes also survives, and that is not equivalent by any argument:
a wake consumed by the closer, waiting on the same variable for the count to
reach zero, leaves the chain stopped and the rest of the pollers sitting out
their timeouts. It passes on wait-queue ordering — the pollers are queued before
the closer, so a FIFO head is never the closer until it is the only waiter left.
Recorded as unreached, with the reason, and deliberately not as unreachable:
that claim has been wrong twice in this campaign. A scenario cannot force it
while the first premise holds, since the only way to put the closer ahead of a
poller in that queue is a poller that blocks after the close began. Reaching it
would need a seam that chooses which waiter a notify goes to. The standard does
not promise FIFO, so what survives here is the implementation's doing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The last pass's only non-comment source change was held by nothing. Reverting the
hold's guard so it fires unconditionally survived all five scenarios, because no
scenario ever had a call in both states at once: the timeout scenario never arms
the hold, and the hold scenarios never let a timeout expire.
`holdtimeout` is the missing state. The hold is armed, the runtime is idle, and
one poll asks for 150 ms while the hold stays armed for 800. A call the wait woke
is the seam's to park; a call whose own timeout expired is on its way out, and
parking it would let `holdclose` report the bridge keeping a host call inside the
library when what kept it was the fixture. The measurement is which of those two
numbers the poll comes back at.
Two entries, and the loosening is the one worth reading: the same two conditions
in the other order. The hold blocks, so the order is the rule — asked first, it
parks the call and then decides whether it should have. Both die on
`holdtimeout` at about 800 ms and on nothing else; `holdclose` and `latecall`
stay green under both, which is the point.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The two bands ruled out a constant and ruled out nothing else. Any mapping that
is the identity at 200 and 900 passed for free, and the late tolerance was a flat
300 ms, so anything within that of the truth passed at every value forever.
Three ordinary ones survived all five scenarios on both targets: clamping the
wait to at most a second, adding 150 ms to it, and scaling it by six fifths.
The clamp is the one that matters. Its failure is verbatim the bug the last pass
claimed to fix — a host asking for five seconds gets one, forever, with nothing
reporting it — and "cap the wait so a shutdown gets noticed" is the most ordinary
spelling that line will ever be given.
So the long probe goes to 5000, above any ceiling a bridge would plausibly clamp
to, which kills the clamp and the scale on the early check and the late one. And
the late tolerance becomes a fraction of what was asked for — 60 ms plus an
eighth — which is what kills the offset: 150 ms is 75% late at the 200 ms probe
and there is no flat gap left to hide in. Measured overshoot on either target is
about 10 ms, against 85 ms allowed at the short probe and 685 at the long.
The table gets the clamp as an entry of its own, because the whole lesson here is
that a table of constants says nothing about a wait computed from the argument,
and a table of constants would go green again the moment these numbers drifted
back. Its honest limit is stated where the tolerance is defined: a scaling
smaller than an eighth survives, and nothing here claims otherwise.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The container recipe carried a PowerShell backtick continuation in a snippet
whose next paragraph invites a POSIX shell to run it, where a backtick is not a
continuation. A continuation has to pick a shell, so the line does not have one.
The Dockerfile's own copy of the recipe still mounted one machine's checkout by
absolute path; it mounts whatever checkout is running it, like the other two.
And the requirement a newcomer meets by tripping over it: on win32-x64 the
harness needs the checkout somewhere short. MSBuild's file tracker gives out on
long paths, so a run from a deep temporary directory fails the build for every
mutant, the no-op control included. That is loud — the harness stops on a red
control and prints what the child said — but what the child says is MSBuild, not
the path.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`* text=auto` and nothing else, so a Windows clone writes every shell script with
carriage returns. The documented Linux path is a container that runs one of them,
and it dies in under a second with a complaint about an interpreter named
`bash\r`. Nothing in that failure names line endings, and the same clone builds
and tests fine everywhere else, so it reads as a broken image.
`scripts/build-quic-native.sh` is the only shell script in the repository and the
only file on those paths that an interpreter reads by shebang; the mutation
runner is Node and handles carriage returns itself, and the Dockerfile builds
either way.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
No constant sits in both bands, which is true and is not the rule. A wait that
caps itself at a second passes every scenario, and capping a wait so shutdown
gets noticed is the most ordinary way that line will ever be written. An offset
and a scale survive beside it.
A fresh Windows clone also cannot run the documented Linux path at all, because
nothing tells git to leave shell scripts alone and the container dies on a
carriage return in the interpreter line. The mount fix works; this sat behind
it.
The equivalence holds, but on two premises nobody had written down, and
narrowing both wakes survives while being equivalent to nothing.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The blind spot was not where the hold sat. No scenario ever let a timeout
finish a poll, because both ended their waits with a close, so the duration
governed nothing anyone could see. A scenario that polls an idle runtime and
measures elapsed against what it asked kills all three constants, and it asks
twice at values whose bands do not overlap, because one value cannot tell a
bridge honouring the host from one waiting on that same number of its own.
The probe that reached the two-phase poll through a closed port is committed,
the wake finally has the entry the table had been claiming, and a surviving
mutant is argued equivalent and handed to Soul rather than swapped for an
easier one.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Three rules the bridge has always promised, none of which anything committed
could see.
The timeout was the structural one. Both scenarios end their waits with a close,
so the duration the bridge waits on governed nothing either of them observes: a
hard-coded 2000 ms survived the whole Linux matrix and Windows, and a host asking
for five seconds would have got two forever with nothing reporting it. The fix is
not another entry at another constant. A timeout's only observable is how long
the call takes, so `polltimeout` lets an idle runtime's poll be ended by its own
timeout and measures it, twice at different values, because no single constant
sits in both bands. The seam's hold moves behind that too: it parks only a call
the wait woke, so a poll whose own timeout expired is on its way out rather than
held there by the fixture and counted as the bridge's doing.
The refusal during a close (the header's section 4) had no scenario at all. A
call admitted after the closer has begun increments the in-flight count behind a
wait that has already read it. `latecall` holds the close open with the seam and
makes a call into it.
And the two-phase poll was written off as needing a listener, a credential and a
connection. It needs a closed loopback port: the failure always carries a reason,
which is an 87-byte payload, in about a millisecond. Moving the pop above the
copy crashes on both targets.
18 killed, 0 survived, 0 skipped on linux-x64; 16 killed, 0 survived on
win32-x64, where the two ThreadSanitizer entries are honestly skipped. Both runs
restored the source to the same sha256, b163cab7d48ebc93.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The mount was this machine's own path, so the one documented way to build and
check the Linux bridge needed editing before it would run anywhere else. It is
the repository root now, named as the shell names it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A reference whose target contradicts its declaration is noticed on read in this
cut and refused by name. Writing the same check into the put path is intended
rather than optional: the declaration already names the target and the writer
already holds the record, so it is cheap, and leaving it to consumers means
every tool that writes records reimplements it and a tool that forgets looks
exactly like one that did not until something refuses.
The consumer story rests on less than it sounds. There is no Studio repository
and no project by that name; the tooling that enforces this today is one
console project in one game. That makes now the cheapest moment to move the
rule to its owner.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A generator runs against published schemas, and the published schema entry is
flat: the cache walks the base types to collect members and then throws the
hierarchy away. Fed that, a generator emits one flat class per leaf with the
inherited members copied into each, which is the opposite of building the
hierarchy up for you.
So the catalog carrying each leaf's base chain is a prerequisite of that
campaign rather than a nicety for naming families, and families fall out of it
once it exists. The campaign itself waits until this cut settles the shape it
would be generating against. No generator exists in the repository today.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A caller needing exact pairs asks once per schema. The limit is stated rather
than answered with a third addressing list.
Naming a family is parked, but not because there is nothing to name. In one
process a family is an assignability check; over the wire there is no type
system, and nothing publishes the hierarchy where a caller can see it. The fix
is a catalog entry naming each leaf's base chain, which is small work that
moves the catalog's shape, so it belongs in a cut already touching it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Floats are fragile, and every value in this vocabulary is already a string, so
decimal strings keep the wire to one value encoding rather than adding a
second. The recommendation this reverses had that backwards.
What it costs is a comparison rule, and that is where the danger sits: compared
as text, ten is less than nine. The cut owes a canonical decimal form and a
numeric comparison, both pinned, and it no longer owes anyone a sentence about
where exactness runs out.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A stranger can now rerun both targets from the committed container and the
README, which was the batch's point and holds. The rules are another matter.
The scenario holds its callers after the wait whose duration is the rule, so
a hard-coded two seconds walks through the whole matrix while the committed
entry only ever died because one constant was smaller than another.
The gap recorded as unreachable is reachable in fifteen lines against a closed
port, and the mutant crashes a plain optimised build on contact. Second gap
falsified today. The next batch fixes the cause once instead of adding another
entry at another constant.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The operator ruled ordinals, the two deletions and the leaf titles earlier
today, in the other campaign's map. This one still carried them as open, and
a stale live question is exactly what a later agent reads as a decision still
to make.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The cache already declares and persists what type each member is, so numbers
cost no migration. What they do cost is a public way to read a declared value
at all, because the evaluator today cannot reach one from outside the cache,
which is why the existing snapshot loop reflects. That is a hundred and fifty
lines, not sixty, and the ledger moves the wrong way to say so.
Three claims inherited from the earlier pass were wrong and were cut rather
than designed around. A fifth selector engine turned up in Rust. And the
cache's index maps keep one key per value, so an evaluator that reached for
them would answer a query about a whole family with a single row.
Watching a selection is mapped as its own cut, where nine overloads that are
a query language in disguise collapse into one.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The map carried a one-line invocation that installed no JavaScript runtime,
so it could not run the mutation runner, and said nothing about disabling
randomisation or relaxing seccomp. That is how the Linux path came to be red
for everyone except the agent that had found the flags by hand. The recipe now
lives with the image and the bridge README, and the map points at them.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The image the bridge is built and mutated in existed nowhere in the repository,
and the recipe that stood in for it installed a C++ toolchain and no JavaScript
runtime — so the container could not run the harness that checks the bridge it
had just built. It is a file now, pinned to the same debian:13 digest as the
release workflow, carrying ninja, Node, setarch and MsQuic's four runtime
dependencies.
The harness configures with `-G Ninja` on linux-x64, which is what
scripts/build-quic-native.sh uses. CMake's Linux default is Makefiles, the image
carries ninja rather than make, and the default configured nothing at all: the
first run in the committed container died in cmake. It said so, because the
control now prints what the child said, which is the whole point of that change.
