Skip to content

armgit / 67.js

Public Imported from GitHub
0

3D engine that AI agents drive

Release 0.17.0 (#743)

Nguyen Xuan An76510af2a991
diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
index 3b3a40ba..b9bc8a39 100644
--- a/.claude-plugin/marketplace.json
+++ b/.claude-plugin/marketplace.json
@@ -1,23 +1,36 @@
{
"name": "67js",
- "owner": { "name": "CodePawl", "url": "https://67js.codepawl.com" },
- "metadata": { "description": "67js, a 3D engine an AI agent drives through MCP." },
+ "owner": {
+ "name": "CodePawl",
+ "url": "https://67js.codepawl.com"
+ },
+ "metadata": {
+ "description": "67js, a 3D engine an AI agent drives through MCP."
+ },
"plugins": [
{
"name": "67js",
"source": "./skills",
"strict": false,
"description": "The 67js MCP server and the build-67js-scene and render-67js-media skills.",
- "version": "0.16.0",
+ "version": "0.17.0",
"category": "development",
"homepage": "https://67js.codepawl.com",
"repository": "https://github.com/codepawl/67.js",
"license": "MIT",
- "skills": ["./build-67js-scene", "./render-67js-media"],
+ "skills": [
+ "./build-67js-scene",
+ "./render-67js-media"
+ ],
"mcpServers": {
"67js": {
"command": "npx",
- "args": ["-y", "-p", "@codepawlhq/67js", "67js-mcp"]
+ "args": [
+ "-y",
+ "-p",
+ "@codepawlhq/67js",
+ "67js-mcp"
+ ]
}
}
}
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 2d4fb41e..70cb9ed3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,110 @@
# Changelog
+## 0.17.0
+
+- bake.gridFog: the sun the fog scatters is dimmed by the fog between it and the sun (#543)
+- bake.gridFog: true (with irradianceGrid) lets the irradiance grid scatter in the scene's volume fog, off by default, for sunlit closed rooms with fog (#543)
+- An irradiance grid bake runs on the kernel worker with its 90 s limit instead of the page's 20 s, so 8 grid bounces fit a hall like temple-hall (#543)
+- Volume fog: opt-in `scatterOctaves`, `multiScatter`, `multiExtinction`, `multiAnisotropy`, `keyTransmittance` and `surfaceTransmittance` approximate multiple scattering of the sun in the haze and dim the sun by the haze between it and a point or surface. Off by default; the scenes without them render the same (#543)
+- The publish workflow waits up to five minutes for the registry to serve a new version instead of two. The 0.16.0 publish succeeded, but the registry took longer than two minutes, so the check failed and skipped the release page.
+- `applyAnimation(time, clip, { checks: false })` and `renderer.updateParts(parts, { checks: false })` skip the per-frame pose checks for clips you trust, and frames on the same clip no longer reset and snapshot every node. Eight animated characters: update CPU 1.6 ms to 1.3 ms with checks, 0.9 ms without.
+- Skinned character frames cost less CPU: palette math reads joint matrices in place and a normal pose allocates nothing, so eight animated characters update about 0.1 ms faster (#572)
+- Animating an imported glTF character costs about a third of the CPU it did: `applyAnimation` prepares the pose once instead of twice, reuses its rollback snapshot, and skin validation checks each distinct set of joint weights once instead of every vertex. Eight RobotExpressive characters went from 4.5 ms to about 1.5 ms per frame.
+- Add a per-frame density and velocity `field` to volume fog (also `renderer.setFogField`), flow-advected noise (`flowStrength`, `flowPeriod`), a `profile` choice, and scattering from up to 8 point and spot lights in the fog (issue #573). A fog without a field or lights draws exactly as before.
+- Breaking: a renderer that draws `environment.volumeFog` needs `import '@codepawlhq/67js/addons/fog-shader'` first (recipes, the film and the MCP server load it on their own), which takes about 2.5 KB gzip out of the core (#577)
+- Motion blur and bloom no longer leave a ghost of the previous frame after a seek (#578)
+- A few mirrored skin vertices (a finger pinched between two joints) no longer fail the frame. They draw pinched, and `renderer.diagnostics` lists `skin-reflected-vertices` with the count and the joints and weights of the first one (#581)
+- Materials accept `alphaMode: 'add'` for additive glow, and the opacity of blend and add parts changes each frame through `updateParts` without a new batch. (#586)
+- Volume fog takes `detailOctaves`, `detailAmount`, `warp` and `contrast` for wispier mist, `profileSoftness` for where a layer top starts to fade, and `keyScattering` to scale only the key light shafts (#586)
+- Metal with detailed normal and roughness maps no longer sparkles from frame to frame at grazing angles; `environment.specularAntiAliasing` sets the strength (#588)
+- resolveCoplanar: resolveFlushParts no longer layers pairs that lie on one plane to within a micrometre (the depth buffer already settles them) and settles chains of equal-sized parts, so it stops adding flicker on dense scenes (#593)
+- A film with a very long lens (far / near over 10,000, such as near 0.08 and far 9000) now switches to reversed-Z float depth on its own: `Renderer67` loads `reversed-depth.js` as a separate chunk on the first such frame and every frame after it draws with it (`await renderer.reversedDepthReady` to have it from frame one; `reversedDepth: false` keeps the 24-bit buffer). On the metro film, 24-bit depth could not tell surfaces 2 micrometres apart at 2.6 m, and stairs, canopy panels, handrails and steel flickered between frames; those pairs went from thousands of flipping pixels to under 60 (#593)
+- `resolveCoplanar` is now a function, `resolveFlushParts` from `67js/addons/coplanar-overlap`, so the plane-keyed search stays out of the core bundle. Its depth layers also work under reversed depth: the slope part of the offset is 16 times larger there, because the constant part moves a surface by under a micrometre in a float depth buffer (#593)
+- motion.constraints: adds copy_transforms, limit_distance, limit_scale, stretch_to, locked_track, transformation and action, each one scene_motion call (#595)
+- Recipes gain `cone`, `icoSphere`, `circle` (filled or ring) and `line` geometries, and `plane` is accepted as an alias of `grid` (#596)
+- Fields can read other surfaces: `nearest` (distance, point or normal of the closest point on a surface) and `raycast` (hit, distance, point or normal along a direction, with a `miss` value), plus `attr: group` to read a vertex group back. In `instances` the surface is a geometry id, so a scatter can avoid a path mesh and follow terrain; in `set_position`, `displace`, `vertex_group` and `vertex_color` it is a geometry record, and `vertex_group` stores the result by name (#597)
+- Force fields: a `fields` list on the recipe `particles` and `cloth` sections accelerates particles, hair and cloth through one shared evaluation, with wind, turbulence, vortex, drag and magnet types, `strength`, `falloff`, `radius` and `seed`. The same seed gives the same frames (#598).
