Official Codebase of 3DCodeVerse
Python
skills: the bundles state no gate thresholds, time or triangle budgets — their claims go with the text
gaoypeng6d12e82f6a88
diff --git a/harness/codeverse3d/skills/_claims/c3d-glsl-craft.toml b/harness/codeverse3d/skills/_claims/c3d-glsl-craft.tomldeleted file mode 100644index d55c785a..00000000--- a/harness/codeverse3d/skills/_claims/c3d-glsl-craft.toml+++ /dev/null@@ -1,42 +0,0 @@-# gl_frames thresholds quoted in the body, pinned to spatial/frame_stats.py.--[[claim]]-key = "static_diff"-text = "0.002"-python = "codeverse3d.spatial.frame_stats:STATIC_DIFF"-scale = 1-format = "{:.3f}"--[[claim]]-key = "flicker_diff"-text = "0.35"-python = "codeverse3d.spatial.frame_stats:FLICKER_DIFF"-scale = 1-format = "{:.2f}"--[[claim]]-key = "low_detail_edge"-text = "0.002"-python = "codeverse3d.spatial.frame_stats:LOW_DETAIL_EDGE"-scale = 1-format = "{:.3f}"--[[claim]]-key = "black_lum"-text = "0.03"-python = "codeverse3d.spatial.frame_stats:BLACK_LUM"-scale = 1-format = "{:.2f}"--[[claim]]-key = "blown_lum"-text = "0.98"-python = "codeverse3d.spatial.frame_stats:BLOWN_LUM"-scale = 1-format = "{:.2f}"--[[claim]]-key = "duplicate_diff"-text = "0.0001"-python = "codeverse3d.spatial.frame_stats:DUPLICATE_DIFF"-format = "{:g}"diff --git a/harness/codeverse3d/skills/_claims/c3d-opengl-pipeline.toml b/harness/codeverse3d/skills/_claims/c3d-opengl-pipeline.tomldeleted file mode 100644index d4b25ccf..00000000--- a/harness/codeverse3d/skills/_claims/c3d-opengl-pipeline.toml+++ /dev/null@@ -1,19 +0,0 @@-# gl_frames thresholds, pinned to the module the gate reads them from.--[[claim]]-key = "gl_static_diff"-text = "0.002"-python = "codeverse3d.spatial.frame_stats:STATIC_DIFF"-format = "{:.3f}"--[[claim]]-key = "gl_flicker_diff"-text = "0.35"-python = "codeverse3d.spatial.frame_stats:FLICKER_DIFF"-format = "{:.2f}"--[[claim]]-key = "gl_low_detail_edge"-text = "0.002"-python = "codeverse3d.spatial.frame_stats:LOW_DETAIL_EDGE"-format = "{:.3f}"diff --git a/harness/codeverse3d/skills/_claims/c3d-scene-composition.toml b/harness/codeverse3d/skills/_claims/c3d-scene-composition.tomlindex 28db0c60..968c1a01 100644--- a/harness/codeverse3d/skills/_claims/c3d-scene-composition.toml+++ b/harness/codeverse3d/skills/_claims/c3d-scene-composition.toml@@ -1,4 +1,4 @@-# scene_frames thresholds quoted in the body, pinned to spatial/frame_metrics.py.+# The camera-placement numbers quoted in the body, pinned to spatial/frame_metrics.py.[[claim]]key = "near_hit"@@ -7,20 +7,6 @@ python = "codeverse3d.spatial.frame_metrics:NEAR_HIT_M"scale = 1format = "{:.1f} m"-[[claim]]-key = "content_min_establishing"-text = "20%"-python = "codeverse3d.spatial.frame_metrics:CONTENT_MIN_ESTABLISHING"-scale = 100-format = "{:.0f}%"--[[claim]]-key = "content_min_authored"-text = "10%"-python = "codeverse3d.spatial.frame_metrics:CONTENT_MIN_AUTHORED"-scale = 100-format = "{:.0f}%"-[[claim]]key = "eye_min_above_ground"text = "0.3 m"@@ -34,17 +20,3 @@ text = "80 m"python = "codeverse3d.spatial.frame_metrics:EYE_MAX_ABOVE_GROUND_M"scale = 1format = "{:.0f} m"--[[claim]]-key = "flat_modal_warn"-text = "85%"-python = "codeverse3d.spatial.frame_metrics:FLAT_MODAL_FRAC"-scale = 100-format = "{:.0f}%"--[[claim]]-key = "flat_modal_error"-text = "92%"-python = "codeverse3d.spatial.frame_metrics:FLAT_MODAL_ERROR"-scale = 100-format = "{:.0f}%"diff --git a/harness/codeverse3d/skills/_claims/c3d-scene-lighting.toml b/harness/codeverse3d/skills/_claims/c3d-scene-lighting.tomldeleted file mode 100644index 916a750c..00000000--- a/harness/codeverse3d/skills/_claims/c3d-scene-lighting.toml+++ /dev/null@@ -1,45 +0,0 @@-# The four thresholds this bundle quotes are the ones scene_frames actually applies.-# Move one and the test that ships this skill fails, which is the point.--[[claim]]-key = "scene_dark_mean_lum"-text = "0.12"-python = "codeverse3d.spatial.frame_metrics:DARK_MEAN_LUM"-format = "{:.2f}"--[[claim]]-key = "scene_dark_frac"-text = "35 %"-python = "codeverse3d.spatial.frame_metrics:DARK_FRAC"-scale = 100-format = "{:.0f} %"--[[claim]]-key = "scene_blown_frac"-text = "20 %"-python = "codeverse3d.spatial.frame_metrics:BLOWN_FRAC"-scale = 100-format = "{:.0f} %"--[[claim]]-key = "scene_flat_modal_frac"-text = "85 %"-python = "codeverse3d.spatial.frame_metrics:FLAT_MODAL_FRAC"-scale = 100-format = "{:.0f} %"--[[claim]]-key = "scene_flat_modal_error"-text = "92 %"-python = "codeverse3d.spatial.frame_metrics:FLAT_MODAL_ERROR"-scale = 100-format = "{:.0f} %"--# The contrast floor that separates "dim but lit" (WARN) from "unlit or broken" (ERROR).