The seccomp requirement is stated where someone hitting it will read it — the
bridge's README, beside the recipe — rather than being knowledge the runner
assumes its caller already has.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three selector engines evaluate the same two allowlists today, two of them
carrying identical copies of the same reflection, and projection is a delegate
no production code ever constructs. The map replaces all of it with one
selection a caller can state: any-of over the aliases the cache already
declares, a projection it asks for, an order, a cursor, and one hop along a
declared reference.
The audit is reported as measured rather than as hoped. About 326 of 21,720
lines collapse in CultMesh, and the repository grows by about 370, because the
Rust runtime gains a capability it never had.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Section 4 forbade any host call that begins after `cultmesh_quic_runtime_close`
starts, and promised in the same sentence that such a call is refused with -1;
section 6 repeated the promise for `cultmesh_quic_next_event`. The refusal lives
behind the runtime's own gate, so once the close has returned there is nothing
left to read it: the case the sentence describes is exactly the case where the
promise cannot hold.
The prohibition stands alone. The refusal is stated for what it is — a call that
races the start of a close, while the close is still running, is refused and
touches nothing — and no longer as a guarantee covering the call the section
forbids.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four entries the native target did not have. Two for the rule that a blocking
crossing is a counted host call: the revert deletes the call scope from
`cultmesh_quic_next_event`, and the loosening keeps the scope and its refusal but
drops it at the door, before the call blocks. Two for the rule that a positive
timeout blocks until an event or the close: the revert inverts the comparison,
and the loosening waits a token millisecond. All four were green before this,
because nothing checked that the pollers were inside the library at all.
The bounded-wait loosening goes from 1 ms to 50 ms. A 1 ms bound only ever lost
a race — the margin was a few hundred lock handoffs — and the same loosening at
50 ms survived on both targets. `holdclose` holds its pollers for 500 ms, so any
bound shorter than the hold now expires with calls still inside. A bound longer
than the hold is not distinguished, which is this entry's honest limit.
A run that skips entries covered those rules with nothing, and exiting 0 let a
win32-x64 run — where every ThreadSanitizer rule is skipped — be read as a green
gate. It exits 2 and says so, with a killed/survived/skipped line.
A red control arrived as a bare colour with the child's output thrown away,
which is how a configuration that could not start at all stayed invisible. The
control now prints what the child said. On linux-x64 the sanitizer
configurations are re-executed through `setarch -R`, because ThreadSanitizer
dies before `main` on the kernel's address-space randomisation; both that and
the seccomp denial behind `personality(ADDR_NO_RANDOMIZE)` stop the run with the
cause and the required invocation instead of being reported as kills.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The closerace scenario pinned the order of the count and the notify but never
that a host call is counted at all. Deleting the call scope from
`cultmesh_quic_next_event` outright leaves every configuration green — the
pollers finish before the teardown reaches them — and so does inverting the
comparison that makes a positive timeout block, which empties the library before
the close even begins. A scenario that only sleeps and hopes cannot tell any of
that apart from the real thing.
So the bridge says it itself, in development builds only (section 7 of the
header, behind CULTMESH_QUIC_DEBUG_ASSERTS, absent from every shipped binary):
the peak number of host calls counted inside the library, the number counted
when the close began its wait, and a hold that parks a call inside the library
until the scenario releases it. A held call returns the 0 an empty queue would
have and touches the runtime no further, so a bridge that frees the runtime
under it is reported by the scenario rather than crashing inside it.
`holdclose` drives the quiesce through that seam: the calls are counted, they
were still counted when the close began, and the close did not return while they
were held. `closerace` keeps hunting the race and now also refuses to pass when
a poller left before the close began, which is a scenario that raced nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The lifetime fix holds under a thread sanitizer and the assertion kills the
deleted wait on both targets. The harness defending all of it does not hold.
Every native kill this cut had reported came from scratchpads that no longer
exist, the committed Linux path dies before its scenario starts and hides the
reason, and the scenario stays green when calls are not counted at all or
when nothing ever parks.
Two rules are defended by nothing committed. They are recorded as undefended
rather than given a number that would look better.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The mutation suite covered only the TypeScript rules. The bridge's two runtime
lifetime rules now have entries of their own, each with a revert that puts back
the wrong spelling and a loosening that keeps the shape and gives away the
guarantee: a notify moved outside the gate, a notify outside it that fires only
on the last call out, the quiesce wait deleted, and the quiesce wait made
bounded. Their killer is the scenario runner, built through CMake rather than
npm, so the runner now takes a check per target instead of assuming a workspace,
and a target without a dist/ keeps the verified restore without the sentinel.
Every native entry names the platforms it is honest on, and a run elsewhere
reports it as skipped rather than killed. There is no ThreadSanitizer for MSVC,
so the notify rule cannot be seen on win32-x64 at all. Soul's Cut 3 pass found
this the hard way, reporting two mutations killed from a win32-x64 release run
where both in fact survive.
Naming a target on the command line runs only that one, which is what lets the
native entries run inside the Debian 13 container where the TypeScript workspaces
have no installed dependencies.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The bridge's close path had no committed test. Whether a close is safe against
host threads already inside the library is the one thing about it a host cannot
establish by reading, and every check of it so far lived in a session scratchpad
and is gone.
The scenario runner under tests/ drives that path and nothing else: it parks
pollers in cultmesh_quic_next_event, closes the runtime under them, and asserts
that the close is what woke them. It consumes the bridge through the public
header, as a host does. CULTMESH_QUIC_BUILD_TESTS builds it, off by default,
because a delivery build has no business producing an executable.
CULTMESH_QUIC_DEBUG_ASSERTS, also off by default, makes the bridge state the
quiesce invariant itself rather than leaving it to whatever a sanitizer happens
to catch. Deleting the quiesce wait is otherwise falsifiable only by luck:
AddressSanitizer saw it 4 times in 48 attempts with pollers blocked mid-copy of
large frames, and never on release. With the assertion and pollers blocked it
fails on the first close.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The README claimed a rebuild reproduces the committed Unity plugin DLL. It does
not. That plugin is 40,448 bytes, predates both the static CRT and the move off
Schannel, and still imports VCRUNTIME140.dll and MSVCP140.dll; the current source
builds to 302,080 bytes with no CRT import.
What /Brepro buys is that a build reproduces itself, and the README now says
that, along with the fact that the committed plugin stays the older build until
the Unity package build replaces it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Section 3 and cultmesh_quic_last_status both told a host that the portable answer
is the return code. What the mapping makes portable is the sign. The magnitude is
not: the same MsQuic failure carries different numbers on different platforms, a
TLS error being -12032 on win32-x64 and -1126 on linux-x64, and the 16-bit mask
gives QUIC_STATUS_USER_CANCELED and QUIC_STATUS_FILE_NOT_FOUND the same -1002 on
Windows.
The header now says the sign is what a host may branch on, that the magnitude is
diagnostic and platform-specific, and that switching on a particular code
identifies nothing reliably.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every error sentence widened its QUIC_STATUS straight to uint64_t. On Windows
that status is a signed HRESULT, so the sentence carried 0xFFFFFFFF80072B18 where
the status is 0x80072B18. The status is 32 bits on both platforms and now goes
through uint32_t on its way in.
Diagnostics only: the v1 managed connector matches on the wording of these
sentences, not on the hex.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A call scope used to drop the lock after decrementing the in-flight count and
notify afterwards. That left a window: the closer's predicate is already true
once the count reaches zero, so a scope preempted between the unlock and the
notify could come back to find the runtime torn down and signal a condition
variable that no longer exists. ThreadSanitizer reports it as a broadcast racing
the destroy.
The decrement and the notify now happen under one lock guard. The closer cannot
return from its wait until the scope releases the gate, so the condition
variable is alive for as long as anything can signal it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The cut does not close. The lifetime model was right in shape and wrong by
one line: the last call notifies after it has released the lock, so the
runtime can be destroyed in between and the wake lands on freed memory. A
sanitizer found it on the unmutated bridge.
The mutation that survived by construction is ruled killable rather than
accepted. Under a sanitizer with pollers blocked mid-copy it fails four times
in forty-eight, so a dev-only assertion makes it fail every time instead. Two
further mutations are honest only on the sanitizer target and the table must
say so, because a mutation whose kill depends on the target is not killed
until it runs there.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The bridge keeps an object alive across the call that names it, closes
each handle exactly once, and quiesces before it frees. The contract is a
header hosts and tests are checked against. Return codes are negative on
every platform, one completion per frame, and every connection's end is
observable. One mutation survived by construction and is handed to Soul
to rule on.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Q9 ruled A. The Schannel MsQuic refuses QUIC_CREDENTIAL_TYPE_CERTIFICATE_PKCS12
with QUIC_STATUS_NOT_SUPPORTED, and PKCS12 is the only credential type a
provider has on both platforms, so a Windows provider could not open a
listener against the runtime this repo was shipping. The same bridge
binary passes all twenty smoke checks and the v1 gate against the OpenSSL
build, which is what Soul verified and what this now downloads.
Two digests, not one. The zip is pinned by FEE9A664 and 19,179,817 bytes;
the msquic.dll inside it by C17C6581 and 4,181,856 bytes. The earlier
report and the cut map's first record put the zip's digest on the DLL,
which is a pin that checks the archive and then ships a file nobody looked
at. Sizes ride along with both, because a digest that fails tells you less
than a digest and a length that fail together.
The CRT is linked statically. A Node or Unity host without the Visual C++
redistributable cannot load a DLL that imports vcruntime140.dll, and
Windows reports that as "the specified module could not be found" while
naming the module that was found. Confirmed on the rebuild: msquic.dll,
CRYPT32, ntdll, KERNEL32, and no CRT import at all. The DLL goes from
83,456 to 302,080 bytes, which is the price of the host not needing
anything installed.
The Unity plugin's committed msquic.dll is still the Schannel one and this
does not touch it; the next build-unity-package.ps1 run is what replaces
it, and that is not in this batch.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Event 7 was not one per frame. The bridge writes the stream's kind byte
itself, that send completed as an anonymous event 7, and a host counting
completions saw six for five frames with nothing to tell them apart. The
send request now says which send it was; the kind byte's completion is the
bridge's own business and is dropped. A send MsQuic released without
putting on the wire reports it: event 7 carries a send result in `code`,
completed or canceled, so an aborted send is accounted for rather than
silently missing. Measured: 32 sends queued under an abort, 32 event 7, 28
canceled and 4 completed.