+- Ocean modifier: the `ocean` mesh edit and stack node turn a flat grid into a seeded sea (Phillips spectrum for the wind, `choppiness`, `size`, `seed`, `time`) and write the vertex group `foam` where the surface folds. The same seed and time give identical vertices. Part of #599.
+- Per-object motion blur: `effects.motionBlur.objects` (0 to 1, off by default) smears each part that moves during the shutter along its own screen path, while the still background stays sharp; it replaces camera motion blur on that frame and loads on demand (#600)
+- render jobs: passes now include `diffuse`, `glossy`, `shadow`, `ao` and `mist` next to albedo, normal, emission, depth and objectId; diffuse + glossy + emission add back to the lit image, and `exr` keeps values above 1 (#601)
+- Three named post steps join defocus (`effects.depthOfField`) in the look: `effects.colorBalance` (`lift`, `gamma`, `gain`, each three numbers), `effects.glare` (`streaks`, `count`, `angle`, `fogGlow`, `threshold`, `size`) and `effects.lensDistortion` (`distortion`, `dispersion`). Each is off until a recipe names it and draws nothing at its neutral values, so a recipe without them renders the same pixels as before. Their code is a lazy module that loads only for a recipe that names one; `installPostSteps(renderer)` from `src/post-steps.js` adds the stages (#602)
+- `scene_inspect` reports a `bake-stale` warning when motion has moved a light, the sun or a part off the pose a light bake was traced at; a scene without motion or without a bake pays nothing. `bake.rebake: true` bakes each pose a render or seek samples (the last 4 are kept, about 0.7 s each on a small room), so the baked light matches the frame and the warning does not apply. Material animation is not checked, see docs/light-bake.md (#603)
+- MCP scene_create and set-environment now accept the 16 point, 8 spot and 7 area lights the engine renders, read from one limits file, not 8 point and 4 area (#604)
+- Motion tracks animate lights, the sun, sky and fog, exposure and focus distance, so a sunset, a lamp flicker or a rack focus plays from a recipe (#605)
+- Nodes take visible (keyable in a motion track), castShadow, receiveShadow, holdout and shadowCatcher, so films hide parts without parking them and renders composite over footage (#606)
+- Scene cameras take projection orthographic with orthoScale, focalLength with sensorWidth (instead of fov), shift [x, y] and fStop for depth of field, in recipes and set-camera (#607)
+- environment.hdri and a .cube environment.look: shipped HDRIs (sky-clouds, studio-softbox) and looks (teal-orange, noir), or catalog files by ID, work from recipes and MCP (#608)
+- Material graphs: new outputs `opacity` (base texture alpha), `clearcoat` and `transmission` (per-pixel masks), and mixing nodes `layer` (mask-blended layers such as paint over rust), `hue` and `curve`. The rasterizer reads opacity and clearcoat, the CPU path tracer all three. Graphs without them bake the same (#609)
+- A recipe mesh takes `shapeKeys` (named offsets or move-only edits), animated by the existing morph tracks and written as morph targets in a GLB (#610)
+- Render jobs gain four optional output settings, all off by default so a plain job writes the same bytes: `transparent: true` (PNG with alpha, for `image` and `stills`), `format: "png16"` or `"exr"` (16-bit PNG and OpenEXR from a half-float frame, passes included: depth in metres as float32 `Z`, object IDs as float32 `id`), `region: { x, y, width, height }` (a crop of the whole frame, for `image`, `stills` and `video`) and `fps` up to 60. The encoders are small hand-written Node code with no new dependency. `docs/agent-render-jobs.md` has the details.
+- Area lights can cast soft shadows: `castShadow: true` on an `areaLights` entry takes one tile of the local shadow atlas, and the penumbra widens with the rectangle and with the gap between blocker and receiver. Cost is 16 blocker taps and 32 filter taps per lit pixel, only for a shadowed area light.
+- `environment.sunAngularRadius` (radians, 0.0005 to 0.2) softens the sun shadow in the raster through the cascades (at least one slice is used) and sets the sun disk of the path tracer, which was fixed at 0.0087. Without the field nothing changes.
+- IES profiles: `ies` on spot and area lights takes a bundled name (`wide-flood`, `downlight`, `narrow-spot`, `batwing`) or `{ angles, candela }`; `parseIesPhotometry` reads an LM-63 file into that shape. The raster and the CPU path tracer read the same 16 samples. The GPU path tracer and the light bake ignore profiles.
+- `vertex_group` and `vertex_color` mesh edits write named weights and vertex colors; `displace` and cloth `pinGroup` read a group, and colors export as `COLOR_0` (#613)
+- Particle collision and look (#614): an emitter system takes `colliders` (static root nodes, bounced off by triangle), `bounce` and `friction` (also for the floor), `spin`, and `color` and `alpha` over life. A fast particle cannot tunnel through a thin mesh, a slow bounce settles, and the same seed gives the same frames after any seek. Alpha multiplies the vertex colors, a fade to black, because the renderer has no per-vertex alpha.
+- Field math and curves: field expressions gain `rotate`, `align` and `transform` for rotation and matrix math. Instances take a `curve` source (`points`, `count`, `closed`, `fillet`) that places points at equal arc length, rounds corners and, with `alignToNormal`, turns each instance along the curve. The `bounding_box` and `convex_hull` modifiers replace a mesh with its box or hull (#615).