-# The skill tells authors to aim at this instead of the 0.15 mean in a practical-lit-# night scene, so it must not drift from the gate.-[[claim]]-key = "scene_lowkey_min_std"-text = "0.12"-python = "codeverse3d.spatial.frame_metrics:LOWKEY_MIN_STD"-format = "{:.2f}"diff --git a/harness/codeverse3d/skills/_claims/c3d-scene-motion.toml b/harness/codeverse3d/skills/_claims/c3d-scene-motion.tomldeleted file mode 100644index 233b04ff..00000000--- a/harness/codeverse3d/skills/_claims/c3d-scene-motion.toml+++ /dev/null@@ -1,29 +0,0 @@-# Every motion threshold the body quotes, pinned to the constant it came from.-# If frame_motion.py moves a number, the test that ships this skill fails before-# an agent is ever told the old one.--[[claim]]-key = "motion_pixel_delta"-text = "8 of 255"-python = "codeverse3d.spatial.frame_motion:PIXEL_DELTA"-format = "{:.0f} of 255"--[[claim]]-key = "motion_moving_frac"-text = "0.4 %"-python = "codeverse3d.spatial.frame_motion:MOVING_FRAC"-scale = 100-format = "{:.1f} %"--[[claim]]-key = "motion_strong_delta"-text = "60 of 255"-python = "codeverse3d.spatial.frame_motion:STRONG_DELTA"-format = "{:.0f} of 255"--[[claim]]-key = "motion_strong_frac"-text = "0.08 %"-python = "codeverse3d.spatial.frame_motion:STRONG_FRAC"-scale = 100-format = "{:.2f} %"diff --git a/harness/codeverse3d/skills/c3d-blender-forms/SKILL.md b/harness/codeverse3d/skills/c3d-blender-forms/SKILL.mdindex 8ee41ab9..7760bdf5 100644--- a/harness/codeverse3d/skills/c3d-blender-forms/SKILL.md+++ b/harness/codeverse3d/skills/c3d-blender-forms/SKILL.md@@ -98,8 +98,8 @@ Two more from the same lint that cost a whole run when they fire: indexingyour own references to everything you create. (1 run lost time to this.)* **The exporter takes every *visible* mesh.** A leftover cutter, a construction guide or aduplicate ships as a part. Delete it or set `hide_render`/`hide_viewport`.-* **Detail budget over triangle count.** The prompt gives a target/floor/ceiling; a bevel- and a chamfer on the right edges read as more refinement than 200k triangles of subsurf.+* **Detail over triangle count.** A bevel and a chamfer on the right edges read as more+ refinement than a dense subsurf.## Before you finishdiff --git a/harness/codeverse3d/skills/c3d-cadquery-forms/SKILL.md b/harness/codeverse3d/skills/c3d-cadquery-forms/SKILL.mdindex 3ae6a049..92b41fd9 100644--- a/harness/codeverse3d/skills/c3d-cadquery-forms/SKILL.md+++ b/harness/codeverse3d/skills/c3d-cadquery-forms/SKILL.md@@ -69,7 +69,7 @@ solid — a revolved profile faked from stacked boxes reads as a blocky toy.penetration thresholds are stated once, in `c3d-part-contact`, and not repeated here.6. **Keep solids closed.** Export runs through STL as well as GLB; `.shell()` keeps a solidwatertight, hand-assembled faces do not.-7. Feature sizes >= 0.5 mm (OCC's tolerance eats smaller), <= 40 parts, build under 120 s.+7. Feature sizes >= 0.5 mm (OCC's tolerance eats smaller).## The four traps our own tooling is written to catchdiff --git a/harness/codeverse3d/skills/c3d-glsl-craft/SKILL.md b/harness/codeverse3d/skills/c3d-glsl-craft/SKILL.mdindex 751710e1..9a79112d 100644--- a/harness/codeverse3d/skills/c3d-glsl-craft/SKILL.md+++ b/harness/codeverse3d/skills/c3d-glsl-craft/SKILL.md@@ -5,7 +5,7 @@ license: Apache-2.0compatibility: GLSL 330 core via moderngl, headless. One image pass (src/shader.frag), an optional src/common.glsl and one feedback pass (src/buffer_a.frag). The harness declares the uniforms; the contract lists them.metadata:evidence: mixed- evidence_note: "Gate thresholds and rubric weights are read from live code (spatial/frame_stats.py, judges/rubrics/shader_v1.yaml). The corpus is thin: 9 graded glsl_shader runs, so every count below names its n and no rate is generalised."+ evidence_note: "Gate checks and rubric weights are read from live code (spatial/frame_stats.py, judges/rubrics/shader_v1.yaml). The corpus is thin: 9 graded glsl_shader runs, so every count below names its n and no rate is generalised."verified: "2026-08-25"corpus: "9 graded glsl_shader runs (18 judged rounds) + 5 opengl_python runs, mined 2026-08-25"target_metric: "mean_edge_density"@@ -26,13 +26,13 @@ corpus actually name.