fix: correct the inverted EEVEE engine id and close the coverage gap that hid it

.github/workflows/validate.yml

@@ -136,6 +136,12 @@ jobs:
       - name: Check product exit codes are in README tables
         run: python3 tests/check_exit_code_readme.py
 
+      - name: Check EEVEE engine-id mappings
+        run: python3 tests/check_engine_id.py
+
+      - name: Check falsifiers declare the budget they target
+        run: python3 tests/check_falsifier_targets.py
+
       - name: Validate template Python syntax
         run: |
           echo "Checking template Python syntax..."

CONTRIBUTING.md

@@ -223,6 +223,39 @@ Three checks that appear on PRs are deliberately **excluded**:
 The two `Socket Security` checks come from a third-party GitHub App. An
 outage or an uninstall would deadlock merges, so they are advisory.
 
+## Falsifiers must fail the budget they target
+
+Every falsifier declares the budget it aims at, and it must fail **that**
+budget. A falsifier that exits non-zero on some earlier check proves
+nothing about its target: the run is red for the wrong reason, and the
+budget it was built for was never reached.
+
+Fix a collision by changing the model or the falsifier. **Never widen a
+band so an ill-aimed falsifier lands.**
+
+Worked example. `showcase/stone-archway`'s arch falsifier began as
+`--flat-arch`, laying the voussoirs as a flat lintel. A lintel is 0.62 m
+shorter than the arch, so it tripped the bounding-box budget (exit 8) and
+never reached the intrados circle fit (exit 19) it existed to break.
+Widening the bbox tolerance to let it through would have destroyed a real
+budget to rescue a bad falsifier. It was replaced by `--off-circle`, which
+keeps the angles, joints, materials, triangle count and envelope identical
+and wanders only the intrados radius. Four other falsifiers in the same run
+needed the same treatment: a stray vertex moved inside the silhouette, a
+`--short-skids` that floats one runner of three instead of all of them, a
+`--same-seed` split into design and placement RNG streams, and a
+`--sink-keystone` that no longer changes the Y envelope.
+
+`tests/check_falsifier_targets.py` enforces the declaration. Its default
+static mode reads each showcase piece's falsifier table and asserts the
+flags are real argparse flags, the declared exit codes appear in that
+piece's exit-code table, and every falsifier names a target budget. Its
+`--run BLENDER` mode executes each falsifier and asserts the **observed**
+exit equals the declared one — the mode that catches an ill-aimed
+falsifier. Runtime costs one Blender launch per falsifier, measured at
+276 s for the whole showcase tree on one version, so it is an authoring
+and cron tool rather than a per-PR smoke step.
+
 ## Exit codes
 
 Three roles, not one global table. Do not copy a code from one script into

examples/gn-sdf-remesh/gn_sdf_remesh.py

@@ -116,7 +116,7 @@ def main():
     # EEVEE-id inversion witnessed for real: the OTHER era's id must be
     # rejected by this build, the helper's accepted
     eid = get_eevee_engine_id()
-    wrong = 'BLENDER_EEVEE_NEXT' if bpy.app.version >= (5, 0, 0) else 'BLENDER_EEVEE'
+    wrong = 'BLENDER_EEVEE_NEXT' if bpy.app.version >= (5, 0, 0) else 'BLENDER_EEVEE'  # engine-id-exempt: the wrong-era id this example asserts is rejected
     try:
         bpy.context.scene.render.engine = wrong
         print(f"ERROR: wrong-era EEVEE id '{wrong}' was accepted", file=sys.stderr); return 5

examples/swatch-grid/swatch_grid.py

@@ -265,7 +265,7 @@ def main():
     # Cycles. Witness the inversion for real: the OTHER era's id must be rejected
     # by this build, and the helper's id must be accepted.
     eid = get_eevee_engine_id()
-    wrong = 'BLENDER_EEVEE_NEXT' if bpy.app.version >= (5, 0, 0) else 'BLENDER_EEVEE'
+    wrong = 'BLENDER_EEVEE_NEXT' if bpy.app.version >= (5, 0, 0) else 'BLENDER_EEVEE'  # engine-id-exempt: the wrong-era id this example asserts is rejected
     try:
         sc.render.engine = wrong
         print(f"ERROR: wrong-era EEVEE id '{wrong}' was accepted by this build — "

examples/turntable/turntable.py

@@ -135,7 +135,7 @@ def main():
     # the EEVEE-id mapping is asserted regardless of whether we render: the
     # OTHER era's id must be rejected by this build, the helper's accepted
     eid = get_eevee_engine_id()
-    wrong = 'BLENDER_EEVEE_NEXT' if bpy.app.version >= (5, 0, 0) else 'BLENDER_EEVEE'
+    wrong = 'BLENDER_EEVEE_NEXT' if bpy.app.version >= (5, 0, 0) else 'BLENDER_EEVEE'  # engine-id-exempt: the wrong-era id this example asserts is rejected
     try:
         bpy.context.scene.render.engine = wrong
         print(f"ERROR: wrong-era EEVEE id '{wrong}' was accepted", file=sys.stderr); return 5