+- Render jobs take `settings.renderer.backend: "gpu"` to path trace on the GPU, film and stills jobs can path trace every frame with one kernel compile, and recipe bakes run on the GPU when the page can, with `scene_inspect` naming the backend (#622)
+- The light tree splits by the surface area orientation heuristic, draws 90 percent by the angle-aware measure, runs on the WebGPU tracer (megakernel and wavefront), and `lightSelection` defaults to `"auto"`: the tree for a scene with more than 8 local lights, uniform otherwise (#623)
+- Path tracer `lightSelection: "tree"`: a light tree after Conty and Kulla (2018) as Cycles builds it, on the CPU reference and the WebGL2 kernel. On the new 24 lamp `many-lights` scene it scores 0.0617 against Cycles at 64 samples (uniform 0.1471) and takes 38 ms per full size sample against 72 ms; it stays off by default because it scores worse than uniform on `brick-room` and `light-linking` (issue #623)
+- `67js/addons/path-tracer-progressive` draws the GPU path tracer one sample per pixel per step, starting at a quarter of the size, and `samplingPattern: "blue-noise-first"` and `lightSelection: "importance"` are new GPU options, both off by default (#623)
+- Real-time WebGPU path tracer (#624): `update({ parts })` with the same part count keeps the history and the radiance cache, follows a part that only moved or turned (its matrix changed) through a per-part motion table, and rejects history on a part whose geometry changed, so a moving object leaves no trail; `followPartMotion: false` turns the following off for comparison
+- `createRealtimeTracer(scene, { backend: 'webgpu' })` runs the real-time path tracer (#624) as WebGPU compute: the WebGPU megakernel with the radiance cache lookup and record built in, a cache update with atomic fixed point sums, and the temporal denoiser, a-trous filter and upsample as compute passes in one command buffer, with the picture copied to `canvas`. On the RTX 3060 the demo room at its own camera takes about 20 ms a frame at 1920 by 1080 (traced at 960 by 540) against 36 ms on WebGL2, and the kernel compiles in about 0.5 s against 11 s. WebGL2 stays the default and the fallback: without WebGPU the tracer's `backend` is `"gpu"` and `fallbackReason` says why. On the WebGPU backend `frame(camera, { wait: true }).done` and `readFrame()` are promises. The denoiser also rejects history where the material at the first hit differs (WebGPU backend only)
+- `67js/addons/path-tracer-realtime` (`createRealtimeTracer(scene, options)`, then `frame(camera)`) is a real-time view of the WebGL2 path tracer: one path per pixel and frame at half the output size, two bounces whose last hit reads a spatial hash radiance cache, and a temporal denoiser (reprojection, variance, a-trous passes) that upsamples to the output size and draws to a canvas; the scene is static, motion vectors come from the camera, and primary hits are traced rather than rasterized (#624)
+- The WebGPU path tracer (`backend: 'webgpu'`) now traces volume fog and subsurface scattering, compiled into the megakernel only for a scene that uses them, and `bakeLightmapsAsync(..., { backend: 'webgpu' })` bakes lightmap texels on WebGPU with the same report fields. On the `fog` and `subsurface` reference scenes it is within 0.0039 FLIP of WebGL2 (480 by 270, 32 samples, no denoiser), and the `lit-interior` bake gives the same light as WebGL2 in 0.54 s against 0.76 s. The wavefront kernel still refuses fog and subsurface with a clear error. No default changes (#625)
+- The path tracer has a third backend, `backend: 'webgpu'`: the WebGL2 kernel written as WGSL compute (`src/path-tracer-webgpu.js`, loaded only for this backend, so the core bundle does not grow). It compiles in about a second instead of 8 to 20, matches the WebGL2 backend within 0.002 FLIP on the 16 reference scenes, and traces the demo room faster. It throws a clear error without `navigator.gpu`, and names volume fog and subsurface scattering (not ported yet) so callers fall back to `gpu`. No default changes (#625)
+- Recipe light bakes run on the native ray query tracer when `packages/render-native` is built: `bake.backend` is `native`, with the page's GPU kernel as the fallback (`bake.backendNote` says why). `bakeLightmapsAsync` takes `backend: "native"` with a `texelTracer`.
+- `renderer.maxSeconds` is honoured by `renderer.backend: "native"`: a trace that passes it fails the job instead of falling back.
+- The native tracer asks the hardware about alpha cutouts itself: cutout triangles are their own geometry and the kernel confirms a candidate hit when its colour map alpha reaches the cutoff, so rays no longer step through transparent texels (8 steps at most before).
+- Shadow rays on the CPU, WebGL2 and WebGPU tracers no longer leak light through dense alpha cutouts. Their any-hit walk took the first hit traversal met as if it were the nearest, so a transparent texel far along the ray let it skip closer leaves. A solid any-hit still ends the walk; a transparent one now retraces for nearest hits. On 300 masked leaf cards the four backends agree within FLIP 0.0004, down from 0.043 between native and the rest.
+- The native tracer (renderer.backend "native", recipe bakes) now traces volume fog and subsurface scattering with the same walk as the WebGPU megakernel, compiled in only for a scene that uses them, and honours `noiseThreshold` on the GPU: after 16 samples, and then every 8, a converged pixel draws no more, and `metrics.pathTracer.averageSamples` reports the mean a pixel drew. `JS67_NATIVE_TRACER=off` keeps a job on the gpu or cpu backend whatever is installed; a native job without the package gives the cpu backend's bytes and report fields (#626)
+- Fixed: the CPU, WebGL2 and WebGPU path tracers gave up after 8 transparent cutout texels on one ray, so dense foliage showed holes and lost shadows; the limit is now 64, and the foliage reference scene matches the native tracer to FLIP 0.001
+- renderer.backend "native" on render jobs (image, stills, video): hardware ray queries through the optional @codepawlhq/67js-render-native package, then the usual wavelet or OIDN denoise and PNG. Without the package, its binary or a Vulkan ray query adapter the job runs on gpu (cpu on a cpu runtime) and metrics.pathTracer.backendNote says why (#626)
+- `@codepawlhq/67js-render-native` is an optional Rust addon that traces a scene on the GPU with Vulkan ray queries, with a bench against the browser WGSL kernel (#626)
+- The path tracer draws the textures and cutouts of loaded glTF files, which it used to trace untextured (#635)
+- The agent bench has a committed baseline for main: 28 tasks, 3 runs each, on Haiku (pass@1 99%) and Sonnet (pass@1 95%).