`spatial/frame_stats.py` renders five frames — t = 0, 1, 2.5, 4, 6 s — and measures them:-| what | constant | fires at |-|---|---|---|-| static | `STATIC_DIFF` | mean absolute pixel difference between consecutive frames **< 0.002** |-| flicker | `FLICKER_DIFF` | any consecutive pair **> 0.35** |-| duplicate frames | `DUPLICATE_DIFF` | a pair under 0.0001 (INFO) |-| near-flat image | `LOW_DETAIL_EDGE` | mean edge density **< 0.002** (fraction of pixels whose luminance gradient exceeds 0.02) |-| black / blown | `BLACK_LUM` 0.03, `BLOWN_LUM` 0.98 | > 97% of pixels in every frame; ERROR, and the gate hint asks for mean luminance 0.2-0.6 |+| what | fires when |+|---|---|+| static | consecutive frames barely differ |+| flicker | a consecutive pair jumps by a large share of the image |+| duplicate frames | two frames are identical (INFO) |+| near-flat image | almost no pixel sits on a visible luminance edge |+| black / blown | nearly every pixel is black (or white) in every frame; ERROR |Measured over the 9 graded glsl runs (mined 2026-08-25), **the gate fired zero findings**:mean frame-to-frame difference ranged 0.013 to 0.182 (median 0.082) and edge density 0.031diff --git a/harness/codeverse3d/skills/c3d-opengl-pipeline/SKILL.md b/harness/codeverse3d/skills/c3d-opengl-pipeline/SKILL.mdindex 2cb71324..0cd68e48 100644--- a/harness/codeverse3d/skills/c3d-opengl-pipeline/SKILL.md+++ b/harness/codeverse3d/skills/c3d-opengl-pipeline/SKILL.md@@ -45,7 +45,7 @@ The fix is to make the frame a pure function of `t`, and get the iterations insi* keep the state and the time it represents in `state`; each call advances from`state["t_sim"]` to `t`, and **resets to the seed whenever `t < state["t_sim"]`** so anout-of-order or repeated sample still gives the same image;-* cap the work per call (a few thousand steps) so the build does not time out;+* bound the work per call (a fixed maximum number of steps);* inject sources as a function of `t`, not once at `frame == 0`.## Trap 2: motion has to be continuous *and* visiblediff --git a/harness/codeverse3d/skills/c3d-opengl-pipeline/references/sampled-time.md b/harness/codeverse3d/skills/c3d-opengl-pipeline/references/sampled-time.mdindex 93a1b5b4..b48e3feb 100644--- a/harness/codeverse3d/skills/c3d-opengl-pipeline/references/sampled-time.md+++ b/harness/codeverse3d/skills/c3d-opengl-pipeline/references/sampled-time.md@@ -21,7 +21,7 @@ produce the same image.**```pythonDT = 1.0 / 120.0 # simulation step-MAX_STEPS = 4000 # keep the build inside its timeout+MAX_STEPS = 4000 # bounded work per calldef _reset(state):state["read"].write(state["seed_bytes"])@@ -61,24 +61,22 @@ our runs down for *"ink is only injected at t=0s"*.`codeverse3d/spatial/frame_stats.py`:-| metric | threshold | finding |+| metric | fires when | finding ||---|---|---|-| mean abs frame-to-frame delta | < 0.002 | static (WARN) |-| mean abs frame-to-frame delta | > 0.35 | flicker (WARN) |-| consecutive frames identical | delta < 1e-4 | duplicate (INFO) |-| mean edge density | < 0.002 | near-flat gradient (WARN) |-| mean luminance | < 0.03 all frames | black (ERROR) |-| mean luminance | > 0.98 all frames | blown (ERROR) |+| mean abs frame-to-frame delta | frames barely differ | static (WARN) |+| mean abs frame-to-frame delta | a pair jumps by a large share of the image | flicker (WARN) |+| consecutive frames identical | two frames are the same | duplicate (INFO) |+| mean edge density | almost no visible edges | near-flat gradient (WARN) |+| mean luminance | near-black in all frames | black (ERROR) |+| mean luminance | near-white in all frames | blown (ERROR) || any NaN / Inf pixel | any | ERROR |-The gate's own advice for luminance is to aim for a mean between 0.2 and 0.6.-## Measured deltas from our own runs`graphics_v1_flash` + `graphics_v2_flash`, 5 graded opengl_python runs, mined 2026-08-25.The gate fired on exactly one of them - `ogl_hard_voxel_city`, final score 0.342 - but itfired repeatedly: flicker in three separate rounds, with max frame-to-frame deltas of-0.373, 0.392 and 0.416, each barely over the 0.35 line, and each caused by a+0.373, 0.392 and 0.416, each a flicker finding, and each caused by adiscontinuity between two samples 1.5 to 2 seconds apart rather than by per-frame noise.The same run also produced `edge density 0.0000`, an entirely smooth image. So the gate isa weak signal at n=5; the judged criteria below are the stronger one.diff --git a/harness/codeverse3d/skills/c3d-scene-composition/SKILL.md b/harness/codeverse3d/skills/c3d-scene-composition/SKILL.mdindex 8ccad5d9..3daba175 