showcase/README.md

@@ -45,6 +45,18 @@ entry in `showcase/gallery.json`, and a rendered still.
   A budget with no falsifier witnesses nothing:
   prove each one fails once, and check the exit code, not just
   non-zero.
+- **A falsifier must fail the budget it targets.** Declare the target in
+  the piece README's falsifier table — flag, target budget, exit code —
+  and make the run exit *that* code. A falsifier that trips an earlier
+  check is red for the wrong reason and proves nothing about its target.
+  Fix a collision by changing the model or the falsifier; never widen a
+  band so an ill-aimed falsifier lands. `stone-archway`'s `--flat-arch`
+  laid the voussoirs as a lintel, which is 0.62 m shorter and so failed
+  the bounding box (exit 8) instead of the intrados circle fit (exit 19);
+  `--off-circle` replaced it by keeping the whole envelope and wandering
+  only the radius. `tests/check_falsifier_targets.py` checks the
+  declaration statically, and with `--run BLENDER` executes each falsifier
+  and asserts the observed exit matches.
 - **Hygiene budgets.** Copied combinatorics from
   `examples/mesh-hygiene-audit` (do not import the example). Every piece
   asserts on the generated mesh: non-manifold edges 0, loose verts 0,
@@ -164,6 +176,46 @@ entry in `showcase/gallery.json`, and a rendered still.
   `measure_framing_deviation` and assert at the call site. Do not move
   or modify `gallery_framing.py` — import it by resolving the repo root
   (see the shipping-crate script).
+- **Contact sheet (required).** Composite the candidate hero beside the
+  pinned calibration set — canonical membership is in `CLAUDE.md`
+  § Quality Gates — commit it as
+  `docs/gallery/contact-sheets/<name>-contact-sheet.webp`, link it in the
+  PR body, and report a per-criterion verdict: stage darkness, wedge
+  warmth, subject fill, saturation, thumbnail legibility, plus mean
+  luminance against the calibration band. A claim without the committed
+  composite is not evidence.
+
+  Required because it has caught a real defect in showcase work. The
+  first sheets for `crate-stack` and `stone-archway` showed both wedge
+  pools reading as cool grey bands rather than the warm pool the house
+  style calls for; both were relit as a result. Nothing else in the
+  pipeline looks at the still beside its peers, so nothing else could
+  have seen it.
+
+- **Asset sheet (required).** Render the hero alone — neutral
+  three-quarter view, plain studio lighting, no staging tricks, no
+  labels, no comparison props — composite it beside the pinned
+  asset-quality reference set rendered the same way, commit under
+  `docs/gallery/asset-sheets/`, and report a verdict. The piece ships
+  only if it is not identifiable as the least-designed object in that
+  lineup.
+
+  Required because showcase is *entirely* game props, which is exactly
+  the scope `docs/VISUAL-STYLE.md` § Asset quality names, and because it
+  covers something no other gate here does. Budgets measure geometry
+  conformance; the contact sheet measures staged presentation. Neither
+  removes the scene, and a strong scene carries a weak model. The
+  recorded evidence is `socket-attach-points`: it passed every
+  measurable floor on its first draft — `edge90` 0.000, ten materials,
+  no default datablock names — and was then judged bad by eye and
+  rebuilt from scratch. The floors scored the bevels, not the design.
+
+  `examples/gallery_asset_quality.check_asset_quality` returns **11** on
+  violation, the same call pattern as `gallery_framing`. Showcase
+  numbering already spends 11 on the collider-triangle ceiling, so remap
+  the return at the call site rather than letting two budgets share a
+  code.
+
 - **Composition.** The README names which shipped skills and snippets the
   piece composes. Duplicated helpers stay inlined or copied; showcase
   scripts do not import snippets as a package.

showcase/iron-cauldron/iron_cauldron.py

@@ -115,7 +115,14 @@
 
 
 def eevee_engine_id():
-    return "BLENDER_EEVEE_NEXT" if bpy.app.version >= (4, 2, 0) else "BLENDER_EEVEE"
+    """EEVEE id: 'BLENDER_EEVEE' on 5.0+, 'BLENDER_EEVEE_NEXT' on 4.2-4.5.
+
+    Mapping matches examples/swatch-grid.get_eevee_engine_id, which asserts it
+    against the running build (copied, not imported). Inherited the inverted
+    form from showcase/shipping-crate; it returned 'BLENDER_EEVEE_NEXT' on 5.x,
+    where that id does not exist.
+    """
+    return "BLENDER_EEVEE" if bpy.app.version >= (5, 0, 0) else "BLENDER_EEVEE_NEXT"
 
 
 def fail(msg, code):

showcase/shipping-crate/shipping_crate.py

@@ -101,7 +101,14 @@
 
 
 def eevee_engine_id():
-    return "BLENDER_EEVEE_NEXT" if bpy.app.version >= (4, 2, 0) else "BLENDER_EEVEE"
+    """EEVEE id: 'BLENDER_EEVEE' on 5.0+, 'BLENDER_EEVEE_NEXT' on 4.2-4.5.
+
+    Mapping matches examples/swatch-grid.get_eevee_engine_id, which asserts it
+    against the running build (copied, not imported). The old form here keyed
+    on >= (4, 2, 0) and so returned 'BLENDER_EEVEE_NEXT' on 5.x, where that id
+    does not exist — the render path raised TypeError on 5.1 and 5.2.
+    """
+    return "BLENDER_EEVEE" if bpy.app.version >= (5, 0, 0) else "BLENDER_EEVEE_NEXT"
 
 
 def fail(msg, code):