+- bench: `bench/agent-modeling/bench.mjs` runs a 28-task agent bench (modeling, revision, scene, look, light, camera, animation and film tasks, each with an automatic grader) k times per model through headless `claude -p`, and prints pass@1, pass^k, calls, tokens and time per model; `compare-bench.mjs` diffs a branch against a baseline file (#647)
+- MCP tool schemas shrink from 196,826 to about 66,000 bytes (#648). The tools list common fields and one line naming the rest, the strict schema still checks every call, and scene_capabilities answers with a short index and serves each recipe part by topic (motion, environment, materialFields, meshEdit, meshEdit.<op> and more). CI now has absolute ceilings for the schema total and each tool.
+- The `build-67js-scene` and `render-67js-media` skills are rewritten for agents that use 67js through MCP: short entry files, references for look, light, camera, film and animation, render jobs and error recovery, and a working tool call for lens and depth of field, light caps, animated lights, sun and sky, catalog HDRIs and looks, `bake.gridFog`, fog multiple scattering, depth fixes and render flags. Evals under `skills/*/evals` run through `claude -p` with `scripts/run-skill-evals.mjs` (issue #649)
+- A Claude Code plugin: `/plugin marketplace add codepawl/67.js` then `/plugin install 67js@67js` adds the 67js MCP server (started with `npx` from the npm package) and the build-67js-scene and render-67js-media skills (#650)
+- `npm run plugin:sync` copies the package.json version into the plugin manifests, and a test fails when they differ (#650)
+- scene_apply and mesh_edit answer with one entry per operation (index, op, target, what changed, a warning when nothing changed) and a diff of node, material and geometry IDs added, removed and changed; entries past 20 are counted, not listed (#651)
+- scene_review, scene_look and the render result warn, with a field and a value to try, on an empty frame (`empty-frame`), a subject out of frame or too small (`subject-out-of-frame`, `subject-too-small`) and a black or clipped frame (`exposure-black`, `exposure-clipped`). The `bake-stale` warning of #603 now also arrives in a look, a review shot and a render job's metrics, not only in scene_inspect. They are in the frame's `diagnostics`, and a render job keeps one entry per warning (#652)
+- Mesh edit validation errors (cut, deform, generator, modifier, normal and attribute edits, and the parameter checks of the edit log) now read `edits[2].divisions: got 9; expected a whole number from 2 to 8, such as 2.`: the field path, the value received and a range, a list of choices or an example. Inside a recipe the path starts at the geometry (`geometries[0].edits[2].levels`). Size-limit errors for subdivide, array and subdivide_edges now name the field to lower (#653)
+- Errors for motion tracks and IK, clip layers, part relations, branches, path queries and crowds now name the field, the value and an example: `nodes[1].relations[0].side: got "+w"; expected one of +x, -x, +y, -y, +z, -z, the face of "a" it sits against.` The shared motion helpers (`bounded`, `denseArray`, `knownKeys`) keep their wording for now (#653)
+- The same readable errors now cover vegetation (`nodes[3].vegetation.scatter.density: got 9; expected a number from 0.0001 to 4, such as 1.`), `environment.wind`, particles and hair, cloth sheets and colliders, and image sequence maps: each names the full field path, the value received and an accepted example or range (#653)
+- The same readable errors now cover physics bodies, joints and ragdolls, vehicle drives (`motion.drive`, wheels, path, speed, lights), asset recipes and operations (root, overrides, animation, inspect, checkpoints), placed assets and the shrinkwrap edit: `motion.drive.wheels[1].node: got "Nope"; expected the name of a node in the asset, such as one scene_inspect lists.` The mesh-edit parameter checks that still say `Mesh edit 2 (...)` are not converted yet (#653)
+- Validation errors in the scene core now read `nodes[1].decal: got "yes"; expected true or false.`: scene operations (remove, set, inspect, checkpoints, revision), camera lens fields, geometry fields (points, holes, base, cutter), armatures, cloth and particle checks (`Cloth gravity: got [0,1]; expected three numbers, such as [0, 1, 0].`), terrain (`terrain.size: got 0; expected a number from 1 to 2000, such as 40.`) and mesh regions. Size limits and engine invariants keep their older wording (#653)
+- Validation errors for materials, nodes, the camera, lights, fog, bake, mesh edits and motion now read `materials[2].emissiveColor: got [-1,0,0]; expected three numbers of 0 or more, such as [1, 0.5, 0.2].`: the full field path, the value received (cut at 40 characters) and a range, a list of choices or an example. Schema errors from the MCP tools add the received value too (`recipe.materials.0.color.2: Too big: expected number to be <=1 (got 2)`). Mesh edit errors keep the position of the edit (`edits[1].segments`). Not every message in the engine is converted yet; the rest keep their older wording (#653)
+- llms.txt is now a task map; its tool list is generated from the MCP registry (`npm run llms`) and a test fails when it is stale (#654)
+- mcp: every tool has a title and annotations (only scene_dispose, scene_render_release and scene_release_checkpoint are destructive), six tools declare a small outputSchema with structuredContent, and scene_render_result returns the PNG as image content plus a resource link to each file (#655)
+- set-environment accepts the optional volumeFog fields (scatterOctaves, multiScatter, multiExtinction, ...) after creation, and the MCP schema names scatterOctaves and multiScatter (#661).
+- set-environment accepts the opt-in volume fog fields (scatterOctaves, multiScatter, multiExtinction, multiAnisotropy, keyTransmittance, surfaceTransmittance and the rest) on a scene that did not set them at creation, and the MCP schema lists them (#661).