100644--- a/harness/codeverse3d/skills/c3d-scene-composition/SKILL.md+++ b/harness/codeverse3d/skills/c3d-scene-composition/SKILL.md@@ -1,11 +1,11 @@---name: c3d-scene-composition-description: "Use when a scene_threejs session owns the environment, a zone's placement or the camera list, or when scene_frames reported a camera problem. Lay out the scene so the harness's deterministic instruments agree with the picture - one height function everything samples, a world that does not end in shot, and cameras computed from measured bounds instead of guessed. Includes the sky/ground/content classifier that silently decides two gate findings, and the exact scene_frames thresholds and score caps for camera placement and frame coverage."+description: "Use when a scene_threejs session owns the environment, a zone's placement or the camera list, or when scene_frames reported a camera problem. Lay out the scene so the harness's deterministic instruments agree with the picture - one height function everything samples, a world that does not end in shot, and cameras computed from measured bounds instead of guessed. Includes the sky/ground/content classifier that silently decides two gate findings, and the camera-placement rules and score caps scene_frames checks."license: Apache-2.0compatibility: track scene, language scene_threejs (three r182, headless Chrome, the harness builds the cameras from your plain camera objects).metadata:evidence: mixed- evidence_note: "Gate thresholds, score caps and the backdrop classifier are read from live code (spatial/frame_metrics.py, judges/rubrics/scene_v1.yaml, runtime_js/lib/backdrop.mjs, runtime_js/lib/scene_host.mjs). The corpus behind them is thin: only 3 graded scene runs exist, so every count below names its n."+ evidence_note: "Camera checks, score caps and the backdrop classifier are read from live code (spatial/frame_metrics.py, judges/rubrics/scene_v1.yaml, runtime_js/lib/backdrop.mjs, runtime_js/lib/scene_host.mjs). The corpus behind them is thin: only 3 graded scene runs exist, so every count below names its n."verified: "2026-08-25"corpus: "3 graded scene_threejs runs with gate artefacts, mined 2026-08-25"target_metric: "camera_placement_findings"@@ -56,7 +56,7 @@ Three consequences you must design around:`ground_y = 6.21 m` and three authored cameras placed at absolute eye height came back as`camera is 1.0 / 4.6 / 4.8 m below the ground level` — three ERRORs in one run.* **Instanced scatter always counts as content.** The cheapest way to lift a thin- establishing shot over the 20% content floor is more instanced planting and props, not a+ establishing shot is more instanced planting and props, not abigger ground plane (which counts against you).## Cameras are measured, and the numbers are hard@@ -69,11 +69,11 @@ turns those ERRORs into score ceilings:|---|---|---|---|| nearest surface < 0.5 m, or the camera is inside a mesh box | `NEAR_HIT_M` | `camera inside / touching geometry` | **0.50** || eye below ground and the frame corroborates it (dark, empty or one-colour) | — | `camera_underground` | **0.50** |-| establishing shot with content < 20% of the pixels | `CONTENT_MIN_ESTABLISHING` | `establishing shot shows too little content` | **0.60** |-| any authored shot with content < 10% | `CONTENT_MIN_AUTHORED` | `shot shows little content` | WARN |+| establishing shot that is mostly sky/ground | — | `establishing shot shows too little content` | **0.60** |+| any authored shot that is almost all sky/ground | — | `shot shows little content` | WARN || eye less than 0.3 m above ground | `EYE_MIN_ABOVE_GROUND_M` | `ant's-eye view` | WARN || eye more than 80 m above ground | `EYE_MAX_ABOVE_GROUND_M` | `satellite view, not a shot` | WARN |-| one luminance band holds > 85% of the frame | `FLAT_MODAL_FRAC` | `flat frame` | WARN, ERROR and **0.65** past 92% |+| one luminance band dominates the frame | — | `flat frame` | WARN; ERROR and **0.65** when it holds nearly all of it |So compute, never guess:diff --git a/harness/codeverse3d/skills/c3d-scene-composition/references/measured-cameras.md b/harness/codeverse3d/skills/c3d-scene-composition/references/measured-cameras.mdindex 731bbc2d..f1140fad 100644--- a/harness/codeverse3d/skills/c3d-scene-composition/references/measured-cameras.md+++ b/harness/codeverse3d/skills/c3d-scene-composition/references/measured-cameras.md@@ -79,5 +79,4 @@ scene_views -> camera_checks per camera: mean_lum, modal_frac, content_frac,nearest_hit_m, eye_height_m, ground_y```-Six numbers per camera. Read them against the thresholds; the gate that scores you reads-exactly the same six.