The runtime no longer trusts the authoring tool for a bin: a NaN Floor weighed NaN and nulled every
draw of its design. AetheriaStores.Open names the field and echoes no value.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The all-weights-zero branch of ChooseProduct and the required-item fallback of RandomProducts weighed
every product 1, so a bin doubled its maker's presence there. Both now weigh by Share alone (the
preference drops out, the bins' shares do not). The sum of a line's bins' pass fractions is computed
once per weighting pass instead of scanning the product table per candidate.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Aetheria.cc written only by the command, on the pinned CultLib. Replay the command at merges; never merge the blob.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A bin gets its line's design and maker, Bin = { Line, Floor }, and no roles or assembly. Resolve-then-write
as the other lines do; a replay is a no-op. The catalog is not yet applied.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
FactionProductData slot 7 Bin (schema id unchanged, additive as slot 6). Mint derives a bin's
distributions from its line on every lot, truncated by NextGaussianAbove (64 tries, then the cutoff).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Its effective quality .15 sits under Common's .2, which GetTier counts as -1; not a rule this test is for. Builds.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Kills the Stryker survivors in the lines this cut touches that a quality-reading test can reach. Builds.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Kills the hand mutant that prices a lot at its best part (the single-fill sleeper could not tell max from mean). Builds.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Heat now reads effective quality, which narrowed the Generated fit's drive spread below the test's 5 percent premise (probe: thermal exponents 1.227/.809 at base, 1.623/1.318 at head; NominalThrust ratio 1.0525 then 1.0436). The test fits the two Large Drives with lots staged at the low and high end of the makers' assembly range, chosen before any run, through the ship's own unequip and equip, and asserts the spread as a premise with its own message. Unverified until the Yggdrasil run.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Price, tier, equipped thermal and durability exponents, wear and the unequipped durability exponent now read the one effective quality; no reader keeps a path to the assembly alone. Tests pin each reader against a lot whose part sits below its assembly. Unverified until the Yggdrasil run.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Code from eureka/aetheria-release-faction-curated-place-names. Conflicts kept both sides: Faction holds Assembly (key 19) and Places (key 21); AetherDb dispatches assembly-from-maker, place-names and doctrine-catalog, and its usage names both; FlightTests counts the Faction wire at 22 slots (19 is Assembly, 21 is Places).
The two LFS files are the merge's resolution, replayed on the post-assembly bytes, never merged as text or by side.
GameData/Aetheria.cc, per h1's two-step replay: (1) AetherDb place-names apply built at the branch's commit 1 (740e1ee7, merged onto the assembly merge so key 19 survives), run on the post-assembly catalog: purges the Markov NameFile records while their type exists and authors 15 place lists; (2) place-names apply built at this merged tree, run on step 1's output: authors Miss Terri's list (1 list). Each step's second apply said nothing to do and left the bytes identical; the whole chain run twice from the assembly catalog gave bytes equal except the record write timestamps (masked, both hash alike). Decoded independently with cultcache-py: 16 factions, key 19 Assembly filled on the 15 makers (Ewan Hart has none), key 21 Places 20 names on all 16, key 9 nil on all, no namefile type among the 321 records.
Assets/Resources/Settings.asset: the post-assembly file (Rarity values removed) with the four NameGeneratorSettings lines removed (NameGeneratorSettings, MinLength 5, MaxLength 10, Order 3).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Code from eureka/aetheria-release-assembly-from-maker-r2. The two LFS files are not merged as text or by side; they are the merge's resolution, made as follows.
GameData/Aetheria.cc: master's catalog (0f5def17's, sha256 a4ff0eff...ebee9) is unchanged by master since, so the branch's blob was checked against the replay. AetherDb assembly-from-maker apply, built from this merged tree, run on master's catalog twice from fresh copies and once more on its own output: 15 factions written; the second pass said No changes and left the bytes identical. The replayed bytes equal the branch's blob (sha256 2b18ff7f...90e... ) except for the 333 record write timestamps, each of which a fresh write restamps (masked to a constant, both hash alike). The branch's blob, which h5's full suite tested, is taken.
Assets/Resources/Settings.asset: master's file (b004d902..., which dropped AsteroidVerticalOffset) with the branch's Rarity cleanup replayed: the five Rarity values (1, 0.5, 0.15, 0.05, 0.01) removed, nothing else touched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pins the survivors soul-ai-consumes-item-stream and soul-patrol-same-for-every-pilot of verdict s1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pins the survivors soul-derive-quadratic-agrees-on-fixtures and soul-derive-floored-mean of verdict s1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pins the survivors soul-cap-assembly-floored, soul-cap-tolerance-band and soul-product-override-mean-only of verdict s1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ScenarioStage.Approach(entity, target) states a piloted entity's task as a follow task on the target, passed through Zone.Admit to the flight as the member's baseline task; it replaces Patrol(entity, orbit) and the orbit index, and RunStart's post-admit circuit overwrite. LockOn stages at .6, the quality Crossing and the other RunStart scenarios use (its brief states none), not .8 picked by outcome.
The bearing check keeps well-law r4's intent (measure against the hostile's actual pose, not the staged line) but reads the poses of the tick before the paint was first seen, which is the tick it was stamped in, with a .05 deg limit for float error; the live-pose check at .5 deg spent its budget on the paint inbox's deliberate one-tick delay. LockOnHoldsUnderOnePercentOfEachStagedInput moves hostile x, y and quality by 1% each way.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The unseeded Random.Range prune leaves ZoneRenderer: the kept set is a function of the candidates, the kept count and the belt key.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Replaces RockMeshing/AssignRockMeshes and the renderer's slot bucketing (findings scale-lookup-unpinned,
renderer-index-wiring-unpinned). The build gate message names the mesh and the field, not the radii.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Seeds 7, 11 and 12 reproduce the ring-inside-Roche-margin defect on the base
generator; seeds 1-6 do not. Seed 10 shows a separate defect (a bodyless
non-ring node whose children are never placed), so the range stops at 9.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A failed Assert.Same or Assert.All over cyclic Galaxy objects overflowed xunit's formatter and crashed the host.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Roche-margin check on a rosette's ring read heavy.Mass before any member had
settled; each member's own Settle later folds its overflowing children into it,
so a ring could stand with a member inside its Roche margin. Settle now places
and settles the ring's members first and judges the ring on ring.Max(Mass); the
queue skips members already settled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ScenarioStage.Patrol(entity, near) states a piloted entity's patrol order (the orbit nearest a point) and RunStart applies it over the drawn circuit. LockOn stages its fit at .6 and the hostile on the orbit by the player. Tests: every SimulationDice stream draws independently of every other (they already did: the stream is mixed into the hash), the pilot's sample offset is on its own stream's grid, a patrol circuit does not move with items minted first, Crossing.ProjectorRange is read against the live range.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Combat.SampleDps and Zone.CreateAgent's patrol circuit drew from the item stream, so minting items moved
how a pilot fights and where it patrols. Both now draw SimulationDice.For(CombatSeed, stream, Agent.Id);
Agent.Id is the zone-scoped ordinal Zone.NextAgentId gives a pilot at admission (ruling
ai-sample-offset-off-item-stream).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