+- MCP: shipped HDRI and .cube look names (sky-clouds, teal-orange and the rest) load in the packaged server built by npm run build:lib; they failed with "The URL must be of scheme file" (#664)
+- Parts with a mirrored (negative-determinant) matrix, such as a flipped billboard, now render with the right side lit and culled instead of ghosted or pale; the renderer batches them apart and glTF no longer flips winding itself (#680, #681)
+- Add multi-bone `chain` IK (with `pole`, `poleAngle`, `stretch`, `iterations` and per-joint `limits`) and `spline` IK along Catmull-Rom curves to `addons/skeleton-pose` and recipe `motion.ik` (#689).
+- Motion tracks take `extrapolation` (`constant` or `linear`) and `modifiers`: `cycles` (repeat, offset, mirror, with counts), `stepped`, `noise` (seeded), `generator`, `limits` and `envelope`, sampled the same in sampleMotion, camera paths and render jobs (#690)
+- Add `maintain_volume`, `damped_track`, `pivot`, `follow_path`, `clamp_to`, `shrinkwrap` and `armature` motion constraints, plus `floor` `useRotation` and `copy_transforms`/`transformation` `mix` modes (#691).
+- Add `cylinder`, `cone` and `compound` collider shapes, `spring` joints, joint `motor` controls, timed `events` (`impulse`, `kinematicPose`, `jointMotor`) and `fields` (force fields on dynamic rigid bodies) to physics definitions and recipe `physics` sections (#692).
+- Instancing: `instances` takes a `place` record with Poisson-disk `spacing`, a `density` field, `variants` of geometry and material picked at random or by a `variant` field, per-point `rotation` and `size` fields, and `at: "vertices"` or `"faces"` on a surface source to place on a mesh's own vertices and face centres. All of it is seeded (#698).
+- Add `curve.length` spacing and `realize` (into one mesh per material) for recipe `instances`.
+- `environment.sky: { type: 'physical', sunElevation, sunRotation, turbidity, groundAlbedo, aerialPerspective }` drives the key light direction and colour, the sky map, the background gradient and the distance haze from one sun, so a time of day is one parameter. Recipes without `sky` render byte for byte as before. (#705)
+- Add `ozone`, `sunDisc`, and time-of-day `environment` motion tracks for `environment.sky` (`type: "physical"`).
+- Add `position`, `velocity`, `uv`, `transmission`, `volume`, `environment`, `cryptomatte` and `aov` render passes (`material.aov` and material graph `aov` outputs), plus `multilayer: true` OpenEXR render jobs (#706).
+- Unprojected ambient occlusion samples in view space rather than world space so cameras far from the origin do not quantize distant wall normals into vertical stripes (#713).
+- `scene_apply`, `scene_restore` and `mesh_edit` answers no longer carry the node inspection: they return the revision, the render status, the scene diagnostics when there are any, a result per operation and the diff. The results and diff of a 20-operation batch stay under 2 KB (the render status adds about 1 KB on top) instead of about 8 KB for the whole answer. Call `scene_inspect` to read the nodes (#718)
+- Keep mesh-leaf tree and blade grass vertex and material colors within `[0, 1]` while preserving their rendered `baseColor * vertexTint` product so `exportGlb` accepts mesh vegetation directly.
+- Report interior vertex valence (regular and poles), quad quality metrics and a normalized `topologyScore` in `lintMesh` and `mesh_inspect` (#731).
+- Fixed: automatic `objectReflections` skip with a reported reason when a scene has more than 256 other parts instead of failing `createAgentScene`, and `Renderer67.dispose()` deletes `liquidProgram` so transmissive scenes release every WebGL program (#733, #734).
+- The README opens on a Bistro street render under a Poly Haven sky, with two more Bistro views and the three showcase worlds below it. The Bistro scene is credited under CC BY 4.0.
+- Add a Blender 5.2 inventory script and a report that compares it with the ledger and agent surface, with a ranked list of missing features
+- The scene skill lists the film glitches to check before a film goes out: specular sparkle, matte rails, depth of field motes, lens ghosts, sun shaft sources and dot-matrix signs.
+- The goal and its scope now live in docs/goal.md instead of a tracking issue
+- The README banner is a full-width Bistro street render with the wordmark on the left, and the showcase table lists the airport, data center and port worlds again.
+- The README follows the same layout as the other CodePawl repos: a banner with the wordmark and three renders, flat-square badges, a showcase table, an install section that starts with a prompt to paste into an agent, and a License heading.
+- The changelog groups its entries under headings, and 0.16.0 is rewritten as a short summary
+- The scene skill points agents at the scene_capabilities topics for field detail, drops the notes for the fixed fog schema (#661) and packaged catalog (#664) gaps, and its HDRI, look and fog eval checks the shipped names and 3 scatter octaves again.