The initiator of a shutdown learned nothing. MsQuic gives it only
shutdown-complete, and the bridge swallowed that, so a host could end its
own connection and never see the id end. Event 4 now has one emitter —
shutdown-complete — and the two initiated-by callbacks record why rather
than announcing it. Each connection id gets exactly one event 4, whoever
ended it, with the code its side passed, and it is emitted after the id
has stopped resolving, so a host told a connection is gone cannot still
resolve it on the next line.
Return codes inverted by platform. QUIC_STATUS is a negative HRESULT on
Windows and a positive errno on POSIX, so a Linux host checking for a
negative code saw every failure as success, against a README that promised
negative. A MsQuic refusal is now -(1000 + (status & 0xffff)) on both, and
cultmesh_quic_last_status hands over the raw bits for diagnostics. On
win32-x64 a garbage PKCS12 returns -12032 with last_status 0x80072B18.
The v1 exports move with it; the managed connector tests the value against
zero and formats it, so the contract it was shipped with holds.
Smaller ones, from the same pass: AppendTrace is v1-only and bounded,
because a long-lived v2 connection grew that string once per MsQuic event
for nothing to read; a listener configuration failure takes the connection
out of the map instead of leaving it naming a handle MsQuic is about to
close; and next_event refuses a null payload with a non-zero capacity
rather than treating the lie as a discard.
Soul's C consumer now compiles against the header and asserts its own
struct and its eleven bound prototypes against it, and the koffi smoke
checks its declarations and the thirteen exports the same way: 20 of 20,
seq clean, cancel clean, the v1 gate 3 passed 1 skipped as before.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
FindStream, FindConnection and FindListener took the lock, dropped it, and
handed back a raw pointer. MsQuic's worker freed the object on
shutdown-complete and the host thread dereferenced it on its next send.
One release run in three crashed on each target under an ordinary
peer-driven stream teardown; ASan caught it three times out of three, with
the free inside DestroyStream on a worker and the read inside
stream_send_frame on the host thread. A null check cannot fix that: the
pointer is not null, it is dead.
The maps hold shared_ptr now and a lookup returns a reference the caller
owns for the length of its call. The worker's destroy takes the object out
of its map and drops the map's reference only. The MsQuic handle follows
the object: the destructor closes it exactly once, under the exchange
guard, so the handle is alive for as long as any holder is. A stream holds
its connection, which orders every stream handle's close before its
connection's without a rule anywhere saying so.
runtime_close quiesces. It marks the runtime closing, wakes every blocked
poller, and waits for every in-flight call to return before it closes a
handle or frees a byte; a call that arrives after that is refused with -1.
Waking a poller without waiting for it was how close returned a poller
into freed memory while looking like it supported the pattern.
Two things fall out of the model rather than being fixed on top of it: a
listener that fails partway through opening leaves no map entry and no
handle, because every failure path just drops the local reference, and a
stream is in the map before StreamStart, because the shutdown-complete
that StreamStart can reach on a worker has to find it.
Cut with it: Listener::stopped, V1Client::connection (an unread raw
pointer into the same hazard), and FindListener.
Event 7 is still one too many and the shutdown initiator still learns
nothing; that is the next commit. The rest of Soul's C consumer passes on
win32-x64, including race at 400 streams, which crashed before.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Soul wrote a C consumer of this ABI from the spec text, because there was
nothing to include. The struct layout, the exports and the conventions
lived in the .cpp and in prose, which is three places for one contract and
no way to notice when they part.
include/cultmesh_quic_native.h is now the contract: the 64-byte event with
every field offset asserted, the exports with their signatures, the id and
return-code conventions, the object lifetime rules, and the threading
rules. The .cpp includes it and keeps none of its own copy. The README
points at it rather than restating it.
The lifetime and return-code sections describe where the bridge is going,
not where it is; the two commits after this one make them true.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The bridge's lifetimes are wrong: a lookup hands the host a pointer the
worker can free before the next call. Release crashes on both platforms
under an ordinary stream teardown. Three ABI gaps ride with it, and the
return-code contract inverts between Windows and POSIX. One digest in
this map was the zip's, not the DLL's. All in a fix batch before Cut 4.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
One credential type on both platforms, the provider role works, the ABI
is unchanged. The tree pays four megabytes more and the Unity plugin's
copy diverges until its rebuild. The map's claim that PKCS12 was one type
for both providers is withdrawn.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The bridge is a runtime-neutral two-role event queue, built and smoked
on both platforms with the Linux artifact named as development-loop only.
Two hangs the old shape never exercised are pinned. One fork stops the
Windows provider: the pinned Schannel flavour does not implement the
PKCS12 credential the spec assumed was one type everywhere, while the
same bridge passes beside the OpenSSL flavour. Recommended: pin OpenSSL
on Windows and pay the tree size.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
There is no Linux MsQuic binary in any NuGet package, and the v2.5.9 GitHub
release carries only test bundles, so the Windows script's approach does not
transfer. What does exist is Microsoft's Debian 13 pool, which is the same
distribution Yggdrasil runs, and the msquic source tree at the matching tag for
the POSIX headers the NuGet omits.
`scripts/build-quic-native.sh` fetches the pinned `libmsquic_2.5.9_amd64.deb`
and the three headers `msquic.h`, `msquic_posix.h` and `quic_sal_stub.h`, and
verifies all four by SHA-256 the same way `build-quic-native.ps1:14` verifies
the NuGet. A digest that stops matching is a changed dependency, not a warning.
Two details the deb forced. It carries three library files and no licence text,
so the MIT licence is copied from the one this repo already committed for the
Windows plugin — the same text for both platforms. And it ships only
`libmsquic.so.2.5.9`, with no `libmsquic.so` for `-lmsquic` to resolve against,
so the script makes that link in the extracted tree rather than in the image.
The Linux CMake branch links `-fvisibility=hidden`, so only the twelve v2
exports leave the `.so` and the Windows-only v1 five are absent by construction
rather than by discipline, and sets `RPATH $ORIGIN` so the `libmsquic.so.2`
shipped beside the bridge resolves without being installed system-wide.
`build-quic-native.ps1`'s default output directory becomes
`artifacts/quic-native/win32-x64` so the two platforms' outputs sit beside each
other instead of on top of each other. Both existing callers —
`build-unity-package.ps1:80-82` and the Native.Tests csproj `:20-22` — pass an
explicit `-OutputDirectory`, so neither is affected.
Built and smoked in the workstation's Docker `debian:13` engine pinned by
digest: g++ 14.2.0, glibc 2.41, `libgamecult_mesh_quic_native.so` 85,456 bytes
with `NEEDED libmsquic.so.2` and `RUNPATH $ORIGIN`, and the same 15-check v2 FFI
smoke that runs on Windows passes under Node 24 on that image. That is a Linux
build on a Linux host and nothing more: Q6 ruled the release artifact comes from
the Actions job, and this one is the development loop.
A Linux host still needs MsQuic's own runtime dependencies present — libssl3t64,
libnuma1, libxdp1, libnl-route-3-200 — which is now written down, because the
first smoke run failed on a missing libxdp.so.1 and the reason was not obvious
from the error.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The bridge was one platform, one role and one host: Windows by construction
(`windows.h`, `__declspec`, `CryptHashCertificate2`), client-only, and shaped
around the managed connector that polls it every 2 ms. Nothing about MsQuic
required that; the shape came from having exactly one caller.
It now carries a C ABI v2 of twelve exports over a single per-runtime event
queue: open a runtime, open listeners and connections on it, open streams and
send frames, and take typed events out of one blocking wait. The event struct is
64 bytes described by layout rather than by a shared header, so a host declares
it in its own FFI and no header travels with the ABI. `next_event` keeps the v1
two-phase convention — return 2, leave the event at the head, report the size
the caller needs — because that is the one part of v1 that was already
host-agnostic.
The rule that shapes everything else is that nothing calls out. MsQuic's worker
threads only push onto the queue and wake that wait, so no host runtime is
entered from a thread it does not own, and the bridge needs to know nothing
about what is on the other side. That is what makes one binary serve both a
Node process through FFI and a Unity player through P/Invoke.
Trust moved out. A v2 client connection defers validation, emits the DER as
event 3 and answers MsQuic only when the host calls
`connection_certificate_complete`; if the host never answers, MsQuic's handshake
timeout closes the connection and event 4 says so. The C code no longer has an
opinion about certificates — except in the v1 shim, where it must.
The five v1 exports are unchanged in signature and behaviour and are now
`#ifdef _WIN32`, matching `CanConnect`'s own Windows gate. A v1 client is a
private v2 runtime holding one connection, and the pin comparison is the single
thing it still decides in C, because that is the contract the shipped Unity
package was built against. `windows.h`, `wincrypt.h` and `CryptHashCertificate2`
appear nowhere outside that block.
Two things the rewrite had to get right that the old shape never exercised.
The v1 pin must be attached before `ConnectionStart`, not after it returns: the
handshake runs on an MsQuic worker thread and reaches the certificate event
immediately on loopback, so a pin applied afterwards loses the race and the
connection silently takes the host path nobody is listening on. And teardown
must close the connection handles itself — `RegistrationClose` blocks until the
app has closed every child handle, and a silent shutdown delivers no shutdown
events at all, so leaning on the callbacks to do it deadlocks the caller
forever. Both were found by the tests hanging, and both now have a mutation that
reproduces them.
Verified: the four v1 tests in tests/GameCult.Mesh.Quic.Native.Tests pass
unchanged (3 passed, 1 skipped for want of an external endpoint), including the
wrong-pin refusal and the departed-client broadcast test. A 15-check FFI smoke
drives the whole v2 path — listener, deferred certificate, both sides connected,
latest-only stream, one frame with FIN, identical bytes out — and passes on
Windows x64 and on Linux x64. Ten source mutations, ten killed, two green
no-op controls.
One open fork, reported separately and not decided here: the pinned Windows
MsQuic is the Schannel build, which refuses
QUIC_CREDENTIAL_TYPE_CERTIFICATE_PKCS12 with QUIC_STATUS_NOT_SUPPORTED, so
`listener_open` cannot open a provider against it. The map assumed PKCS12 was
"one credential type for Schannel and OpenSSL"; the header declares it but the
Schannel provider does not implement it. The identical bridge binary passes the
full provider smoke against OpenSSL MsQuic 2.5.9, so this is a question about
which flavour Windows pins, not about the bridge.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Both comments around `STRICTER_THAN_CSHARP` called the divergences host
repairs. Most are not. Of the eleven, three carry a backslash and two have no
authority for `System.Uri` to accept; `refusedByCSharpUriShape` refuses those
five on the raw string, before there is a host to repair. Four are genuine host
repairs `URL` would make and C# would not — two trailing dots, two fullwidth
hosts — and the last two throw in `URL` before any rule of ours runs.
