@@ -264,8 +264,8 @@ <h1>cart</h1>
 <p><strong>Composes</strong> skills <code>mesh-editing-and-bmesh</code>, <code>bake-high-to-low</code>, <code>depsgraph-and-evaluated-data</code>, <code>engine-export-presets</code>, and snippets <code>bake_normal_high_to_low.py</code>, <code>setup_bake_target_image.py</code>, <code>lod_chain.py</code> / <code>decimate_to_budget.py</code>, <code>convex_hull_collider.py</code>, <code>export_preset_unity.py</code> (helpers copied, not imported as a package).</p>
 <h2>Budgets</h2>
 <p>Declared as named constants; every gate <strong>recomputes</strong> from the mesh, materials, UVs, evaluated LOD, collider, or export file.</p>
-<p>| Axis | Declared | Measured (4.5.11 / 5.1.2 / 5.2.1) | | --- | --- | --- | | Base triangles | 2760–2960 | 2856 / 2856 / 2856 | | LOD1 ratio | 0.32–0.62 of base | 0.5000 / 0.5000 / 0.5000 | | LOD2 ratio | 0.10–0.35 of base | 0.2200 / 0.2200 / 0.2154 | | Materials | exactly 2 distinct, metal ≥ 24, wood ≥ 800 faces | 2 slots, floors met | | UVs | in <code>0..1</code>, AABB overlap ≤ 1e-5 | in range, overlap 0 | | Outer AABB | (1.539, 0.749, 0.640) m ± 0.01 | (1.5393, 0.7486, 0.6400) | | Grounded | bbox min Z within 1e-4 of 0 | 0.0000 / 0.0000 / 0.0000 | | Named supports | each tyre&#x27;s own zmin within 1e-4 of the floor | +Y 0.000000, −Y 0.000000 | | Hygiene | loose V/E, non-manifold, zero-area, doubles @1e-5, n-gons: all 0 | 0 / 0 / 0 on every axis | | Z-fighting | coplanar cross-shell face pairs = 0 | 0 / 0 / 0 | | Tyre seat | 0.0030–0.0055 m interference, per angular station | [0.00400, 0.00400] over 48 stations | | Wheel mirror | the two wheels match in X, Z and extent within 5e-5 m | 0.00000 mm | | Material-island gap | metal↔wood min distance ≤ 0.008 m | 0.00000 / 0.00000 / 0.00000 | | Collider tris | ≤ 360 | 122 | | Export | written, size &gt; 0 | 201048 / 201048 / 201032 bytes |</p>
-<p>DECIMATE COLLAPSE triangle counts are <strong>not</strong> identical across series — 5.2.1 is more aggressive on LOD2. The gate is a ratio band, not an exact count. Bake pixels are stochastic; the gate is <code>has_data</code> plus operator <code>FINISHED</code>, not byte-identity. Construction uses no RNG. Export byte counts differ by 8 B on 5.2.1 (glTF serializer), not a gated axis.</p>
+<p>| Axis | Declared | Measured (4.5.11 / 5.1.2 / 5.2.1) | | --- | --- | --- | | Base triangles | 2760–2960 | 2856 / 2856 / 2856 | | LOD1 ratio | 0.32–0.62 of base | 0.5000 / 0.5000 / 0.5000 | | LOD2 ratio | 0.10–0.35 of base | 0.2199 / 0.2199 / 0.2157 | | Materials | exactly 2 distinct, metal ≥ 24, wood ≥ 800 faces | 2 slots, floors met | | UVs | in <code>0..1</code>, AABB overlap ≤ 1e-5 | in range, overlap 0 | | Outer AABB | (1.539, 0.749, 0.640) m ± 0.01 | (1.5393, 0.7486, 0.6400) | | Grounded | bbox min Z within 1e-4 of 0 | 0.0000 / 0.0000 / 0.0000 | | Named supports | each tyre&#x27;s own zmin within 1e-4 of the floor | +Y 0.000000, −Y 0.000000 | | Hygiene | loose V/E, non-manifold, zero-area, doubles @1e-5, n-gons: all 0 | 0 / 0 / 0 on every axis | | Z-fighting | coplanar cross-shell face pairs = 0 | 0 / 0 / 0 | | Tyre seat | 0.0030–0.0055 m interference, per angular station | [0.00400, 0.00400] over 48 stations | | Wheel mirror | the two wheels match in X, Z and extent within 5e-5 m | 0.00000 mm | | Material-island gap | metal↔wood min distance ≤ 0.008 m | 0.00000 / 0.00000 / 0.00000 | | Collider tris | ≤ 360 | 154 | | Export | written, size &gt; 0, removed after measuring | 221960 bytes on 5.2.1 |</p>
+<p>DECIMATE COLLAPSE triangle counts are <strong>not</strong> identical across series — 5.2.1 is more aggressive on LOD2. The gate is a ratio band, not an exact count. Bake pixels are stochastic; the gate is <code>has_data</code> plus operator <code>FINISHED</code>, not byte-identity. Construction uses no RNG. Export byte counts differ by a few bytes across series (glTF serializer); the gate is &quot;written and non-empty&quot;, not a byte count, and the file is removed once measured.</p>
 <h2>Falsifiers</h2>
 <p>Each flag breaks one stage so a <strong>named</strong> budget fails and the piece exits *that* code. Magnitudes are sized so no earlier gate can steal the failure: the two wheel moves stay inside <code>BBOX_TOL</code>, so the AABB gate (exit 8) still passes.</p>
 <p>| Flag | Target budget | Exit | | --- | --- | --- | | <code>--skip-decimate</code> | LOD1 ratio band (ratio becomes 1.0) | 9 | | <code>--lift-z</code> | AABB grounded zmin (whole mesh up 0.05 m) | 16 | | <code>--flush-tyre</code> | Coplanar cross-shell pairs — sets the tyre&#x27;s inner radius equal to the felloe&#x27;s outer radius, which is the construction bug this piece was rebuilt to remove (0 → 48 pairs) | 15 | | <code>--sink-tyre</code> | Tyre seat band — buries the hoop 9 mm into the felloe so it reads as one body | 18 | | <code>--float-wheel</code> | Named supports — lifts one wheel 3 mm while the other still grounds the AABB | 16 | | <code>--skew-wheel</code> | Wheel mirror — pushes one wheel 6 mm out along the track | 19 |</p>
@@ -299,17 +299,21 @@ <h2>Source</h2>
 export.
 
 Budgets are declared below and recomputed from the generated result.
-They are not API-contract witnesses. ``--skip-decimate`` skips the LOD
-DECIMATE stage so the LOD-ratio budget fails. ``--lift-z`` raises the
-mesh so the grounded-zmin hygiene budget fails.
+They are not API-contract witnesses. Each falsifier flag breaks one
+stage so a named budget fails and the piece exits that budget&#x27;s code:
+``--skip-decimate`` (LOD ratio, 9), ``--lift-z`` (AABB grounded, 16),
+``--float-wheel`` (named supports, 16), ``--flush-tyre`` (coplanar
+cross-shell pairs, 15), ``--sink-tyre`` (tyre seat band, 18),
+``--skew-wheel`` (wheel mirror, 19). The two wheel moves stay inside
+``BBOX_TOL`` so the AABB gate cannot steal the failure.
 
 No RNG. Construction is closed-form. DECIMATE COLLAPSE triangle counts
 are not byte-identical across Blender versions — the LOD gate is a
 ratio band, not an exact count.