+Six numbers per camera. Read them; the gate that scores you reads exactly the same six.diff --git a/harness/codeverse3d/skills/c3d-scene-lighting/SKILL.md b/harness/codeverse3d/skills/c3d-scene-lighting/SKILL.mdindex ab1f7ed8..6cc2101c 100644--- a/harness/codeverse3d/skills/c3d-scene-lighting/SKILL.md+++ b/harness/codeverse3d/skills/c3d-scene-lighting/SKILL.md@@ -1,11 +1,11 @@---name: c3d-scene-lighting-description: "Use when lighting a scene_threejs baseline, env or refine session, or when scene_frames called a frame too dark, blown out or flat. Light the scene so the frame gate passes and the mood still reads. Gives the exact thresholds the gate measures, the key / fill / practical recipe with numbers, and the one thing hex colours hide - how dark an albedo really is once the renderer works in linear light."+description: "Use when lighting a scene_threejs baseline, env or refine session, or when scene_frames called a frame too dark, blown out or flat. Light the scene so the frame gate passes and the mood still reads. Gives what the frame gate measures, the key / fill / practical recipe with numbers, and the one thing hex colours hide - how dark an albedo really is once the renderer works in linear light."license: Apache-2.0compatibility: three r0.182, headless WebGL; renderer fixed at ACES Filmic tone mapping, exposure 1.0, sRGB output.metadata:evidence: mixed- evidence_note: 'Thresholds and renderer settings are read from live code, namely spatial/frame_metrics.py and runtime_js/lib/browser/renderer.js, and the luminance figures are computed from those settings. The corpus half is thin and labelled as such, being 4 scene_threejs runs in scenes_v1_flash mined 2026-08-25, so it is quoted as examples and never as a rate.'+ evidence_note: 'The checks and renderer settings are read from live code, namely spatial/frame_metrics.py and runtime_js/lib/browser/renderer.js, and the luminance figures are computed from those settings. The corpus half is thin and labelled as such, being 4 scene_threejs runs in scenes_v1_flash mined 2026-08-25, so it is quoted as examples and never as a rate.'verified: "2026-08-25"target_metric: "dark_or_flat_frames"target_direction: "down"@@ -22,9 +22,9 @@ The harness renders every camera and scores the pixels (`spatial/frame_metrics.p| finding | fires when | severity ||---|---|---|-| too dark | mean luminance < **0.12** *or* > **35 %** of pixels near black | ERROR on an authored camera, WARN on a harness rig view |-| blown out | > **20 %** of pixels pure white | same |-| flat frame | one luminance band holds > **85 %** of pixels (> **92 %** makes it an ERROR on an authored camera) | WARN / ERROR |+| too dark | dim on average *or* mostly near-black pixels | ERROR on an authored camera, WARN on a harness rig view |+| blown out | a large share of pixels pure white | same |+| flat frame | one luminance band dominates the frame (an ERROR on an authored camera when it holds nearly all of it) | WARN / ERROR |Two consequences people get wrong:@@ -52,14 +52,14 @@ everything else equal:| foliage `0x1d3d22` (albedo 0.037) | frame 0.07 | frame **0.15** | frame 0.27 || plaster `0xf0f4f8` (albedo 0.900) | frame 0.75 | frame 0.87 | frame 0.94 |-A scene made of dark foliage sits on the gate line **when it is fully lit**, and doubling+A scene made of dark foliage stays dim **even when it is fully lit**, and doublingthe key only moves it to 0.27. Neither more light nor more exposure changes the ratio — onlya lighter albedo does. So: lift the albedos of the large surfaces first, then light. Groundalbedo below about 0.25 is a scene that cannot pass without heroic lighting.-The same arithmetic says where the headroom is: mean frame luminance 0.12 corresponds to a-linear scene luminance of only ~0.03, and the tone curve is steep down there — going from-linear 0.02 to 0.05 moves the frame from 0.08 to 0.20. Small increases pay near the floor.+The same arithmetic says where the headroom is: the tone curve is steep at the dark end —+going from linear 0.02 to 0.05 moves the frame from 0.08 to 0.20. Small increases pay near+the floor.