The same slip sits in `isProtectedEndpoint`'s own doc comment, over a different
eleven: it says `URL` repairs eleven spellings `System.Uri` refuses, when only
the four carrying a tab or CR/LF are rewritten and the other seven are admitted
as written.
Wording only; no rule changes. It matters because the next reader deciding
whether the pre-check earns its keep will reach for the reason the comment
gives, and "host repairs" points at the wrong half of the list.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The codec's byte parity rests on two vector files written from opposite sides:
`realtime-frame-vectors.json` by the C# reference and decoded by TypeScript,
and `realtime-frame-vectors.ts-written.json` by the TypeScript codec and
decoded by C#. Neither half ran in CI. The TypeScript suite ran nowhere —
cultmesh-ts was not in any job — and the C# test ran nowhere either, because
`cultnet-interop.yml` is windows-latest and main-only and does not filter for
it. A codec that agreed only with itself would have gone green on every push,
which is the exact failure the two-file scheme exists to prevent.
Both now run in the `reactive-and-resource-lifetime` job beside the authority
steps, on windows-latest and ubuntu-latest, for main, codex/** and pull
requests. That job was already the one place the TypeScript side of CultMesh
gets exercised on Linux at all.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Two mutants the cut-2 pass recorded as equivalent were not. Both notes were
reasoning, and both were wrong in the same way: they argued from the shape of
the code instead of trying the input that breaks it.
The decoder's length refusal reads `payloadLength < 0 || expected !== length`.
The note said the first clause could never fire on its own, because a negative
payload length makes the sum smaller than the frame. It does not: the sum also
carries the three identity lengths, and a negative payload cancels against
them. A 37-byte frame declaring `channelLength=65535, payloadLength=-65535`
sums to exactly 37, matches its own byte length, and without the clause decodes
to three blank identities and an empty payload where `CultMeshRealtimeWireProtocol`
refuses it. Two smaller spellings do the same, `ch=1, pl=-1` over 37 bytes and
`ch=2, pl=-1` over 38. No test covered that shape; one does now, and the
mutation is listed and killed rather than excused.
The payload copy was worse, because the fix and the bug looked identical. The
decoder used `bytes.slice(...)`, which copies a `Uint8Array` — and the existing
test passes one, so it was green — but Node's `Buffer` overrides `slice` with
an alias of `subarray`. A `Buffer` is what a socket read hands a Node caller,
which is exactly how the QUIC plane in cut 4 will reach this function, so every
decoded payload would have been a live view of a frame buffer the transport is
free to reuse.
`Uint8Array.prototype.slice.call(bytes, ...)` was the obvious repair and does
copy, but `%TypedArray%.prototype.slice` takes its result constructor from
`@@species`, and `Buffer`'s is `FastBuffer`: the payload then comes back a
`Buffer` from a `Buffer` frame and a `Uint8Array` from a `Uint8Array` one, so
the same frame compares equal or unequal under `deepStrictEqual` depending on
how it arrived. `new Uint8Array(bytes.subarray(...))` copies and is one type on
both. Three mutations now guard the line — return a view, return a `Buffer`
view, return a species-typed copy — and the first two die only on their own
input type.
Mutation runner: 75 mutations, 75 killed, 0 survivors, three green controls.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`npm run build:ts` and `test:ts` have been dead since 8cb3b72, not 8dd5a45 as
first recorded. That commit renamed the package to @gamecult/cultcache-ts and
left `run-typescript-workspaces.mjs` asking npm for a workspace called
`cultcache-ts`, which no longer exists; npm refused at the first entry in the
list, so neither script reached any workspace at all. npm resolves a workspace
by manifest name, not directory, so the list moves whenever a package is scoped
— said out loud in the source now, since nothing else will say it.
The mutation runner is renamed `scripts/mutate-cultmesh.mjs`. It stopped being
about the route verifier alone when cut 2 gave it a third target, the realtime
frame codec, and a file named for one of its three targets invites the reader to
assume the other two are somebody else's problem.
`docs/typescript-quic-realtime-cut.md:46-47` names the old path; the map is
Self's and is updated separately.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Byte parity held both ways under Soul's own frames. Three small things
did not: a mutant declared equivalent was not, the payload copy is a view
for a Node buffer, and the frame parity has no CI executor. The breakage
of the aggregate scripts dates from the scoped rename, not the version
bump this map first named.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The frame codec exists with parity proven both ways from its first
commit, and its test found a real divergence before any consumer could:
the JavaScript decoder deletes a leading byte-order mark the .NET decoder
keeps. Two housekeeping rulings taken: the aggregate TypeScript scripts
have been broken by a rename for two days and get their one-word fix, and
the mutation runner is renamed for what it now covers.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A third target in the committed runner, `realtime`, over
`packages/cultmesh-ts/src/realtime-wire.ts`, killed by that package's own
suite. It gets the machinery the other two already have: a no-op control
through the same write path, the sidecar and its verified restore, and the
post-restore rebuild that proves `dist/` no longer carries a mutant, with
`0x31545343` as the sentinel its last mutation removes.
Thirty-seven mutations, one per rule the codec owns: each header field's
offset, width and endianness; the field order the reference fixed; the magic,
the stamped header size and wire version; the delivery ordinal; UTF-8
identities and byte-counting length prefixes; the identity and payload
ceilings; the ALPN, the two QUIC error codes and the two stream kinds; the
four encode refusals from the transport contract; each decode refusal and the
order they fire in; the byte-order mark; and the two aliasing rules, that
decode reads only the viewed bytes of an offset view and that the decoded
payload does not alias the frame. A mutation that changes the encoding
symmetrically still dies, because the vector bytes were written on the other
side.
All seventy-two mutations across the three targets are killed and all three
controls are green.
One mutant is not listed and says why in the table, as the P1363 length guard
already did: removing `payloadLength < 0` from the decoder's length refusal is
equivalent. The clause mirrors the reference but cannot fire alone, because
reaching it means the frame is already 37 bytes and a negative length can then
never make the sum match. It survived when tried.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
One codec, `packages/cultmesh-ts/src/realtime-wire.ts`, rule for rule what
`CultMeshRealtimeWireProtocol` does: the 37-byte little-endian fixed header
with magic 0x31545343, the delivery byte as the C# enum ordinal, the i64
epoch and sequence, three u16 identity lengths, the i32 payload length, the
stamped header size and wire version, then UTF-8 identities and the payload.
It touches no socket; QUIC stream framing stays transport-owned. The blank
reading in `Validate` is `IsNullOrWhiteSpace`, whose whitespace set is not
`String.prototype.trim`'s, so it borrows `isNullOrWhiteSpaceCSharp` from
`cultnet-ts` rather than growing a second reading of the same rule.
Byte parity is proven in both directions and in two processes, which is what
the authority vectors cost us to learn. `contracts/cultmesh/realtime-frame-vectors.json`
is written by the new C# test under CULTMESH_WRITE_VECTORS=1 and decoded and
re-encoded here; `realtime-frame-vectors.ts-written.json` is written by
`scripts/write-cultmesh-realtime-frame-vectors.mjs` with this codec, over
different identities and epochs, and decoded by the same C# test with the
reference implementation alone. Both files share one field shape, so each
side reads both with one reader. The two cases that would put megabytes into
a committed contract file, identities at exactly 65,535 UTF-8 bytes and a
64 MiB payload, are recorded as a build rule plus the digest of the whole
encoding; both sides build them the same way.
Three things the C# reference gets from its type system and TypeScript does
not, each refused rather than left to surface as a stray runtime error: a
`bigint` outside the i64 range, a delivery string the union cannot spell, and
a leading U+FEFF, which `TextDecoder` deletes by default and
`Encoding.UTF8.GetString` keeps. The last one is a real divergence the test
caught; the decoder now sets `ignoreBOM`. U+FEFF is not whitespace to .NET,
so such an identity passes the transport contract and reaches the wire.
There was no C# test for this codec before; this is also its first.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A SIGKILL leaves the sidecar behind, and the next run wrote it over whatever
the file then held. That is right after a kill and destructive after one: a fix
written to the source between the kill and the next run was reverted to bytes
from before the kill, with a line claiming a repair, and the run then reported
mutation results for code nobody had written.
The mutation table is the record of what a live run can leave behind: the
original (which is the sidecar, and also what the no-op control writes) or the
original with exactly one of that target's mutations applied. Repair now
recognises those and nothing else. A file matching neither stops the run with
its own digest and the sidecar's and changes nothing.
Reproduced both halves against the shared target. Killed mid-mutation, then
appended a line to the source: the rerun exits 1, prints both sha256s, and the
hand-written line is still there afterwards. With the mutant put back and the
sidecar unchanged, the rerun repairs as before and now names the mutation the
dead run was inside.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The unit separator that joins protocol ids sat in the source as a literal
U+001F: once in `cultmesh-authority.ts` and twice in its test. Every editor,
every diff and every review renders it as nothing, so `join("")` reads as a
join on the empty string and the transcript looks wrong to anyone checking it
against the C# reference, which spells the same separator as an escape.
The bytes were right, which is exactly why nothing caught it.
The three become escapes. The bytes do not move: the C#-written and the
TypeScript-written route vectors both still verify through the shared module,
and the sorted-transcript test still builds the same expected bytes.
A new test reads both files and refuses any byte below 0x20 that is not LF or
CR, so the class recurs loudly rather than silently. It earned its keep on the
first run by catching a literal U+001F in the comment introducing it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`isLoopbackEndpoint` hands its endpoint to `refusedByCSharpUriShape` before
`URL` sees it; `isProtectedEndpoint` did not, and `URL` repairs eleven
spellings `System.Uri` refuses outright. Every one of them answered protected
here and unprotected in the C# reference, which is the dangerous direction:
the channel-protection gate opened on a route `IsProtected` would have closed.
`wss:x`, `wss:host`, `wss:host:8443`, `wss:/host` and `wss:///host` have no
authority for `System.Uri`; `wss://host\path` and `wss://host:8443\` carry a
backslash; `wss://ho\tst/m`, `wss://host\t/m` and `quic://ho\tst` a tab, and
`wss://host\r\n/m` a CR/LF, all of which `URL` deletes.