 
     blender --background --python cart.py --
     blender --background --python cart.py -- --skip-decimate
-    blender --background --python cart.py -- --lift-z
+    blender --background --python cart.py -- --flush-tyre
     blender --background --python cart.py -- --output cart.png
 &quot;&quot;&quot;</span>
 <span class="k">import</span> argparse
@@ -1295,6 +1299,15 @@ <h2>Source</h2>
         os.remove(export_path)
     export_unity(export_path, [low, collider])
     export_size = os.path.getsize(export_path) <span class="k">if</span> os.path.isfile(export_path) <span class="k">else</span> <span class="n">0</span>
+    <span class="c"># Blender sets TMPDIR from its own preference, which resolves to the</span>
+    <span class="c"># working directory on a stock portable build — so gettempdir() is the</span>
+    <span class="c"># repo root under CI and every run left a .glb behind. The budget only</span>
+    <span class="c"># needs the byte count, so drop the file once it is measured.</span>
+    <span class="k">if</span> os.path.isfile(export_path):
+        <span class="k">try</span>:
+            os.remove(export_path)
+        <span class="k">except</span> OSError:
+            <span class="k">pass</span>
 
     print(<span class="s">f&quot;</span><span class="s">blender=</span>{tuple(bpy.app.version)}<span class="s"> skip_decimate=</span>{skip_decimate}<span class="s">&quot;</span>)
     print(
 
@@ -881,7 +881,7 @@ <h2><a href="bake-normal-high-to-low/">bake-normal-high-to-low</a></h2>
         <div class="card-body">
           <h2><a href="gn-socket-rename/">gn-socket-rename</a></h2>
           <p class="teaches">Compare and Random Value socket identifiers collapsed onto reused names in 5.2; enabled-name lookup wires on 4.5, 5.1, and 5.2</p>
-          <p class="witnesses"><span class="tag">witnesses</span> A_INT and Min_001 present on 4.5.11 and 5.1.2, absent on 5.2.1. --legacy-ids exits 5 only on 5.2. Evaluated 16 verts and gauge_h 1.80 on eight column verts</p>
+          <p class="witnesses"><span class="tag">witnesses</span> A_INT and Min_001 present on 4.5.11 and 5.1.2, absent on 5.2.1. --legacy-ids exits 5 only on 5.2. Evaluated 16 verts and gauge_h 1.80 on eight column verts; the still draws the column graduations from gauge_h, so an unwired Random Value renders a blank column</p>
           <a class="card-link" href="gn-socket-rename/">View example <span aria-hidden="true">&rarr;</span></a>
         </div>
       </article>
@@ -1096,12 +1096,12 @@ <h2><a href="signpost/">signpost</a></h2>
       </article>
       <article class="card" data-tags="mesh export showcase">
         <a class="card-media" href="chopping-block/" aria-label="chopping-block showcase piece detail page">
-          <img src="assets/chopping-block-hero.webp" alt="chopping-block — A procedural log-round chopping block with an iron hoop and a felling axe buried in the sawn face, through UVs, bake, LOD, collider, and Unity glTF, asserting…" loading="lazy" decoding="async" />
+          <img src="assets/chopping-block-hero.webp" alt="chopping-block — A procedural log-round chopping block with a riveted iron hoop and a felling axe buried in the sawn face, through UVs, bake, LOD, collider, and Unity glTF…" loading="lazy" decoding="async" />
         </a>
         <div class="card-body">
           <h2><a href="chopping-block/">chopping-block</a></h2>
-          <p class="teaches">A procedural log-round chopping block with an iron hoop and a felling axe buried in the sawn face, through UVs, bake, LOD, collider, and Unity glTF, asserting recomputed budgets rather than an API contract.</p>
-          <p class="witnesses"><span class="tag">witnesses</span> Recomputed: 1472 tris, three materials with 252 bark / 252 grain / 288 metal faces, UVs in 0..1 with zero AABB overlap, outer AABB 0.526 x 0.526 x 1.012 m, LOD ratios in band, convex collider 434 tris, non-empty glTF. Hygiene: zero non-manifold edges, loose geometry, doubles, zero-area faces, n-gons and coplanar disjoint face pairs, grounded at zmin 0, log plumb to 1e-7. Joint fit: four shells, 18.0 mm haft clearance in the eye, 78.7 mm bit bury, hoop biting 3.7-4.0 mm into the log on every segment. --fat-haft exits 17 on eye clearance; --round-band exits 18 on the hoop bite band.</p>