## The recipe@@ -76,7 +76,7 @@ linear 0.02 to 0.05 moves the frame from 0.08 to 0.20. Small increases pay nearit from its own practicals instead: emissive on the lamp, a PointLight at each one, alow but non-zero base colour on dark materials so they read as material and not as void,and shadows left dark. The gate now tells these two cases apart — a dim frame WITH- contrast (`lum_std` >= 0.12) is reported as "dim but lit" and is only a WARN.+ real contrast is reported as "dim but lit" and is only a WARN.* **The sun is white at noon and warm only near the horizon.** A warm sun at high elevationreads as an error, not as mood.* **The fill is the opposite hue from the key.** Hemisphere light is sky colour above,@@ -93,18 +93,17 @@ gate's fix hint: the starter's `sunRig({ mood, intensity, fill, sunColor, fillSk— pick the mood (`day | golden | night | overcast`; a set sun is the night rig), raise`intensity` / `fill` (the rig clamps requests below its floors UP, so under-lighting is notreachable through it), never a second DirectionalLight on top — plus `emissiveIntensity`-**2-6** with a PointLight **0.5-2** per practical, and keep mean luminance **>= 0.15** so-you are not sitting on the threshold. If the frame is blown instead: `intensity` at the+**2-6** with a PointLight **0.5-2** per practical. If the frame is blown instead: `intensity` at thefloor, `fill` <= 1.2, sky below 0.9 white, `emissiveIntensity` <= 4 on large surfaces.For a scene lit from its PRACTICALS the target is contrast, not average — but it is a-BAND, not a floor you can fall through. Aim at `lum_std` **>= 0.12** with mean luminance-**0.15-0.28** and under 10 % pure black on every authored camera. One dim moon/sky key at+BAND, not a floor you can fall through. Aim at real contrast with a dim-but-lit mean and+little pure black on every authored camera. One dim moon/sky key at**<= 0.2** for shape in the shadows, at most one low ambient **0.06-0.12**, and noHemisphereLight and no second DirectionalLight.Both ends of that band are measured failures, one week apart in the same wave:-* Over the top: chasing the sky-lit ">= 0.15 mean" in a lantern-lit courtyard produced a+* Over the top: chasing a sky-lit average in a lantern-lit courtyard produced aflat grey wash whose granite paving read as snow and whose working water shader wasinvisible.* Through the floor: "remove the AmbientLight and the HemisphereLight" applied on its owndiff --git a/harness/codeverse3d/skills/c3d-scene-lighting/references/exposure-math.md b/harness/codeverse3d/skills/c3d-scene-lighting/references/exposure-math.mdindex f71a3ccd..97a81f07 100644--- a/harness/codeverse3d/skills/c3d-scene-lighting/references/exposure-math.md+++ b/harness/codeverse3d/skills/c3d-scene-lighting/references/exposure-math.md@@ -27,9 +27,8 @@ averages into `mean_lum`.| `0xf0f4f8 plaster` | 0.900 | 0.75 | 0.87 | 0.94 |-The gate's dark threshold is 0.12. Dark foliage lands at 0.15 fully lit and 0.07 at half-key — which is why a night woodland scene fails the gate no matter how the lights are-tuned, and the only real fix is a lighter albedo (or emissive practicals that put bright+Dark foliage lands at 0.15 fully lit and 0.07 at half key — which is why a night+woodland scene reads black no matter how the lights are tuned, and the only real fix is a lighter albedo (or emissive practicals that put brightpixels in the frame).## Step 3 — where the headroom is@@ -43,7 +42,7 @@ pixels in the frame).The curve is steep at the bottom: getting from a frame value of 0.12 to 0.20 needs thelinear scene luminance to go from 0.029 to 0.051, i.e. roughly +75 % of light — but getting-from 0.30 to 0.50 needs it to more than double. Near the gate line, modest increases move+from 0.30 to 0.50 needs it to more than double. At the dark end, modest increases movethe number a lot; that is the cheapest place to fix a dark scene.## Step 4 — sanity numbers for the coloured-darkness rows@@ -67,6 +66,5 @@ region carrying the frame's mean.## Step 5 — the check`scene_views` writes `camera_checks` into `metrics.json` with `mean_lum`, `dark_frac`,-`blown_frac` and `modal_frac` per camera — the same four numbers the gate thresholds. Aim-for `mean_lum >= 0.15`, `dark_frac <= 0.25`, `blown_frac <= 0.10`. Anything on the-threshold will flip between rounds.