feat: cool ambient in wreck interiors, plus the #124 optics guard

.github/workflows/deploy.yml

@@ -53,6 +53,22 @@ jobs:
           if-no-files-found: ignore
           retention-days: 7
 
+      # Issue #124: the interiors went flat twice, once per renderer surface,
+      # and nothing caught either. This is the guard, and unlike the Pixi one it
+      # enforces hand-set thresholds rather than recorded frames — see the header
+      # of the script for why recording them would defeat the purpose.
+      - name: Interior optics check
+        run: npm run interior:check
+
+      - name: Upload interior optics diff frames
+        if: failure()
+        uses: actions/upload-artifact@v7
+        with:
+          name: interior-optics-actual-frames
+          path: tests/fixtures/reference-frames/interior-optics/*.actual.png
+          if-no-files-found: ignore
+          retention-days: 7
+
       - name: Run migration client unit tests
         run: npm run test:unit
 

.github/workflows/pr.yml

@@ -58,6 +58,22 @@ jobs:
           if-no-files-found: ignore
           retention-days: 7
 
+      # Issue #124: the interiors went flat twice, once per renderer surface,
+      # and nothing caught either. This is the guard, and unlike the Pixi one it
+      # enforces hand-set thresholds rather than recorded frames — see the header
+      # of the script for why recording them would defeat the purpose.
+      - name: Interior optics check
+        run: npm run interior:check
+
+      - name: Upload interior optics diff frames
+        if: failure()
+        uses: actions/upload-artifact@v7
+        with:
+          name: interior-optics-actual-frames
+          path: tests/fixtures/reference-frames/interior-optics/*.actual.png
+          if-no-files-found: ignore
+          retention-days: 7
+
       - name: Run migration client unit tests
         run: npm run test:unit
 

CONTRIBUTING.md

@@ -20,6 +20,8 @@ your commit will be blocked if lint fails.
 | `npm run lint` | ESLint over `src/*.js` |
 | `npm test` | Runs the in-browser test suite headless via Playwright |
 | `npm run screenshots` | Captures review screenshots to `screenshots/` |
+| `npm run interior:check` | Fails if a wreck or cave interior has gone flat (issue #124) |
+| `npm run interior:update` | Re-records that guard's reference frames — never its thresholds |
 | `npm run license-check` | Fails if any dependency's license isn't in the allowlist below |
 
 ### License allowlist (issue #40)

README.md

@@ -210,6 +210,29 @@ New TypeScript UI copy is keyed in `src/app/i18n/catalog.ts`; direct user-facing
 | `npm run test:e2e` | Builds and tests the migration Worker plus the full legacy Playwright suite |
 | `npm run test:perf` | Captures the opt-in performance baseline |
 | `npm run screenshots` | Captures review screenshots (phone + desktop, setup + in-dive) to `screenshots/` via `scripts/screenshots.mjs` |
+| `npm run interior:check` | Issue #124 optics guard: measures the legacy client's overhead interiors and fails if one has gone flat (CI-gated) |
+| `npm run interior:update` | Re-records that guard's reference frames. Deliberately does **not** touch its thresholds — see the note below |
+
+### Interior optics thresholds (issue #124)
+
+`npm run interior:check` enforces per-scene floors on **per-pixel chroma**
+(`mean(max(R,G,B) - min(R,G,B))`) for the torch-lit wreck and cave interiors,
+plus two open-water control scenes. It exists because those interiors drifted to
+near-neutral grey twice — once per renderer surface — and nothing caught either.
+
+Two things about it are deliberate and easy to undo by accident:
+
+- **The thresholds live in `scripts/interior-optics-check.mjs`, not in a
+  recorded manifest.** `--update` re-records the reference *frames* only.
+  Thresholds derived from whatever is currently on screen describe the current
+  state, and if the current state is the shortfall, recording it locks the
+  shortfall in — which is the sequencing trap #124 explicitly warns about. If a
+  change to the interiors is intended, move the specific band and say why.
+- **The reference frames are review artefacts, not a pixel diff.** Playwright's
+  frames are deterministic per platform but not across them, and there is no
+  linux reference set here. The enforced half is the statistics — which do
+  travel: the same scenes measured on Windows and on `ubuntu-latest` agree
+  exactly on every interior, while their pixels do not agree at all.
 
 **CI pipelines** (GitHub Actions):
 

package.json

@@ -32,7 +32,9 @@
     "prepixi:visual-check": "npm run build",
     "pixi:visual-check": "node scripts/pixi-visual-check.mjs",
     "prepixi:visual-update": "npm run build",
-    "pixi:visual-update": "node scripts/pixi-visual-check.mjs --update"
+    "pixi:visual-update": "node scripts/pixi-visual-check.mjs --update",
+    "interior:check": "node scripts/interior-optics-check.mjs",
+    "interior:update": "node scripts/interior-optics-check.mjs --update"
   },
   "devDependencies": {
     "@playwright/test": "^1.62.0",