+          <p class="teaches">A procedural log-round chopping block with a riveted iron hoop and a felling axe buried in the sawn face, through UVs, bake, LOD, collider, and Unity glTF, asserting recomputed budgets rather than an API contract.</p>
+          <p class="witnesses"><span class="tag">witnesses</span> Recomputed: 1800 tris, four materials with 180 bark / 216 grain / 404 metal / 176 haft faces, UVs in 0..1 with zero AABB overlap, outer AABB 0.526 x 0.526 x 1.006 m, LOD ratios in band, convex collider 538 tris, non-empty glTF. Hygiene: zero non-manifold edges, loose geometry, doubles, zero-area faces, n-gons and coplanar disjoint face pairs, grounded at zmin 0, log plumb to 1e-7, oval haft 1.35 wide-to-thick. Joint fit: six shells, 18.0 mm haft clearance in the eye, 78.7 mm bit bury, hoop biting 3.7-4.2 mm into the log on every segment, two lap rivets seated 1.50 and 1.54 mm. --fat-haft exits 17 on eye clearance; --round-band and --float-rivets exit 18 on their seats; --round-haft exits 19 on the haft section.</p>
           <a class="card-link" href="chopping-block/">View showcase piece <span aria-hidden="true">&rarr;</span></a>
         </div>
       </article>
 
@@ -264,8 +264,8 @@ <h1>stone-well</h1>
 <p><strong>Composes</strong> skills <code>mesh-editing-and-bmesh</code>, <code>bake-high-to-low</code>, <code>depsgraph-and-evaluated-data</code>, <code>engine-export-presets</code>, and snippets <code>bake_normal_high_to_low.py</code>, <code>setup_bake_target_image.py</code>, <code>lod_chain.py</code> / <code>decimate_to_budget.py</code>, <code>convex_hull_collider.py</code>, <code>export_preset_unity.py</code> (helpers copied, not imported as a package).</p>
 <h2>Budgets</h2>
 <p>Declared as named constants; every gate <strong>recomputes</strong> from the mesh, materials, UVs, evaluated LOD, collider, or export file.</p>
-<p>| Axis | Declared | Measured (4.5.11 / 5.1.2 / 5.2.1) | | --- | --- | --- | | Base triangles | 8280–9500 | 9380 / 9380 / 9380 | | LOD1 ratio | 0.32–0.62 of base | 0.5000 / 0.5000 / 0.5000 | | LOD2 ratio | 0.10–0.35 of base | 0.2198 / 0.2198 / 0.2198 | | Materials | exactly 3 distinct | 3 | | Material faces | stone ≥ 3000, wood ≥ 600, metal ≥ 100 | all above | | UVs | in <code>0..1</code>, AABB overlap ≤ 1e-5 | in range, overlap 0 | | Outer AABB | (1.312, 1.312, 1.761) m ± 0.01 | (1.3120, 1.3120, 1.7606) | | Grounded | bbox min Z within 1e-4 of 0 | 0.0000 / 0.0000 / 0.0000 | | Named supports | all 12 bottom-course stones on the floor within 1e-4 | 12/12, worst 0.000000 | | Hygiene | loose V/E, non-manifold, zero-area, doubles @1e-5, n-gons: all 0 | 0 / 0 / 0 on every axis | | Z-fighting | coplanar cross-shell face pairs = 0 | 0 / 0 / 0 | | Post tenon | each post seats 0.015–0.022 m into the measured coping top | 0.01800 / 0.01800 / 0.01800 | | Post placement | each post within 0.02 rad, in plan, of a roof hip corner recomputed from the mesh | 0.00000 | | Bucket clearance | hangs 0.085–0.115 m clear of the measured coping top | 0.10000 | | Material-island gap | stone↔wood and metal↔wood min distance ≤ 0.008 m | 0.00000 / 0.00000 / 0.00000 | | Collider tris | ≤ 380 | 346 | | Export | written, size &gt; 0 | 686672 / 686736 / 686724 bytes |</p>
-<p>DECIMATE COLLAPSE triangle counts are <strong>not</strong> guaranteed identical across series — the gate is a ratio band, not an exact count. This mesh happened to match on 4.5.11 / 5.1.2 / 5.2.1. Bake pixels are stochastic; the gate is <code>has_data</code> plus operator <code>FINISHED</code>, not byte-identity. Construction uses no RNG. glTF byte size differs by a few hundred bytes across series.</p>