+- The tagline is now "3D engine that AI agents drive" on the site, the package and the README banner
+
## 0.16.0
### Highlights
diff --git a/bench/size/baseline.json b/bench/size/baseline.json
index 8230c4b1..7d5d4e8e 100644
--- a/bench/size/baseline.json
+++ b/bench/size/baseline.json
@@ -14,24 +14,29 @@
},
"bundles": {
"entry:core": {
- "min": 273150,
- "gzip": 80973,
- "brotli": 68728
+ "min": 277284,
+ "gzip": 83286,
+ "brotli": 70705
+ },
+ "entry:core+lazy": {
+ "min": 291209,
+ "gzip": 89221,
+ "brotli": 76033
},
"entry:model": {
"min": 40093,
- "gzip": 14025,
- "brotli": 11542
+ "gzip": 14018,
+ "brotli": 11556
},
"entry:nature": {
"min": 51894,
- "gzip": 18329,
- "brotli": 15518
+ "gzip": 18333,
+ "brotli": 15538
},
"entry:motion": {
- "min": 43815,
- "gzip": 15257,
- "brotli": 12916
+ "min": 44022,
+ "gzip": 15350,
+ "brotli": 13040
},
"entry:canvas2d": {
"min": 27190,
@@ -39,14 +44,14 @@
"brotli": 7331
},
"entry:gltf": {
- "min": 157322,
- "gzip": 51497,
- "brotli": 44283
+ "min": 157909,
+ "gzip": 52258,
+ "brotli": 45056
},
"entry:gltf+lazy": {
- "min": 212120,
- "gzip": 69090,
- "brotli": 60070
+ "min": 226804,
+ "gzip": 75858,
+ "brotli": 66223
},
"entry:modeling": {
"min": 21149,
@@ -64,9 +69,9 @@
"brotli": 7937
},
"entry:mesh-query": {
- "min": 45446,
- "gzip": 15618,
- "brotli": 13074
+ "min": 45588,
+ "gzip": 15651,
+ "brotli": 13072
},
"entry:scene-query": {
"min": 50989,
@@ -74,9 +79,9 @@
"brotli": 14875
},
"entry:mesh": {
- "min": 242003,
- "gzip": 81515,
- "brotli": 68855
+ "min": 262361,
+ "gzip": 89080,
+ "brotli": 75043
},
"entry:shrinkwrap": {
"min": 56422,
@@ -84,14 +89,14 @@
"brotli": 15130
},
"entry:surface-attachment": {
- "min": 40497,
- "gzip": 12646,
- "brotli": 10472
+ "min": 41964,
+ "gzip": 13319,
+ "brotli": 11079
},
"entry:physics": {
- "min": 54288,
- "gzip": 17571,
- "brotli": 14952
+ "min": 58926,
+ "gzip": 19149,
+ "brotli": 16374
},
"entry:spume": {
"min": 35636,
@@ -99,29 +104,29 @@
"brotli": 10651
},
"entry:retarget": {
- "min": 54237,
- "gzip": 18386,
- "brotli": 15759
+ "min": 56211,
+ "gzip": 19308,
+ "brotli": 16498
},
"entry:crowd-skinning": {
- "min": 61230,
- "gzip": 18819,
- "brotli": 16133
+ "min": 63669,
+ "gzip": 19708,
+ "brotli": 16936
},
"entry:skeleton-pose": {
- "min": 65851,
- "gzip": 21983,
- "brotli": 19049
+ "min": 75142,
+ "gzip": 25019,
+ "brotli": 21578
},
"entry:vehicle": {
- "min": 46430,
- "gzip": 15744,
- "brotli": 13391
+ "min": 47322,
+ "gzip": 16359,
+ "brotli": 13938
},
"entry:vegetation": {
- "min": 82180,
- "gzip": 29448,
- "brotli": 25177
+ "min": 82264,
+ "gzip": 29482,
+ "brotli": 25232
},
"entry:terrain": {
"min": 26444,
@@ -141,42 +146,67 @@
"entry:grass": {
"min": 24946,
"gzip": 9383,
- "brotli": 8461
+ "brotli": 8458
},
"entry:clutter": {
"min": 27562,
"gzip": 10340,
- "brotli": 9307
+ "brotli": 9310
},
"entry:light-bake": {
- "min": 78676,
- "gzip": 28746,
- "brotli": 25383
+ "min": 80327,
+ "gzip": 29391,
+ "brotli": 25988
},
"entry:light-bake-gpu": {
- "min": 147355,
- "gzip": 49092,
- "brotli": 42626
+ "min": 93804,
+ "gzip": 34276,
+ "brotli": 29949
+ },
+ "entry:light-bake-gpu+lazy": {
+ "min": 252194,
+ "gzip": 79298,
+ "brotli": 69318
},
"entry:ground-bounce": {
- "min": 35795,
- "gzip": 12778,
- "brotli": 11573
+ "min": 35973,
+ "gzip": 12803,
+ "brotli": 11559
},
"entry:floor-sky-visibility": {
- "min": 47397,
- "gzip": 17857,
- "brotli": 16007
+ "min": 47616,
+ "gzip": 17878,
+ "brotli": 16057
},
"entry:path-tracer": {
- "min": 45960,
- "gzip": 18651,
- "brotli": 16483
+ "min": 58313,
+ "gzip": 23104,
+ "brotli": 20444
},
"entry:path-tracer+lazy": {
- "min": 104642,
- "gzip": 35314,
- "brotli": 31264
+ "min": 212719,
+ "gzip": 66459,
+ "brotli": 58455
+ },
+ "entry:path-tracer-progressive": {
+ "min": 103784,
+ "gzip": 33722,
+ "brotli": 29761
+ },
+ "entry:path-tracer-progressive+lazy": {
+ "min": 187494,
+ "gzip": 56648,
+ "brotli": 49795
+ },
+ "entry:path-tracer-realtime": {
+ "min": 118620,
+ "gzip": 38209,
+ "brotli": 33905
+ },
+ "entry:path-tracer-realtime+lazy": {
+ "min": 231917,
+ "gzip": 70640,
+ "brotli": 62422
},
"entry:texture-set": {
"min": 1837,
@@ -184,24 +214,24 @@
"brotli": 762
},
"entry:agent-scene": {
- "min": 188657,
- "gzip": 65732,
- "brotli": 57744
+ "min": 213702,
+ "gzip": 75802,
+ "brotli": 66730
},
"entry:agent-scene+lazy": {
- "min": 951279,
- "gzip": 347587,
- "brotli": 308958
+ "min": 1124424,
+ "gzip": 410069,
+ "brotli": 364262
},
"entry:agent-asset": {
- "min": 298573,
- "gzip": 103804,
- "brotli": 91009
+ "min": 358200,
+ "gzip": 124691,
+ "brotli": 109585
},
"entry:agent-asset+lazy": {
- "min": 996593,
- "gzip": 362598,
- "brotli": 321392
+ "min": 1167627,