Soul's 71 raw-shape spellings go in as a table with the answer the C#
`IsProtected` gave for each on 2026-09-16, and the test asserts the invariant
rather than equality: TypeScript may never call an endpoint protected that C#
refuses. Eleven of the 71 now differ the safe way round, C# protected and this
not, and they are pinned by name rather than excluded, so neither a twelfth
joining them nor one of them flipping passes quietly. Two throw in `URL` before
any rule of ours runs (an IPv6 zone id, a non-breaking-space host); the other
nine are host repairs `refusedByCSharpUriShape` refuses on purpose, because the
same repair would make `isLoopbackEndpoint` lie. None is dialable as a browser
CultMesh route.
The mutation that drops the pre-check from `isProtectedEndpoint` alone, leaving
`isLoopbackEndpoint`'s in place, is killed by that test.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Three Soul passes and three fix batches leave one verifier whose rules
match the C# reference as executed, parity pinned both ways and run in
CI on both operating systems. Three residuals ride with Cut 2. Two
statements this map made about its own batch were overstated and are
corrected: the NUL byte never caused a rewrite, and CI was not green
throughout.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
TypeScript now cleans a route the way the C# constructors do, reads
whitespace with C#'s alphabet, refuses the endpoint spellings the URL
parser would otherwise repair, and names the two signature refusals as
C# names them. Every probe reruns with zero reachable disagreements. Two
scars kept for the postmortem: a sidecar that reverted fixes it was meant
to protect, and a NUL byte that made git read source as binary.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Nothing ran `npm run test --workspace packages/cultnet-ts` in CI. The shared
route verifier's tests, which are the whole TypeScript side of the CultMesh
authority contract, ran only on a developer's machine, and the browser client's
tests ran nowhere. `cultnet-interop.yml` is windows-latest and triggers on main
only, so even that would not have covered Linux or a pull request.
Both now run in `cultmesh-portability.yml`'s `reactive-and-resource-lifetime`
job, beside the step that verifies the same vectors with the C# reference. That
job already matrixes windows-latest and ubuntu-latest, already triggers on
main, codex/** and pull requests, and already installs the TypeScript workspace
dependencies, so the steps cost two npm runs and no new setup.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Soul replaced the shared `isUnsignedCertificate` call in the browser's
session-acceptance check with a local empty-string comparison and every one of
the 18 browser tests still passed, because every test used an empty signature.
A whitespace signature is where the two readings part: the shared route
verifier accepts such a route as unsigned on a loopback local-development
endpoint, and an inline comparison then demands a provider proof for the
session the verifier just let through. The test now opens that session with
three spaces, U+0085 and a no-break-space/tab/newline mix as well as the empty
string, and asserts U+FEFF the other way, since it is whitespace to JavaScript
and a signature to C#.
The mutation runner grew a second target. `shared` is the one verifier, killed
by the cultnet-ts tests; `browser` is `packages/cultmesh-browser/src/index.ts`,
killed by the browser tests. A browser mutation cannot be killed by cultnet-ts
tests, so the killer is the target's own workspace, and each target carries its
own sidecar, control, sentinel and post-restore dist assertion.
Two repairs this pass earned:
- The sidecar is now removed inside the `finally`, right after the verified
restore. A throw mid-run left it behind holding the pre-run bytes, and the
next run "repaired" the file from them, silently reverting everything fixed
in between. That happened here, twice, before it was noticed.
- The NUL in the loopback pre-check's strip regex had been written as a literal
byte rather than an escape. It read the same, but git classified the file as
binary and stopped normalising its line endings, so the whole file diffed as
rewritten. Escaped, along with the ASCII host range beside it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`isLoopbackEndpoint` handed the raw endpoint to `URL` and read the host it got
back. `URL` is a repairing parser and `System.Uri` is not, so seven spellings
were loopback here and remote in C#: a trailing dot on the host, which `URL`
drops; `ws:127.0.0.1`, `ws:/127.0.0.1` and `ws:///127.0.0.1`, whose authority
`URL` infers; a backslash path separator, which `URL` reads as a separator;
fullwidth digits and letters in the host, which `URL` folds to ASCII; and a tab
in the host, which `URL` deletes. Each of those is a local-development bypass:
a route the C# consumer would have demanded a certificate for opened unsigned
here.
`refusedByCSharpUriShape` refuses exactly those, on the raw string, before
`URL` sees it. Leading and trailing C0 controls and spaces around the whole
endpoint are stripped first, because both sides accept them. The spellings
Hands probed earlier and C# accepts -- `127.1`, hex, decimal,
`[::FFFF:7F00:1]`, userinfo, `localhost:8443` -- are untouched.
The doc comment claimed equivalence with `System.Uri.IsLoopback`. It is not
equivalent and now says so: two divergences remain, both unreachable from a
browser CultMesh route, and both named. An IPv6 zone id throws in `URL` and is
a host-local interface name no browser socket can dial; `file:` and `mailto:`
have no `URL` host and are refused by `openSocket`, which takes `ws:` and
`wss:` only.
The test is Soul's endpoint list with the C# answer as the expectation: 82 of
the 88 it probed, the six unreachable ones named in the doc comment instead.
Mutations: the trailing-dot refusal dropped, the scheme-and-authority shape
inverted, the backslash refusal dropped, and the non-ASCII host refusal
dropped -- each killed by the table.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The C# reference signs a transcript over a *constructed* route, and the
constructors clean on the way in: `CultMeshAuthorityRoute.Clean` trims each
protocol id, drops the blank ones, keeps one of each ordinally and sorts
ordinally; `RequireNonEmpty` trims the runtime id and endpoint; `Require`
trims the provider key id and coordinates; `CultMeshRouteCertificate` trims
the Odin key id; a blank generation becomes the runtime id and endpoint
joined by "@". TypeScript cleaned none of it and transcribed the caller's
string, so a route C# signed and accepted was refused here the moment any
field carried padding or a duplicate.
Soul's tamper loop is the evidence: C# accepted a duplicated protocol id, a
padded generation and a padded Odin key id against the committed vector's own
signature, and TypeScript refused all three. The test now reruns those, plus a
padded protocol id, a blank one, a reversed list, a padded endpoint and a
padded provider key id, against the C#-written vector's signature. A padded
Odin key id must also find its root, so the trust map keys on the trimmed id
as the C# dictionary does.
`verseId` is left alone on purpose: it is not a route field, and
`CanonicalRoute` passes the caller's string into `Canonical` untrimmed. The
test pins that too.
`protocolId ?? "cultmesh.documents.v1"` is gone. A C# route with no protocol
ids is `Array.Empty<string>()` and transcribes as the empty string; reading
the handshake's `protocolId` as a one-element list wrote bytes the reference
never writes. `canonicalSession` matches its C# counterpart the other way:
it cleans nothing on the request, because the handshake message is what the
wire carried, and only the endpoint is cleaned, because the one C# passes is
`Route.Endpoint`.
Mutations: the dedupe removed, the blank-id filter removed, the generation
trim removed, the provider key id trim removed, and the Odin key id trim
removed from the lookup -- each killed by its test.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Two edge alignments that have to land together, because the test for the
first asserts the messages of the second.
`String.prototype.trim` is not `char.IsWhiteSpace`. JavaScript counts U+FEFF,
which .NET does not; .NET counts U+0085, which JavaScript does not. Every
`trim()` in the shared verifier stood in for a C# `IsNullOrWhiteSpace` or
`Trim()`, so both were wrong at those two code points: a U+0085 signature was
an unsigned route to C# and a bad signature to TypeScript, and a U+FEFF
signature the other way round. `isCSharpWhiteSpace`, `trimCSharp` and
`isNullOrWhiteSpaceCSharp` are that set, and `isUnsignedCertificate` and the
signature trim now use it. The test pins all 34 code points Soul probed on
both sides against the answers .NET 10 gave on 2026-09-16.
The C# `Verify` also has three signature refusals, not one: a signature that
is not base64, a decoded length that is not 64, and a signature that decodes
but does not verify. TypeScript collapsed the first two into the third and
lost the distinction the reference draws. Split, with each message pinned by
its exact C# string. `verifyP256` keeps its own checks: it is the silent
`VerifySession` path, which returns false for all three.
Mutations: the whitespace-set site reverted to `String.prototype.trim`,
U+0085 dropped from the set, and the two new messages merged back into
"invalid" -- each killed by its test.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The cut closes on its invariants. What is left is the gap between the C#
rules as written and as executed: the route is cleaned before the
transcript on one side and transcribed raw on the other, whitespace and
loopback are decided by different alphabets at the edges, and the browser
guard is enforced by name only. All in a final Hands batch before Cut 4
makes a TypeScript signer real.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The browser's local-development short-circuit reads "unsigned" through the
shared predicate, so the two TypeScript sides cannot disagree with each
other or with C#. The negative grep now names the two functions the fix
batch added. The generated contracts file's line-ending churn is recorded
where unassigned findings live.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The browser's local-development session-proof short-circuit tested
`!route.certificate`, while the shared verifier already treated an empty or
whitespace signature as unsigned, matching C# `Verify` (`:120`). A loopback
route carrying a blank-signature certificate therefore passed the verifier
and was then refused for lacking a provider key, where C# returns
`IsLocalDevelopment` and skips the session proof.
`isUnsignedCertificate` in `cultnet-ts/authority` is now the one owner of
that reading; the verifier and the browser both call it. A browser test pins
the local-development loopback open without a provider proof, the signed
loopback certificate still demanding it, and the remote counterpart refused
in both trust modes. The mutation runner gains two predicate mutations
(whitespace treated as signed; only a missing certificate treated as
unsigned), both killed by the existing whitespace-signature test in
cultnet-ts.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Both parity directions are now pinned in the repo and run in CI, the
public verifier copies its bytes, and the four edge rules match the C#
reference, with the loopback set probed rather than assumed. One scar: the
runner's multi-line anchors had never matched on a CRLF checkout, so two
earlier kill claims rested on an LF-on-disk state. One browser-side
divergence remains and is in Hands.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Every test run inside the runner rebuilds `packages/cultnet-ts/dist/`
from the mutant under test, so after the final restore of the source the
compiled module still carried the last mutant. The runner now rebuilds
the package after the restore and asserts `dist/cultmesh-authority.js`
contains a sentinel the last listed mutation removes, checked against the
list itself so a reordering cannot make the assertion vacuous.
The original bytes are also written to `<target>.mutation-original`
before the first write and removed after the verified restore. A run
killed mid-mutation is repaired from that sidecar at the next run's
start, instead of by whoever notices git status.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`CultMeshAuthorityProofTests` verifies both vector files with the C#
reference, but no workflow ran it. The portability workflow's
reactive-and-resource-lifetime job is the one that already builds and
runs `GameCult.Mesh.Tests` on every push to main and codex branches and
on every pull request, on Windows and Ubuntu, so it gains one step:
`dotnet test --filter FullyQualifiedName~AuthorityProof`. The interop
workflow was not chosen: it runs on Windows only, on main pushes only,
and runs no Mesh tests. It already covers the TypeScript direction
through `npm test --workspace packages/cultnet-ts`.
