The Source gets a linked Metallic and a Transmission Weight of 1, the bake is forced to fail at the
wiring and replace steps and with meshes that have no material, and the cage extrusion and max ray
distance preferences must move the normal map. All commits build; no C# touched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds aetheria.bake_from_source: bakes the Source meshes onto the selected render meshes as
'<ship id>.albedo' (sRGB) and '<ship id>.normal' (Non-Color), packed, wired into every material
(a shared '<ship id>.hull' material for a mesh with none). One routine serves both maps and every
target. The Source's Metallic and Transmission Weight are neutralised for the bake (the glTF importer
sets Metallic 1.0, which bakes black); engine, samples, device, selection and Source are restored in
a finally, and a failure removes everything made and puts every changed link back. Three bake
preferences, a panel button, the tripo-hull.glb LFS fixture and smoke_bake.py (4 passes).
_refuse_loose_meshes now shares the select-offenders refusal with the bake.
All commits build; no C# touched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mechanical merge adaptation: contact-reveals landed on master after missile-stats' base, and its fixture built a LauncherData without the
munition parts ValidateMunitionParts now demands (63 failures at Upsert). Builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Their specs were withdrawn under map-only-what-runs once their deferred
findings moved to follow-ups.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The test compares type and message with SingleFileMessagePackBackingStore.pull_all on the same
file; the add-on's shape refusal (also a ValueError) no longer passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ShipAuthoringStore.ValidateLoaded is the one composition of hull Validate and ValidateProducts;
Read and the draft branch of ship-authoring validate both call it (ValidateProducts is private).
Tests: a draft with an unnamed hull is refused; blank product name and key collision with the
offending product not first (both makers in turn); a negative hull price refused.
Builds untested here; the verify job runs at the end.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Builds. The previous commit's test survived the hand mutant munition-thruster-joins-ship-thrusters because the ship was never stepped.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Test-only; builds. Ports Soul's feedback-1 s2 probes as behavioural tests that
observe where each rule is decided (the published AbsorbEvent, the handler's
view of hull and shield, Feedback.Cue, DrawAhead.Carried). The blast shield
Magnitude is read against an unshielded twin's totalShare so a scale mutant
fails. Covers findings shield-layer-fields-unpinned, blast-order-unpinned,
height-fixtures-at-zero and cue-saturation-edges-unpinned.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The previous commit's ContactTests did not build a Mid hull, so ten tests failed on the missing hull (full run at
806641ae: 1481 passed, 10 failed). The commit builds; those ten tests are red until this one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ruling ship-bounce-restitution-setting: j = -(1+e)(v_rel.n)/(1/m1+1/m2), applied only while closing. The old
exchange formula (PostCollisionVelocity) is deleted. ContactTests now pin momentum, frame shift, restitution,
no impulse while separating, slow closing, tangent of both bodies, inactive hulls, the contact point radius
and the three-body pile. Not yet built or run; verification follows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first version left the package's GLB in place, so validate judged records and model through ReadPackage and the test failed in the mutation baseline. Test-only; builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Findings of verdict cut-ships-addon-guards.s1: guard-not-import (the version now comes from the checkout the imported
cultcache_py lives in, one owner, no pyproject-beside-a-path), prefix-compare-survives (0.3.0.1, 0.3.0rc1, 0.3.00
refused), cap-boundary-unpinned and group-restore-unpinned (smoke_package texture_boundary_cases: at the cap, one over,
height-long, non-square rounding, file/packed/node-group images, failed export restores). A copy whose scale raises is
removed. All commits build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ship_cc.read lets cultcache-py refuse a zero-byte file (CultLib pins that); the test expects the refusal. New behavioural tests: ValidateJudgesProductsOnADraftAndInspectListsThem (through ShipAuthoringCommands.Run), AProductKeyAlreadyInTheCatalogExcludesThePackage, and price 1 in ASoldPackageHullNeedsAPrice. Tests only plus the add-on deletion; all commits build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With equal masses the first exchange stops the visitor and a second contact with the docked hull cannot show.
Builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A docked hull shares its carrier's velocity, so the pair alone never closes; a visitor does expose a second contact.
Builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The pair moved 0.1 each before the contact step and so closed inside the 14.14 bound; start them 15 apart.
Builds; the first commit's only red test was this one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Imagination imagination-aeth-refresh-visibility, session self-2026-10-11-ag2. The model page
amendments are stated in this section; folding them into the Visibility rework rows is owed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Moving the target after Fire never changed the cell: the lateral is drawn in the hull's frame about the aimed point. Builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Contacts.Step exchanges normal velocity between touching hull discs (PostCollisionVelocity moved verbatim
from ShieldManager) and publishes Zone.Contacted; ShieldManager presents it; HullCollider.OnCollisionEnter
and ShieldManager's ship-on-ship branch are deleted. Not yet built or tested (verification follows).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The untested-cultcache test now refuses 0.4.0, 0.3.1 and 0.3 as well as 9.9.9 (a major-only comparison survived).
smoke_package's hull map is 4096x1000, so the scaled size (2048x500, 512x125) is not what a clamp of both sides or of
the other axis would give. Builds: yes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Passes 7 and 8 run mirror_matrix on ships whose grids are 14 (even) and 15 (odd) wide; the expected twin cells are
grid_w - a - w computed in the matrix itself. Pass 6 (5 wide) now runs through the same matrix_pass function. This is
the commit that adds the cases for follow-up mirror-wide-grid-test. Builds: yes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
_export_glb points each Image Texture node of the exported objects' materials whose image is longer than the Max texture
preference (default 2048) at a scaled copy (longest side at the cap, aspect kept) and restores the originals and removes
the copies in its finally block, so the .blend's images are never modified. smoke_package.py reads the GLB's PNG
headers: a 4096x1024 map exports at 2048x512 and at 512x128 under a 512 cap, a 256x256 map exports as it is, the .blend
images keep their size and no copy is left in bpy.data.images. Builds: yes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
_libraries reads [project].version from the pyproject.toml beside the cultcache-py source before importing it and raises
ValueError naming the version found and the packages path when it differs from TESTED_CULTCACHE_PY (0.3.0) or the file is
missing. Builds: yes. test_refuses_a_file_with_no_record fails as before (cultcache-py unpack error on a 0-byte file).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Survivor of the top-speed-any-behavior-counts hand mutant: the cleared fixture hull held no behavior at all. Test-only; builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The committed font is not rebuilt yet, so test_fe0f_twins.py's first test is red against it; the font lands next. Builds nothing in Unity.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
OpacityAt is already saturated to 0..1 by the sim, so a 0.3 vent reads thinner than a full
cloud and nothing above full draws thinner. Builds (Unity compile run after).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Keycaps and the other body-font-based sequences now collapse: 25 failures become 13. The 13 FE0F+ZWJ sequences still show three pieces. Builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The test asked a second cache for the handle of an object loaded by the first (identity lookup, no handle); it now asks the reopened cache for its product. cultcache-py raises an Unpack error on a zero-byte file, so ship_cc.read names the empty file through its own shape refusal. Both commits build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every prefab and scene was re-serialized by Unity 6, so each file also carries the editor's own migrations (m_RootOrder, m_ConstrainProportionsScale, m_enableWordWrapping to m_TextWrappingMode, name quoting). Builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
hull_grid.mirror_position gives the twin Position as grid_w - a - w, so an even-width footprint mirrors to its true
reflection and the cell lookup's tie rule no longer decides a mirror; the twin's object stands in the cell that Position
puts it in, so Package derives the same Position back. test_hull_grid pins the reflection on odd and even grids and
rows of 4 or more cells; smoke_mounts runs footprint widths 1 to 4 at every cell of an even and an odd grid, the
boundary tie, the armour and arc copy, the muzzle's local X, the centre-line threshold, the grid edge and the parent
check, and keeps a mirrored pair of two-cell footprints through Package and the validator. Every commit builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
LockWarningTests builds launcher designs, which the new gate (ValidateMunitionParts) refuses without parts. The golden test
selects the role's Quality term among the stat's heat and durability terms, and the item-read test settles the ship at the
item's optimal temperature so only role quality moves the stat.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>