+<p>| Axis | Declared | Measured (4.5.11 / 5.1.2 / 5.2.1) | | --- | --- | --- | | Base triangles | 8280–9500 | 9380 / 9380 / 9380 | | LOD1 ratio | 0.32–0.62 of base | 0.5000 / 0.5000 / 0.5000 | | LOD2 ratio | 0.10–0.35 of base | 0.2198 / 0.2198 / 0.2198 | | Materials | exactly 3 distinct | 3 | | Material faces | stone ≥ 3000, wood ≥ 600, metal ≥ 100 | all above | | UVs | in <code>0..1</code>, AABB overlap ≤ 1e-5 | in range, overlap 0 | | Outer AABB | (1.312, 1.312, 1.761) m ± 0.01 | (1.3120, 1.3120, 1.7606) | | Grounded | bbox min Z within 1e-4 of 0 | 0.0000 / 0.0000 / 0.0000 | | Named supports | all 12 bottom-course stones on the floor within 1e-4 | 12/12, worst 0.000000 | | Hygiene | loose V/E, non-manifold, zero-area, doubles @1e-5, n-gons: all 0 | 0 / 0 / 0 on every axis | | Z-fighting | coplanar cross-shell face pairs = 0 | 0 / 0 / 0 | | Post tenon | each post seats 0.015–0.022 m into the measured coping top | 0.01800 / 0.01800 / 0.01800 | | Post placement | each post within 0.02 rad, in plan, of a roof hip corner recomputed from the mesh | 0.00000 | | Bucket clearance | hangs 0.085–0.115 m clear of the measured coping top | 0.10000 | | Material-island gap | stone↔wood and metal↔wood min distance ≤ 0.008 m | 0.00000 / 0.00000 / 0.00000 | | Collider tris | ≤ 380 | 346 | | Export | written, size &gt; 0, removed after measuring | 690092 bytes on 5.2.1 |</p>
+<p>DECIMATE COLLAPSE triangle counts are <strong>not</strong> guaranteed identical across series — the gate is a ratio band, not an exact count. This mesh happened to match on 4.5.11 / 5.1.2 / 5.2.1. Bake pixels are stochastic; the gate is <code>has_data</code> plus operator <code>FINISHED</code>, not byte-identity. Construction uses no RNG. glTF byte size differs by a few hundred bytes across series; the gate is &quot;written and non-empty&quot;, not a byte count, and the file is removed once measured.</p>
 <h2>Falsifiers</h2>
 <p>Each flag breaks one stage so a <strong>named</strong> budget fails and the piece exits *that* code.</p>
 <p>| Flag | Target budget | Exit | | --- | --- | --- | | <code>--skip-decimate</code> | LOD1 ratio band (ratio becomes 1.0) | 9 | | <code>--lift-z</code> | AABB grounded zmin (whole mesh up 0.05 m) | 16 | | <code>--stand-posts</code> | Coplanar cross-shell pairs — puts the posts back on the axis *and* stands them on the coping, which is the pair of choices the shipped piece made (0 → 4 pairs) | 15 | | <code>--float-stone</code> | Named supports — lifts one bottom-course stone 4 mm while the other eleven still ground the AABB | 16 | | <code>--shallow-tenon</code> | Post tenon band — seats the posts only 4 mm, too shallow to be a joint but not flush, so no faces share a plane | 18 | | <code>--drop-bucket</code> | Bucket clearance — lowers the bucket 0.26 m back down the shaft | 18 | | <code>--turn-posts</code> | Post placement — rotates the post ring one curb facet off the roof&#x27;s hip corners | 19 |</p>
@@ -1427,6 +1427,15 @@ <h2>Source</h2>
         os.remove(export_path)
     export_unity(export_path, [low, collider])
     export_size = os.path.getsize(export_path) <span class="k">if</span> os.path.isfile(export_path) <span class="k">else</span> <span class="n">0</span>
+    <span class="c"># Blender sets TMPDIR from its own preference, which resolves to the</span>
+    <span class="c"># working directory on a stock portable build — so gettempdir() is the</span>
+    <span class="c"># repo root under CI and every run left a .glb behind. The budget only</span>
+    <span class="c"># needs the byte count, so drop the file once it is measured.</span>
+    <span class="k">if</span> os.path.isfile(export_path):
+        <span class="k">try</span>:
+            os.remove(export_path)
+        <span class="k">except</span> OSError:
+            <span class="k">pass</span>
 
     print(
         <span class="s">f&quot;</span><span class="s">blender=</span>{tuple(bpy.app.version)}<span class="s"> skip_decimate=</span>{skip_decimate}<span class="s">&quot;</span>