CULTMESH_WRITE_VECTORS stays unset in CI; the committed files are the
contract.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`contracts/cultmesh/authority-route-vectors.json` proves C# bytes verify
in TypeScript. Nothing in the repo proved the reverse; Soul's probe did it
once in a scratch directory. It is now a committed check in the same shape,
mirrored:
- `scripts/sign-cultmesh-authority-vectors.mjs` builds the route and
session transcripts with the shared module, signs them with fixed
TypeScript-owned test keys over awkward inputs (a quic scheme with a
query, three unsorted protocol ids, non-ASCII generation text, priority
13), self-verifies, and writes
`contracts/cultmesh/authority-route-vectors.ts-signed.json`.
- `CultMeshAuthorityProofTests.AuthorityRouteVectorsSignedByTypeScriptVerifyHere`
reads that file with keys taken from the file, asserts the protocol ids
are unsorted, that `Validate` and `VerifySessionProof` accept, and that
a priority-tampered route and a replayed nonce are refused.
- The TypeScript vector test runs over both files, so a transcript change
on the TypeScript side fails here before C# could trust a stale file.
Checked negatively: one base64 character changed in the committed
signature fails the C# test.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
verifyP256 handed WebCrypto `view.slice().buffer`. For a Node Buffer,
`slice` is `subarray`, so `.buffer` is the shared pool: a pooled Buffer or
a subarray of a larger allocation verified false while the same bytes as a
plain Uint8Array verified true. Copy exactly the viewed bytes.
Trust rules that had drifted from `CultMeshAuthorityProof.cs`:
- A certificate whose signature is empty or whitespace is an unsigned
route (C# `Verify` line 120): local-development loopback accepts it,
everything else refuses with the unsigned message.
- Loopback is `System.Uri.IsLoopback`: any 127.0.0.0/8 host, `::1`, the
IPv4-mapped `::ffff:127.0.0.1` (only `.1`), and the host names
`localhost` and `loopback`. WHATWG URL normalises `127.1`, `0x7f000001`
and `2130706433` the way System.Uri does. Probed on the C# side through
the public policy before the rule was written; the test pins that set.
- The root lookup runs before the validity window, so an expired route
under an unknown root is refused as untrusted on both sides.
- A policy listing one Odin key id twice is refused at first use, the
closest a plain object gets to the C# constructor throwing.
One test per rule, one mutation per rule in the runner's list. A fixture
whose protocol ids are unsorted and whose expected transcript is the
sorted one means removing the sort is now killed.
The runner now matches anchors in the file's own line ending. On an
autocrlf checkout the two multi-line anchors it already had ("route
transcript field order", "session transcript binds the nonce") matched
nothing, so the run threw at those entries instead of testing them.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The verifier is sound in both directions, the second one by probe. What
can go wrong silently is the verification story: a mutant left compiled
in an ignored directory, a C# vector check with no CI owner, a public
function that reads a pooled buffer wrong, and four edge rules that drift
from the reference. One map claim about error texts is withdrawn.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
One route verifier in cultnet-ts, the browser copy deleted, a C#-written
vector file, and a mutation runner with a control. Soul is on it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
One mutation per rule the cut map names, applied to the source bytes and
restored from the original bytes with a digest check; a no-op control
runs first through the same write path. A survivor is a rule no test
pins and fails the script.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
cultmesh-browser now verifies routes and provider session proofs through
cultnet-ts/authority. Its own verifyAuthorityRoute, canonicalRoute,
canonicalSession, canonicalFields, verifyP256, isLoopbackEndpoint and the
two base64 helpers are gone; the six route and trust types are re-exported
under their old Browser names as aliases of the shared types. randomNonce
stays: it is client behaviour, not verification. All 17 browser tests pass
unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Cut 1 of the TypeScript QUIC realtime plane: one TypeScript
implementation of Odin root lookup, the validity window, the canonical
route and session transcripts, P-256/P1363 verification, loopback
detection and the protected-scheme rule, published as
`cultnet-ts/authority`. Browser-safe by construction: WebCrypto,
TextEncoder, atob/btoa, URL, DataView; no node: import, and a test that
says so.
The protected-scheme rule is the C# rule (wss, https, any scheme
containing "quic") rather than wss-only, so the module is not wrong for
QUIC routes; the browser client still refuses non-ws(s) endpoints itself.
The transcript bytes are pinned by contracts/cultmesh/authority-route-
vectors.json, written by the C# reference under CULTMESH_WRITE_VECTORS=1
with fixed test keys and otherwise re-verified by both runtimes. The
vector carries a route certificate and a provider session proof.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The deploy tarball is a git archive, so committed files are the only ones
that reach Yggdrasil without a second mechanism, and the Windows pair is
already committed for Unity. Four binaries, about eight megabytes, with a
manifest and digests beside them. No operator question is open on this map.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The deploy tarball is a git archive of two package directories, so only
committed files reach Yggdrasil and cultmesh-ts does not ride at all
today. Under C the binding resolves binaries from an environment variable
or the package's own native directory, a workflow builds the Linux artifact
in a debian:13 container and attaches it to a release with a manifest and
digests, and the operator smoke ships its own archive. Registry publication
is parked with its seam named. One question remains: whether the four
binaries are committed or fetched at deploy time.
The corrected npm premise is recorded in the operator's words: reachable,
publication deferred by choice.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Ruling A assumed a publish job whose token is minted by logging in to
npmjs.com, and the operator's subnet is blocked there. That is also why the
existing cultcache-ts publish job has never run. Consumers keep file: paths;
the binaries ship in the CultLib tree the deploy path carries; the Actions
build attaches the Linux artifact to a release. Cut 6 and the lines of Cuts
4 and 7 that assumed a registry are flagged for remapping.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
All three recommendations were taken: unscoped cultnet-ts and cultmesh-ts
with only the native platform packages scoped; the Linux artifact built by
GitHub Actions in a debian:13 container and smoked on Yggdrasil; the lookup
source port plus a static source now, with the Node Odin rendezvous parked.
The status header now says no operator question is open and cut 1 has not
started, so a reader of the header alone knows where the campaign stands.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Seven cuts against main at f2cd2eb4, with the largest liability priced first
rather than discovered later. The native binding was the open question: a probe
loaded the committed msquic.dll through koffi under Node 24, drove a handshake
on MsQuic's own threads and closed cleanly, with no addon build and no toolchain
at load time. It is one cut, not the campaign.
The consumer audit found no package at zero external consumers and every
external consumer on a file: path rather than the registry, which is why
registry naming is an operator question rather than an assumption.
Two corrections against the target are recorded in the map rather than absorbed
silently: a cited transport filename that does not exist under that name, and a
registry path the target assumes for cultmesh-ts that has never existed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Eureka target for giving cultmesh-ts a QUIC realtime provider and consumer,
byte-compatible with GameCult.Mesh.Quic and the Unity native MsQuic connector.
First consumer: the StreamPixels service publishing to its Unity overlay.
Operator rulings 2026-09-16: overlay character, subathon window and scene
intents travel as QUIC state frames and the costume catalog as an HTTPS
content artifact; shared TypeScript route verification moves to a
browser-safe module; Windows x64 and Linux x64 only; datagrams out of scope.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Corrects the 1.0.58/1.3.0 release. Rebuilds the tracked plugins with
scripts/build-unity-package.ps1 -UpdateTemplate so they carry "" as the
unset CultRecordRef wire form in every position (nil still reads as unset)
and the edit-start Record snapshot. The native QUIC connector changes once
to its /Brepro form; later rebuilds reproduce it byte for byte. The Studio
package moves to 1.3.1 and depends on cultlib 1.0.59.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
b85a828 changed x86_64/gamecult_mesh_quic_native.dll with no native
source change. The two copies differ in six bytes: the PE header
TimeDateStamp and the debug directory timestamp. The MSVC link stamps the
build time, so every -UpdateTemplate run rewrote the committed DLL.
/Brepro on compile and link replaces the timestamps with content hashes.
Two clean builds from an emptied build root now produce identical bytes,
and the DLL embeds no build path. The committed DLL changes once to the
reproducible form; after that the release byte check covers it like the
managed plugins.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
BeginEdit cloned Document eagerly but Record lazily from record.Document,
so a cache change in place before Record's first read made the two copies
come from different states. BeginEdit now serializes the document once and
deserializes Document from those bytes; Record deserializes the same bytes
on first read. Record is the state the edit began from, and costs one
deserialize only when a drawer reads it, with no second serialize.
Each commit still spends the edit, so a slider drag begins an edit per
frame; that serialize was already paid for Document, and Record adds a
deserialize only in frames where a drawer reads it.
The new test mutates the cached document between BeginEdit and the first
Record read; it fails on the lazy clone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
452f928 made nil the unset wire form, including dictionary keys: a
Dictionary<CultRecordRef<H>, float> holding the empty key encoded it as C0.
Python msgpack refuses nil map keys under strict_map_key, its default, so
cultcache-py cannot decode such a record. Nil cannot be the one canonical
form; "" can, and every runtime reads it.
The writer always writes "". The reader still takes nil as unset, because
452f928/b85a828 stores and Aetheria's Cut 9 import wrote nil; such a record
converts to "" on its next save and is byte-stable after that. The
store-composition contract names "" for value and key position, requires
readers to accept nil, and tells other runtimes to write "".
The tests pin payload bytes, so each claim is consequential: an unset ref
writes "", reloads through a fresh cache and resaves byte-identical; a
nil-written legacy payload reads as unset and resaves as ""; a ref-keyed
dictionary with an empty key writes "", reads back unset and resaves
byte-identical. The nil writer fails every byte assertion, a reader taking
nil as a set key fails the legacy pair, and a reader taking "" as a set key
fails the read-back assertions.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Rebuilds the tracked plugins with scripts/build-unity-package.ps1
-UpdateTemplate so they carry the unset-ref wire fix and
CultInspectorEdit.Record. The Studio package moves to 1.3.0 for
CultInspector.Record and depends on cultlib 1.0.58.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A drawer saw only its value; Aetheria's schematic-shape drawer needs sibling
members of the owning document (texture path, aspect, hull tints).
The model owns the seam: CultInspectorEdit.Record is a second copy of the
stored document as the edit began, cloned on first read and kept for the
edit. The Studio lowers it as CultInspector.Record, so a runtime CultUI
panel gets the same seam from the same edit. The Draw signature is
unchanged, so existing drawers keep compiling.