+ "gzip": 424311,
+ "brotli": 376048
},
"entry:hdri": {
"min": 18870,
@@ -209,19 +239,19 @@
"brotli": 5826
},
"entry:camera-effects": {
- "min": 24088,
- "gzip": 8403,
- "brotli": 6781
+ "min": 25166,
+ "gzip": 8819,
+ "brotli": 7163
},
"entry:water-reflection": {
- "min": 37369,
- "gzip": 13158,
- "brotli": 11898
+ "min": 37501,
+ "gzip": 13173,
+ "brotli": 11881
},
"entry:render-passes": {
- "min": 3764,
- "gzip": 1749,
- "brotli": 1557
+ "min": 29848,
+ "gzip": 10528,
+ "brotli": 8676
},
"entry:user-pass": {
"min": 19511,
@@ -229,34 +259,34 @@
"brotli": 7227
},
"entry:line-art": {
- "min": 48430,
- "gzip": 14238,
- "brotli": 12815
+ "min": 49461,
+ "gzip": 14526,
+ "brotli": 13071
},
"entry:mesh-blend": {
"min": 12261,
"gzip": 4553,
- "brotli": 4134
+ "brotli": 4140
},
"entry:coplanar-overlap": {
- "min": 6309,
- "gzip": 2952,
- "brotli": 2643
+ "min": 9060,
+ "gzip": 4032,
+ "brotli": 3613
},
"entry:reversed-depth": {
- "min": 29016,
- "gzip": 10788,
- "brotli": 8864
+ "min": 31617,
+ "gzip": 11783,
+ "brotli": 9744
},
"entry:material-layer-shaders": {
- "min": 11378,
- "gzip": 3358,
- "brotli": 3076
+ "min": 11423,
+ "gzip": 3383,
+ "brotli": 3098
},
"entry:view-transform-shader": {
- "min": 4898,
- "gzip": 2299,
- "brotli": 2040
+ "min": 5251,
+ "gzip": 2463,
+ "brotli": 2188
},
"entry:view-transform": {
"min": 3715,
@@ -266,7 +296,7 @@
"entry:liquid-shader": {
"min": 11206,
"gzip": 3979,
- "brotli": 3624
+ "brotli": 3619
},
"entry:world-material-shader": {
"min": 16167,
@@ -281,17 +311,27 @@
"entry:subsurface-shader": {
"min": 3871,
"gzip": 1617,
- "brotli": 1469
+ "brotli": 1466
+ },
+ "entry:fog-shader": {
+ "min": 17252,
+ "gzip": 4956,
+ "brotli": 4542
},
"entry:light-linking": {
"min": 2503,
"gzip": 1117,
- "brotli": 1010
+ "brotli": 1007
},
"entry:cascaded-shadows": {
- "min": 31775,
- "gzip": 10566,
- "brotli": 8629
+ "min": 33377,
+ "gzip": 11024,
+ "brotli": 9083
+ },
+ "entry:light-shapes": {
+ "min": 7782,
+ "gzip": 3269,
+ "brotli": 2967
},
"entry:compact-color": {
"min": 1430,
@@ -304,19 +344,29 @@
"brotli": 9174
},
"entry:replace-textured-parts": {
- "min": 49226,
- "gzip": 13758,
- "brotli": 12311
+ "min": 64294,
+ "gzip": 19327,
+ "brotli": 16536
+ },
+ "entry:replace-textured-parts+lazy": {
+ "min": 78219,
+ "gzip": 25261,
+ "brotli": 21860
},
"entry:texture-update": {
- "min": 51017,
- "gzip": 14364,
- "brotli": 12833
+ "min": 66090,
+ "gzip": 19925,
+ "brotli": 17081
+ },
+ "entry:texture-update+lazy": {
+ "min": 80015,
+ "gzip": 25859,
+ "brotli": 22406
},
"entry:cloth": {
- "min": 10064,
- "gzip": 4238,
- "brotli": 3822
+ "min": 10112,
+ "gzip": 4260,
+ "brotli": 3841
},
"entry:update-part-vertices": {
"min": 1254,
@@ -324,29 +374,34 @@
"brotli": 609
},
"scene:lit-box": {
- "min": 180280,
- "gzip": 51056,
- "brotli": 43893
+ "min": 182265,
+ "gzip": 52410,
+ "brotli": 45258
+ },
+ "scene:lit-box+lazy": {
+ "min": 196190,
+ "gzip": 58344,
+ "brotli": 50586
},
"scene:gltf-box": {
- "min": 267433,
- "gzip": 80675,
- "brotli": 69671
+ "min": 272095,
+ "gzip": 83182,
+ "brotli": 71596
},
"scene:gltf-box+lazy": {
- "min": 322226,
- "gzip": 98267,
- "brotli": 85450
+ "min": 340990,
+ "gzip": 106780,
+ "brotli": 92762
},
"scene:agent-scene": {
- "min": 326519,
- "gzip": 104476,
- "brotli": 91238
+ "min": 347506,
+ "gzip": 113555,
+ "brotli": 99203
},
"scene:agent-scene+lazy": {
- "min": 1055202,
- "gzip": 376567,
- "brotli": 333659
+ "min": 1229297,
+ "gzip": 440419,
+ "brotli": 390224
}
}
}
diff --git a/bench/size/budgets.json b/bench/size/budgets.json
index 6058195b..82dee000 100644
--- a/bench/size/budgets.json
+++ b/bench/size/budgets.json
@@ -1,77 +1,88 @@
{
- "archiveBytes": 934216,
+ "archiveBytes": 1236423,
"goals": {
"lit-box": 50000,
"gltf-box": 80000,
"agent-scene": 100000
},
"budgets": {
- "entry:core": 82600,
- "entry:model": 14400,
+ "entry:core": 85000,
+ "entry:core+lazy": 91100,
+ "entry:model": 14300,
"entry:nature": 18700,
- "entry:motion": 15600,
+ "entry:motion": 15700,
"entry:canvas2d": 9400,
- "entry:gltf": 52600,
- "entry:gltf+lazy": 70500,
+ "entry:gltf": 53400,
+ "entry:gltf+lazy": 77400,
"entry:modeling": 8700,
"entry:sculpt": 2500,
"entry:surface": 10100,
"entry:mesh-query": 16000,
"entry:scene-query": 18000,
- "entry:mesh": 83200,
+ "entry:mesh": 90900,
"entry:shrinkwrap": 18300,
- "entry:surface-attachment": 12900,
- "entry:physics": 18000,
+ "entry:surface-attachment": 13600,
+ "entry:physics": 19600,
"entry:spume": 12900,
- "entry:retarget": 18800,
- "entry:crowd-skinning": 19200,
- "entry:skeleton-pose": 22500,
- "entry:vehicle": 16100,
+ "entry:retarget": 19700,
+ "entry:crowd-skinning": 20200,
+ "entry:skeleton-pose": 25600,
+ "entry:vehicle": 16700,