+`blown_frac` and `modal_frac` per camera — the same four numbers the gate reads. Read them,+and fix whatever finding the gate names.diff --git a/harness/codeverse3d/skills/c3d-scene-motion/SKILL.md b/harness/codeverse3d/skills/c3d-scene-motion/SKILL.mdindex 304bd406..07eb86b4 100644--- a/harness/codeverse3d/skills/c3d-scene-motion/SKILL.md+++ b/harness/codeverse3d/skills/c3d-scene-motion/SKILL.md@@ -1,18 +1,18 @@---name: c3d-scene-motion-description: "Use when a scene_threejs session authors or refines animation, or when scene_frames reported no motion or the judge said nothing_moves. Make the scene visibly move between the two frames the harness photographs, and survive the four mechanisms that silently freeze it. Applies to baseline, zone, refine and compose sessions. Covers the measured motion threshold, the single update hook the harness actually calls, the amplitudes that read at 1024 x 576, and why clearing the motion gate is not the same as scoring animation_life - that is judged per item against the plan animation list."+description: "Use when a scene_threejs session authors or refines animation, or when scene_frames reported no motion or the judge said nothing_moves. Make the scene visibly move between the two frames the harness photographs, and survive the four mechanisms that silently freeze it. Applies to baseline, zone, refine and compose sessions. Covers how motion is measured, the single update hook the harness actually calls, the amplitudes that read at 1024 x 576, and why clearing the motion gate is not the same as scoring animation_life - that is judged per item against the plan animation list."license: Apache-2.0compatibility: "scene track, language scene_threejs. three r182 in headless Chrome; the harness photographs each camera at t = 0 s and t = 1.5 s and diffs the pair."metadata:evidence: mixed- evidence_note: "The mechanism, thresholds and hook are read from live code (spatial/frame_motion.py, spatial/frame_metrics.py, runtime_js/lib/scene_host.mjs) and are exact. The corpus behind them is thin: 3 graded scene runs / 8 judged rounds on scenes_v1_flash, 2026-08-23. Where a number is from that battery it says so."+ evidence_note: "The mechanism and hook are read from live code (spatial/frame_motion.py, spatial/frame_metrics.py, runtime_js/lib/scene_host.mjs) and are exact. The corpus behind them is thin: 3 graded scene runs / 8 judged rounds on scenes_v1_flash, 2026-08-23. Where a number is from that battery it says so."verified: "2026-08-25"pairs_with: "c3d-scene-composition, c3d-scene-lighting, c3d-threejs-shader-traps"target_metric: "min_authored_changed_frac"target_direction: "up"target_unit: "fraction of pixels changed on the WEAKEST authored camera, t=0 → t=1.5 s"target_measurable: "false"- target_baseline: "0.006 median / 0.017 mean on the weakest authored camera; n=3 (bench/out, 2026-08-25). The gate needs only ONE authored camera over the 0.4 pct bar, so a run whose establishing shot is frozen at 0.13 pct still passes - which is why this is a guard, not a target"+ target_baseline: "0.006 median / 0.017 mean on the weakest authored camera; n=3 (bench/out, 2026-08-25). The gate needs only ONE authored camera to move, so a run whose establishing shot is frozen still passes - which is why this is a guard, not a target"---# Motion the harness can measure@@ -30,9 +30,8 @@ in the corpus (8 judged rounds over 3 graded scene runs, `scenes_v1_flash`,`codeverse3d/spatial/frame_motion.py` diffs the first and last frame of everycamera and hands the judge the number:-* a **pixel** counts as changed when any channel differs by more than 8 of 255-* a camera counts as **MOVING** when at least **0.4 %** of its pixels changed, or- when 0.08 % of them changed by more than 60 of 255 (a lit window, a spark)+* the harness counts changed pixels per camera; a camera counts as **MOVING** when+ enough of its frame changes (a lit window or a spark counts)* only **authored** cameras — the ones you returned from `createScene` — count.The harness's own orbit and eye views are measured and shown but never decidethe verdict@@ -40,17 +39,16 @@ camera and hands the judge the number:"nothing moves: the largest change on an authored camera is N% of pixels …(measured, not perceived)", and the judge is told to treat it as fact-So the target is concrete: **one authored camera over 0.4 % of changed pixels-between t = 0 and t = 1.5 s.** On a 1024 x 576 frame that is about 2,360 pixels —-roughly a 50 x 50 patch. A few centimetres of sway on a scene tens of metres wide-does not reach it.+So the target is concrete: **at least one authored camera visibly changes between+t = 0 and t = 1.5 s.** A few centimetres of sway on a scene tens of metres wide+does not do it.## Clearing the bar is necessary, and it is not sufficientRe-measuring the stored frames of all 8 recorded rounds with today's code(2026-08-25) gives an uncomfortable result: **every one of them cleared the bar.