Why not hand drawers the edit copy the Studio already draws: that copy is
what Commit upserts, so a write through it would be saved without passing
through the returned value, GUI.changed, or read-only metadata. The record
copy is read by nothing, so writing to it saves nothing and changes neither
the edit nor the cache. It does not show edits other drawers made earlier in
the same frame.
The Studio's separate document-type field collapses into the held edit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The formatter wrote default(CultRecordRef<T>) as nil but read nil back as
"", so re-upserting a loaded record changed its bytes (0xC0 became 0xA0).
One in-memory unset value: CultRecordKey.Value is never null, so
default(CultRecordRef<T>) equals a ref built from "". One wire form: an
empty key writes nil, and nil or a legacy "" reads back as default.
Load then save is byte-stable; a store holding "" converges to nil on its
next save.
Rust, TypeScript and Python have no reference type: a reference member is a
string in the consumer's own payload, so there is no sibling codec to align.
The store-composition contract now names nil as the one unset wire form.
The new test loads a record with an unset ref through a fresh cache,
re-upserts and flushes, and requires identical payload bytes; it fails on
the old formatter.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CultMath's hash is the shader-style frac(sin(x) * 43758.5453) and
returns float. Unity.Mathematics' hash returned a well-mixed uint, and
seeding from the float hash collapses: a 100x100 position grid gave 1625
distinct values in Aetheria's zone generator.
- math.pcg(uint), pcg3d(int3/float2/float3), and pcg4d(int4/float4)
follow Jarzynski and Olano, "Hash Functions for GPU Rendering"
(JCGT 9(3), 2020). pcg3d and pcg4d are the paper's code from page 32.
pcg is O'Neill's RXS-M-XS 32/32 output over one LCG step, with the
constants checked against pcg-c's pcg_variants.h.
- CultMath has no uint vectors, so pcg3d and pcg4d carry uint bit
patterns in int3/int4. The float overloads hash IEEE-754 bits via the
HLSL intrinsic asuint(float), which is new. pcg3d(float2) holds z at
0, the paper's route for 2 -> N inputs.
- The HLSL mirror gains the same functions, written over scalar uints
so the text compiles both in dxc 1.9.0.5402 (lib_6_3, HV 2021, clean)
and in C#. HlslSourceCompatibilityTests now compares int and uint
arguments and results by value; a wrong shift in the mirror's pcg4d
gives 208 mismatches.
- Known-answer vectors come from an independent Python transcription.
pcg3d separates a 100x100 half-unit grid (the full int3). One
component alone has one birthday collision there, as a 32-bit value
can. The same grid on the integer lattice has none.
- design.md states that the hashes are integer-exact in HLSL and on the
CPU: dxc wraps uint multiply and add modulo 2^32, as C# unchecked does.
- Bump to 0.2.3 and rebuild the tracked Unity DLL and pdb.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
dxc 1.9.0.5402 lowers float2/3/4 normalize(x) to x * Rsqrt(Dot(x, x)),
and DXIL.rst defines Rsqrt as 1 / sqrt(src). A zero vector normalizes
to NaN and a NaN component poisons every component. CultMath divided
by max(length, 1e-20), which returned 0 for a zero vector and a huge
finite component beside a NaN one.
- math.normalize(float2/3/4) is now value * (1 / sqrt(dot(value, value))).
- The HLSL mirror drops cultmath_normalize and its epsilon constant: it
now equals the normalize intrinsic, and nothing consumed it.
- HlslSemanticsTests pins zero and NaN normalize, the rsqrt bit pattern,
and zero length/distance. design.md states the rule and notes that
DXIL.rst's Rsqrt special-value table is a copy of Round's.
- GameCult.Geometry's iso-surface gave degenerate triangles (an inside
vertex on the isovalue) a zero normal through the old epsilon. It now
guards them explicitly, with a test.
- quaternion normalize keeps its identity fallback: quaternion is not
an HLSL type.
- Bump to 0.2.2 and rebuild the tracked Unity DLL and pdb.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Bump the package and assembly version and rebuild the tracked Unity
DLL and pdb from a deterministic build.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Aetheria Cut 8a needs int2(float2), clamp(int, int, int), and
frac(double). Fill them in CultMath as dxc compiles them:
- generate-swizzles.ps1 emits same-size conversion constructors and
math functions between float, int, and bool vectors. dxc lowers
float to int as fptosi (toward zero; NaN and out-of-range undefined),
numeric to bool as != 0 (fcmp une, NaN is true), bool to numeric as
0 or 1. floatN(int) splat keeps floatN(0) from going ambiguous
against floatN(intN).
- Int and int-vector clamp are min(max(x, a), b), matching dxc's
IMin(IMax(x, a), b); inverted bounds return b.
- frac(double) keeps double precision. dxc demotes double frac to
float, as it does exp and lerp; CultMath's existing double overloads
already diverge that way, and design.md now says so.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Bumps org.gamecult.cultlib to 1.0.57 and rebuilds its managed plugins from
main, so the Unity package carries the store-routing CultCache (one home
store per type, AddBackingStore(CacheBackingStore, Type[]), atomic commit)
instead of the Sep 3 assemblies. Install docs point at the new tag.
The build script now publishes without Source Link or the source revision,
with CI path mapping and cleared Release intermediates. The previous DLLs
named the commit and the worktree they were built in, so committed plugins
could never match a fresh build of the commit that holds them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Brings Cuts 1-6 of the Aetheria CultCache migration (one home store per
type, fixed routes, atomic commit with compare-exchange, parked SoA) and
CultMath 0.2.0 onto main, which had moved on with the CultNet contract
consolidation. The two lines touch disjoint files.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
REC-032 Soul finding: dist/contracts.js required "../../../contracts/cultnet/...",
escaping the cultnet-ts package tree entirely. That breaks every installed
consumer: the two-runtime sample script (packs and installs cultnet-ts into a
clean consumer), test-typescript-package-closure.mjs, and Heimdall's pnpm copy
of cultnet-ts (honours `files`, would 404 on its next vendor/CultLib bump).
Only Stonks (file: symlink into the live checkout) happened to still work.
Fix: keep CultLib/contracts/cultnet as the only tracked source, but have
tools/generate-swarm-contracts.mjs wipe and re-copy it into a package-local,
gitignored packages/cultnet-ts/contracts/cultnet/ every time it runs (build,
typecheck, test, test:interop, prepack, the two-runtime script, and Heimdral's
`npm run build` all already invoke it). src/contracts.ts and the swarm-contracts
manifest now import that package-local copy ("../contracts/cultnet/...") instead
of reaching three directories up. package.json gets back `files: ["dist",
"contracts"]` and `prepack` so a packed tarball carries its own schema copy.
Also revert fa00f3f's tsconfig.test.json rootDir widen ("." -> "../..") and the
matching test-script/dist-test path changes and __dirname-relative recomputation
in cultnet-interop.test.ts and runtime-presence-health.test.ts: that widening
was working around dist-test not receiving a copy of the (now out-of-package)
contracts, but nothing in this repo actually needs the wider rootDir once the
schemas are package-local again, and the widened paths never matched what CI's
interop lanes and package.json's own dist-test paths expect --
packages/cultnet-ts/dist-test/test/interop/cultnet-interop.test.js, not
.../dist-test/packages/cultnet-ts/test/interop/.... PR #3 CI run 34766892424 on
7d3a3dc failed the first interop lane on exactly this missing-file defect.
cultnet-interop.test.ts's interopSchemaPath already resolved correctly once
cultNetTsRoot was reverted (it was never touched by fa00f3f) and continues to
point at the canonical CultLib/contracts/cultnet/interop copy, which is fine
since it is only read by the spawned non-TS peer processes, not required() by
this package. The cultmesh-ts shard-catalog $id lookup fix from fa00f3f
(cultnet.shard_catalog_request.v0 -> the CultLib contracts URL id) is unrelated
to the rootDir defect and is left as is.
REC-032, Soul redirect on branch claude/rec-032-cultnet-contracts.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
git grep exits 1 when it finds nothing, and the GitHub pwsh step wrapper
ends the step with that $LASTEXITCODE, so the guard failed on exactly the
clean tree it should accept. End the step with an explicit exit 0 after the
checks; a real drift still throws first.
REC-032.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
REC-032 commit 2: fail cultnet-interop.yml before it installs Node
dependencies if either negative check from commit 1 finds anything:
a *.schema.json file living outside contracts/cultnet/, or a lingering
reference to the retired https://github.com/GameCult/cultnet-ts/contracts
(or .../integration/contracts) namespace anywhere but .voidbot/birth
snapshots. The old-namespace grep uses git grep -I so the deferred
Unity plugin DLL (D4, not rebuilt in this pass) doesn't trip a
permanent false failure.
Verified locally: staging a stray contracts/stray.schema.json and a
stray-id.ts referencing the old namespace makes the guard's logic
report failure for both; removing them makes it pass again against the
real tree.
REC-032.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
REC-032 commit 1 follow-up, found during required verification: moving
the CultNet schemas out of packages/cultnet-ts/contracts into
CultLib/contracts/cultnet broke npm test for cultnet-ts. tsconfig.test.json
compiled with rootDir "." (to cover both src/ and test/ as siblings), and
TypeScript's resolveJsonModule copy-into-output behavior only mirrors
files that fall within the declared rootDir. The old contracts/ folder
was a sibling of src/ (inside rootDir), so it got copied into
dist-test/contracts and the "../contracts/x" import resolved. The new
contracts/cultnet is two directories above the package, outside any
possible rootDir short of the repo root, so nothing got copied and the
"../../../contracts/cultnet/x" import 404'd at runtime under dist-test
(the production "dist" build was unaffected, since dist/ and src/ sit at
the same depth under the package).
Fix: widen tsconfig.test.json's rootDir to the CultLib repo root so
resolveJsonModule can find and copy contracts/cultnet into dist-test,
update the "test"/"test:interop" npm scripts for the resulting deeper
dist-test/packages/cultnet-ts/... output path, and correct the two
__dirname-relative computations in test/interop/cultnet-interop.test.ts
and test/runtime-presence-health.test.ts that assumed the old shallower
depth.
Also fixes packages/cultmesh-ts/test/cultmesh.test.ts, which still
looked up the shard-catalog builtins by their old bare
cultnet.shard_catalog_{request,response}.v0 schemaId; CultNetSchemaCatalog
keys by $id, which D1 changed to the CultLib contracts URL, so the
lookup returned undefined.
