Make the suite honest: digest repin, Crossing/MiningCut3 causes, authored cargo for Arcs and Pursuit, ConstantWeapon HUD, station descriptions #10
2026-11-11
TwoAiDigestIsPinned: repin linux-x64 to 52fbb10d and name the flight-model commits that moved it
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Catalog: mining names Diamond, volatiles names Ammonia (economy author ExtractionKindDescriptions)
Applied through AetherDb: 0 added, 2 changed (station-kind:mining, station-kind:volatiles), 0 removed. economy report fault count unchanged at 126. GameData/Aetheria.cc sha256 956a1573fa2aa69d7f72861e10dd17d7d3255d114e878da81802869efcfe31e9 (before: 0f81d35d4cec8f57178220f20bca98ffaa6709783349efe3cc28aa64ea1f02b1). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
AetherDb economy: Author takes the scratch root; TheTrialCopyIsRemoved audits its own, not the shared temp folder
The test counted aetherdb-economy-* directories in the machine's temp folder, which every parallel test running the command also writes to. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
GetAllChildClasses: the type memo is concurrent
A plain Dictionary written by parallel catalog loads corrupted itself ("Operations that change non-concurrent collections must have exclusive
access") and failed unrelated tests at random, through ValidateStatModifiers.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
MiningCut3Tests: ChunkInfo's pinned bits are per host class; the Linux bits are the base's own
ChunkInfoIsBitIdenticalToItsValueBeforeTheExtraction never passed on Linux: atc86f28e93(the commit that added it) it already measures 1039274088 for the second observer where the recording says 1039274086, and base12dfcef5, the code the recording claims to be taken from, measures the same ...088 on Linux (CultMath 0.4.1 and 0.5.0 alike). The sensor's bearing term (acos, curve) is libm arithmetic, which rounds differently on Windows, where the recording was made. It is a host-class recording read as universal, not a regression: no commit moved it. The pin is now per host class like the two-AI digest (ruling replays-per-host-class): windows-x64 keeps the original recording, linux-x64 is base12dfcef5run on Linux, and the current code equals it. A host with no recording fails and prints its bits. RunStartTests.HostClass is now internal, the one host-class name. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Crossing: the crosser is launched at the speed it should cross at, since a hull no longer decays below top speed
845f6c3aa (aetheria-gravity:ruling:no-drag-below-top-speed) removed hull drag below a hull's top speed. The crosser was staged at the Djinni's 50 m/s cap on the premise that drag slowed it to ~25 m/s at the bow line; it now holds 50 m/s and crosses at 2.57 s, 50 m/s relative, 1.1 rad/s, outside the tuned encounter (3.6 s, 25 m/s, 0.56 rad/s) that the Brief, the pass window and CrossingHoldsAcrossTheStartWindow all describe. CrossingConditions and CrossingHoldsAcrossTheStartWindow failed on that stale tuning, not on a defect.
The encounter is restaged to reproduce its tuning under the ruled physics: launched at 25 m/s from 90 m out (probed over speed x start; constant-speed passes are dx / v seconds, 45 m ahead). The pass lands at 3.63 s, 44.7 m, 24.9 m/s relative, 0.56 rad/s. The test's launch-speed assertions now pin the scenario's 25 m/s instead of "most of the cap"; the time, distance, relative-speed and angular-rate windows are unchanged. If drag below top speed returns, the crosser decays and the pass leaves the window.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Scenarios: the Arcs turret and the Pursuit player carry the crates their guns reload from
Arcs' hostile turret fits two 100-round Light Machine Guns and had no hold, so it fell silent after one magazine; it now fits a bay and carries 30 crates of Ammo (about 8 minutes of both guns firing without pause). Pursuit's authored cargo had no Mines, capping the player at one magazine of six; it now carries 10 crates (sixty-six mines). ScenarioStage.Cargo admits a commodity name as one unit, so authored cargo can state crates. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
HUD: a weapon says what its ammunition label shows; a beam reads its reserve alone
ConstantWeapon names an AmmoType but has no magazine (a crate per AmmoInterval straight from the cargo), so the label printed its unfilled
magazine ("1/N"). Weapon.AmmoReadout owns the label's text, HasMagazine says whether there is a magazine to show, and SchematicDisplay reads it.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
AetherDb: mining and volatiles descriptions name Diamond and Ammonia; the ExtractionKindDescriptions batch restates them
Every commodity a kind extracts (a recipe with no inputs) is named in that kind's description: checked over the shipped catalog, only these two kinds broke it. The batch applies in the catalog commit that follows; until then the catalog tests for it fail. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Test: the Economy console-never-written probes keep only the calling thread's writes
The two tests capture Console.Out to prove the commands write to the writer they are given. Console is process-global, so in a full-suite run other tests' diagnostic prints landed in the capture and failed the author dry-run test. ConsoleProbe keeps only writes made on the thread running the command, and serialises probes so two cannot restore each other's writer. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>