**-The best authored camera changed 2.1 %–4.7 % of its pixels in every round — five-to twelve times the threshold — and `animation_life` still came out between 0.1+The best authored camera changed 2.1 %–4.7 % of its pixels in every round, and+`animation_life` still came out between 0.1and 0.8, averaging 0.431.The reason is in the judge's own words on three of those rounds: "the planned@@ -154,7 +152,7 @@ Never use `Math.random()`, `Date.now()` or `performance.now()` — the contractforbids all three, and a non-deterministic scene makes the two framesincomparable.-## Amplitudes that clear 0.4 %+## Amplitudes that readGive the scene **one unmistakable hero motion** — a turning wheel, a driftingboat, falling leaves, a rotating beam, a flock — and then give every remaining@@ -181,8 +179,8 @@ Call `scene_views` (its default is `times=[0, 1.5]`) and read the measuredmotion table it prints — it names each authored camera and its changed fraction.Then do the harder check the table cannot do for you: put the two sheets side byside and walk the plan's animation list item by item, pointing at each one in the-picture. An item you cannot point at is the one the judge will name. If your best-authored camera is under 0.4 %, or an item is invisible, double the amplitude or+picture. An item you cannot point at is the one the judge will name. If every+authored camera reads STATIC, or an item is invisible, double the amplitude ormove the mover into frame. Do not ship on the strength of having written thecode — and if nothing at all moved, run `scene_probe` first: a frozen scene ismore often a thrown `update` than a small amplitude.diff --git a/harness/codeverse3d/skills/c3d-scene-motion/references/motion_recipes.md b/harness/codeverse3d/skills/c3d-scene-motion/references/motion_recipes.mdindex eb5b599e..c6143ad1 100644--- a/harness/codeverse3d/skills/c3d-scene-motion/references/motion_recipes.md+++ b/harness/codeverse3d/skills/c3d-scene-motion/references/motion_recipes.md@@ -76,8 +76,8 @@ function fallingField(THREE, n, area, { fall = 0.9, cycle = 8, seed = 0 } = {})The modulo is what makes recycling stateless: no counter to drift, no reset tomiss, and `update(1.5)` gives the same frame however many times it is called.-Allocate `pos`, `base` and `ph` once at build time — the contract budgets-`update` at under 4 ms and forbids per-frame allocation and `traverse`.+Allocate `pos`, `base` and `ph` once at build time — the contract forbids+per-frame allocation and `traverse` in `update`.## Registering moversdiff --git a/harness/codeverse3d/skills/c3d-scene-night/SKILL.md b/harness/codeverse3d/skills/c3d-scene-night/SKILL.mdindex eacb6624..cb54e9f1 100644--- a/harness/codeverse3d/skills/c3d-scene-night/SKILL.md+++ b/harness/codeverse3d/skills/c3d-scene-night/SKILL.md@@ -36,12 +36,12 @@ halo at ~0.15 opacity, both `depthWrite: false`; the bulb mesh itself is`MeshBasicMaterial` (a lit bulb does not shade). One sprite reads as a sticker;the two-layer stack is what reads as bloom.-## Street lamps — fake the pool, budget the real lights+## Street lamps — fake the pool, few real lightsThe pool of light on the ground is an additive radial-gradient disc ON the-ground plus a faint gradient cone for the shaft. Real `PointLight`s (decay 2)-are a budget: 4-8 for foreground lamps, no more — WebGL forward lighting costs-per-fragment per-light, and thirty real lights is both slow and flat.+ground plus a faint gradient cone for the shaft. Keep real `PointLight`s (decay 2)+to the foreground lamps — WebGL forward lighting costs per-fragment per-light, and+thirty real lights is both slow and flat.## Light shafts are barely therediff --git a/harness/codeverse3d/skills/c3d-threejs-forms/SKILL.md b/harness/codeverse3d/skills/c3d-threejs-forms/SKILL.mdindex cdb88465..5bec63a3 100644--- a/harness/codeverse3d/skills/c3d-threejs-forms/SKILL.md+++ b/harness/codeverse3d/skills/c3d-threejs-forms/SKILL.md@@ -62,8 +62,8 @@ prism in the render sheet.| small sphere | 16 x 12 || extrude `curveSegments` on a curved outline | 16 |-A `SphereGeometry(r, 64, 64)` repeated a hundred times is 800 k triangles and-blows the ceiling; the budget belongs on the shapes that fill the silhouette.+A `SphereGeometry(r, 64, 64)` repeated a hundred times is 800 k triangles for+nothing; the detail belongs on the shapes that fill the silhouette.## There is no CSG