REC-032, decision D1.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
REC-032 commit 1: move the canonical 33 TS + witness-artifact-bundle
schema files (34 total) into runtime-neutral CultLib/contracts/cultnet,
move the interop-note schema (and delete the zero-consumer discovery-*
schemas per D3, tagged parked/cultnet-discovery-contracts) into
contracts/cultnet/interop, and delete the 27 C# and 24 Rust duplicate
copies after confirming byte-for-byte equivalence (only
cultnet.snapshot-request differs: TS carries shardId/shardEpoch, a
known drift already resolved in favor of the TS shape).
Rewrite every $id from https://github.com/GameCult/cultnet-ts/contracts/
and .../cultnet-ts/integration/contracts/ to
https://github.com/GameCult/CultLib/contracts/cultnet/(interop/); give
the two shard-catalog schemas URL ids per D1, replacing the bare
cultnet.shard_catalog_*.v0 identifiers; leave ghostlight.agent-state's
Ghostlight-owned $id untouched per D5.
Repoint every runtime's load path and id literal at the new location:
C# csproj EmbeddedResource + LogicalName, Rust include_str! paths,
TS contracts.ts imports + swarm-contracts manifest/tooling + package.json
files/prepack, and the id/path literals in the C#, Rust, Python and
Kotlin interop tests and peers. Python and Kotlin still build their own
schema JSON in-process (D2); only their base constants, id literals,
and (for Kotlin) contentHash literals were swapped. Every Kotlin
contentHash was recomputed from the final canonical files using the
runtime's own stable-stringify + SHA-256 function; all matched the
planner's precomputed values except cultnet.schema-catalog-response,
where the planner's proposed value did not match recomputation and the
freshly computed value was used instead.
REC-032, decisions D1/D2/D3/D5.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-09-13 17:27:41 +02:00
301 changed files with 40711 additions and 3300 deletions
throw "CultNet schema contracts drifted out of contracts/cultnet/ (REC-032). See CultLib/contracts/cultnet/ and rewrite any `$id back to https://github.com/GameCult/CultLib/contracts/cultnet/."
}
# git grep exits 1 when nothing matches, and the runner's pwsh wrapper
# ends the step with that $LASTEXITCODE; a clean tree must pass.
"description":"R-X (docs/cultnet-selection-cut.md, fix batch 4): CultNetErrorMessage is a MessagePack-CSharp map-mode object, so every declared key is written on the wire, nil included - routingHint, code and details are `required` here the same as cultnet.raw-document-header.schema.json's nullable fields, not optional just because a given refusal leaves them null.",
"type":"object",
"required":[
"schemaVersion",
"error",
"routingHint",
"code",
"details"
],
"additionalProperties":false,
"properties":{
"schemaVersion":{"const":"cultnet.error.v0"},
"error":{"type":"string","minLength":1},
"routingHint":{
"description":"Optional shard routing detail for an authority failure. Null for every selection/cursor refusal code (R-N).",
"type":["object","null"],
"additionalProperties":false,
"required":["shard","reason"],
"properties":{
"shard":{"type":["object","null"]},
"reason":{"type":"string"}
}
},
"code":{
"description":"Machine-readable refusal code (R-N). Additive: a v0 peer that does not know this field ignores it.",
"description":"R-S (docs/cultnet-selection-cut.md, fix batch 3): CultNetRawDocumentRecord minus payload/payloadEncoding - CultNetRawDocumentHeader has no such members at all, so a header must not $ref the record schema (which requires them). Matches MessagePack-CSharp's map-mode output: every declared key is written, nil included, so every property here is `required` even where its value may be null.",
"description":"R-S (docs/cultnet-selection-cut.md, fix batch 3): matches MessagePack-CSharp's map-mode output for CultNetRawDocumentRecord - every declared key is written, nil included, so every property here is `required` (present on the wire) even where its value may be null.",
"description":"Absent means every schema. Present and empty is refused at the door (selection_invalid), never treated as \"every schema\": omit the field instead."
},
"keys":{
"type":"array",
"items":{"type":"string","minLength":1},
"minItems":1,
"description":"Absent means every key. Present and empty is refused at the door (selection_invalid), never treated as \"every key\": omit the field instead."
},
"fields":{
"type":"array",
"items":{"$ref":"#/$defs/fieldPredicate"}
},
"cites":{"$ref":"#/$defs/citation"},
"cited":{"$ref":"#/$defs/incoming"},
"projection":{
"type":"string",
"enum":["header","document"],
"default":"header"
},
"descending":{"type":"boolean","default":false},
"limit":{
"type":"integer",
"minimum":1,
"maximum":200,
"description":"Clamped to 1..=200 by the answering server, not refused outside that range: an absent value or one outside the bound is treated as the nearer bound (absent = 200)."
"description":"Q-J: a canonical decimal string, never a JSON number. Matches ^-?(0|[1-9][0-9]*)(\\.[0-9]*[1-9])?$ and is not \"-0\"; any other spelling is refused typed at the door (selection_invalid), never normalised.",
"title":"CultNet Snapshot Response Raw Message v1",
"description":"R-S (docs/cultnet-selection-cut.md, fix batch 3): matches MessagePack-CSharp's map-mode output for CultNetSnapshotResponseRawV1Message - every declared key is written, nil included, so every property is `required` (present on the wire) even where its value may be null. `headers`/`documents` are still exactly one populated by projection (C5), but the OTHER key is present too, as an explicit null, not absent - the prior `oneOf: [{required:[headers]},{required:[documents]}]` matched both branches on every real page (both keys always present) and so never validated at all.",
"//":"CultNet typed selection, Cut 1 fix batch (docs/cultnet-selection-cut.md, Self's rulings for the Cut 1 fix batch, R-E and the shared-fixture ruling, 2026-09-22). The one row set both the C# reference and cultnet-rs build their parity vectors from. schemaId is each schema's real SHA-256 content-hash id, exactly as GameCult.Caching.CultDocumentRegistry computes it for the fixture's C# document types (leaf_a=ParityLeafA, leaf_b=ParityLeafB, citer=ParityCiter, citer_narrow=ParityCiterNarrow, tests/GameCult.Networking.Tests/SelectionParityVectorTests.cs) - CultCache.cs's Sha256(semanticFingerprint), which carries no '.vN' suffix of its own; the version is folded into the hash, never appended as text. Both writers read this file instead of building their own fixture rows.",
"//aliases":"Two DIFFERENT alias forms, now deliberate rather than a cross-runtime defect (S-12, 2026-09-22: the Rust alias port made this so - see docs/cultnet-selection-cut.md, 'Self's ruling, 2026-09-22: the C# reference's alias rule is the rule'). Both runtimes resolve a schema alias the same one way: strip a trailing .vN and compare the remainder against the descriptor's declared SchemaName (CultNetSchemaAliasMatching.Matches(candidate, CultDocumentDescriptor), src/GameCult.Networking/CultNetDatabase.cs:74-87; cultnet_rs::selection::schema_alias::matches ports it name-for-name and deletes the hash-stripping overload it used to carry). nameAlias is <SchemaName>.v9 and resolves through both runtimes under that one rule. hashAlias is <schemaId>.v1 and is refused as unmatched by both - it is not a wire form either runtime's production call sites ever produce or accept, and it is kept in the fixture specifically to pin that neither runtime resolves it.",
"//references":"R-AF, 2026-09-23: a reference's `references[].targetKey` is the only thing a stored edge actually carries on the wire - GameCult.Caching.MessagePack.CultRecordRefFormatter<T> serializes a CultRecordRef<T> as a bare key string (probed: MessagePackSerializer.Serialize on a real Design field produced the single msgpack string 'a-eq', nothing else). This fixture used to also carry a `targetSchema` per reference; it was never anything but documentation - SelectionParityVectorTests.BuildDocument and cultnet-rs's build_row both read only targetKey to construct the edge - and it is removed rather than kept as a field the wire does not have. Resolving which row an edge names is by targetKey alone, filtered to the declared role's target leaves (D9); see citer_narrow below for the shared-key shape that rule is pinned against.",
"//":"R-AF: a citer whose declared reference target is one leaf only (leaf_a), so a record key shared by a leaf_a row and a leaf_b row has exactly one candidate inside the declared target and exactly one outside it - the shape neither runtime's parity vectors covered before this cut (zero duplicate record keys across the original ten rows).",
"//":"R-D: 2233759.25 is exactly representable as an f32 and sits on a round-to-even tie between two shortest-form spellings; its exact canonical decimal is 2233759.25 either way. kind=tie so no weapon/armor/shared case can sweep it up.",
"//":"R-C: BMP private-use key (U+E000), same ordinal as the astral row below so the tiebreak - not the ordinal - decides order. kind=unicode, mass=0 so no unfiltered or mass>1 case reaches these.",
"//":"R-C: astral key (U+1F602, encoded as a surrogate pair whose high half 0xD83D is less than 0xE000 - a UTF-16 code-unit compare sorts this first; code-point order sorts it after the BMP row above).",
"//":"R-AF shape 1: the only row at this key is outside citer_narrow's declared target (leaf_a only) - resolving the edge below must refuse reference_outside_target in both runtimes, not silently accept an empty page. kind=narrow-fixture (not \"shared\") so this row never sweeps into shared_index_value_three_rows, which pins an exact count of 3.",
"//":"R-AF shape 2: two rows share this key, one inside citer_narrow's declared target and one outside it - resolving the edge below must accept, deterministically picking the leaf_a row, in both runtimes.",
"//":"R-AJ: a tied ordinal across two DIFFERENT schemas (leaf_a and leaf_b), not two rows of the same leaf like the astral/BMP pair above - the shape the row order's schemaId tiebreak needs and the original ten rows never had. Keys are anti-correlated with the real schema ids on purpose: leaf_b's real id sorts before leaf_a's (sha256:b1f2... < sha256:ff35...), so leaf_b gets the alphabetically-*later* key (\"zz-tie-schema\") and leaf_a gets the *earlier* one (\"aa-tie-schema\") - a schemaId-first order and a recordKey-first order disagree here on purpose, so only the correct composition produces leaf_b's row first.",
"//":"R-AO, 2026-09-23: 'dup-key' already carries two rows (leaf_a ordinal 11, leaf_b ordinal 12) - both are in citer's Design role's declared target ([leaf_a, leaf_b], unlike citer_narrow's narrow_ref above, which only ever had one in-target candidate at a shared key). This citer is the first row in this file (hence in both runtimes' row order) to name 'dup-key' through Design, so resolving it must deterministically pick the leaf_a row - the first in-target candidate in the caller's order - never the leaf_b row, in both runtimes.",