"entry:vegetation": 30100,
"entry:terrain": 10300,
"entry:heightfield": 1500,
"entry:navmesh": 3400,
"entry:grass": 9600,
"entry:clutter": 10600,
- "entry:light-bake": 29400,
- "entry:light-bake-gpu": 50100,
+ "entry:light-bake": 30000,
+ "entry:light-bake-gpu": 35000,
+ "entry:light-bake-gpu+lazy": 80900,
"entry:ground-bounce": 13100,
"entry:floor-sky-visibility": 18300,
- "entry:path-tracer": 19100,
- "entry:path-tracer+lazy": 36100,
+ "entry:path-tracer": 23600,
+ "entry:path-tracer+lazy": 67800,
+ "entry:path-tracer-progressive": 34400,
+ "entry:path-tracer-progressive+lazy": 57800,
+ "entry:path-tracer-realtime": 39000,
+ "entry:path-tracer-realtime+lazy": 72100,
"entry:texture-set": 900,
- "entry:agent-scene": 67100,
- "entry:agent-scene+lazy": 354600,
- "entry:agent-asset": 105900,
- "entry:agent-asset+lazy": 369900,
+ "entry:agent-scene": 77400,
+ "entry:agent-scene+lazy": 418300,
+ "entry:agent-asset": 127200,
+ "entry:agent-asset+lazy": 432800,
"entry:hdri": 6700,
- "entry:camera-effects": 8600,
+ "entry:camera-effects": 9000,
"entry:water-reflection": 13500,
- "entry:render-passes": 1800,
+ "entry:render-passes": 10800,
"entry:user-pass": 8300,
- "entry:line-art": 14600,
+ "entry:line-art": 14900,
"entry:mesh-blend": 4700,
- "entry:coplanar-overlap": 3100,
- "entry:reversed-depth": 11100,
+ "entry:coplanar-overlap": 4200,
+ "entry:reversed-depth": 12100,
"entry:material-layer-shaders": 3500,
- "entry:view-transform-shader": 2400,
+ "entry:view-transform-shader": 2600,
"entry:view-transform": 2100,
"entry:liquid-shader": 4100,
"entry:world-material-shader": 5800,
"entry:splat-shader": 7000,
"entry:subsurface-shader": 1700,
+ "entry:fog-shader": 5100,
"entry:light-linking": 1200,
- "entry:cascaded-shadows": 10800,
+ "entry:cascaded-shadows": 11300,
+ "entry:light-shapes": 3400,
"entry:compact-color": 700,
"entry:gaussian-splats": 11400,
- "entry:replace-textured-parts": 14100,
- "entry:texture-update": 14700,
+ "entry:replace-textured-parts": 19800,
+ "entry:replace-textured-parts+lazy": 25800,
+ "entry:texture-update": 20400,
+ "entry:texture-update+lazy": 26400,
"entry:cloth": 4400,
"entry:update-part-vertices": 800,
- "scene:lit-box": 52100,
- "scene:gltf-box": 82300,
- "scene:gltf-box+lazy": 100300,
- "scene:agent-scene": 106600,
- "scene:agent-scene+lazy": 384100
+ "scene:lit-box": 53500,
+ "scene:lit-box+lazy": 59600,
+ "scene:gltf-box": 84900,
+ "scene:gltf-box+lazy": 109000,
+ "scene:agent-scene": 115900,
+ "scene:agent-scene+lazy": 449300
}
}
diff --git a/bench/tool-cost-budgets.json b/bench/tool-cost-budgets.json
index 95ecdb8d..49500a8a 100644
--- a/bench/tool-cost-budgets.json
+++ b/bench/tool-cost-budgets.json
@@ -1,17 +1,18 @@
{
- "schemaTotalBytes": 68590,
+ "schemaTotalBytes": 69410,
"schemaToolBytes": {
- "scene_apply": 20900,
- "scene_create": 12870,
- "scene_create_asset": 8900
+ "scene_apply": 21200,
+ "scene_create": 12990,
+ "scene_create_asset": 8950,
+ "scene_render_submit": 4270
},
"schemaOtherToolBytes": 4000,
"resultBytes": {
- "scene_capabilities": 1410,
+ "scene_capabilities": 1440,
"scene_create": 1200,
"scene_inspect": 1150,
- "mesh_inspect": 940,
- "mesh_edit": 2050,
+ "mesh_inspect": 1110,
+ "mesh_edit": 2220,
"scene_apply": 100,
"scene_export_recipe": 2720
}
diff --git a/changes/543-fog-sun.md b/changes/543-fog-sun.md
deleted file mode 100644
index 0fce794a..00000000
--- a/changes/543-fog-sun.md
+++ /dev/null
@@ -1 +0,0 @@
-- bake.gridFog: the sun the fog scatters is dimmed by the fog between it and the sun (#543)
diff --git a/changes/543-grid-fog.md b/changes/543-grid-fog.md
deleted file mode 100644
index 3296186a..00000000
--- a/changes/543-grid-fog.md
+++ /dev/null
@@ -1 +0,0 @@
-- bake.gridFog: true (with irradianceGrid) lets the irradiance grid scatter in the scene's volume fog, off by default, for sunlit closed rooms with fog (#543)
diff --git a/changes/543-grid-fog.size.md b/changes/543-grid-fog.size.md
deleted file mode 100644
index ec4166a6..00000000
--- a/changes/543-grid-fog.size.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- scene:agent-scene+lazy: the opt-in grid fog (a volumetric path in the fog, gridFog) brings temple-hall-a and -b closer to Cycles; the code loads only with a bake.
-- goal:agent-scene: gridFog adds a validated bake field (36 bytes gzip); the fog trace itself is in the lazy bake module.
diff --git a/changes/543-grid-worker.md b/changes/543-grid-worker.md
deleted file mode 100644
index 96af3476..00000000
--- a/changes/543-grid-worker.md
+++ /dev/null
@@ -1 +0,0 @@
-- An irradiance grid bake runs on the kernel worker with its 90 s limit instead of the page's 20 s, so 8 grid bounces fit a hall like temple-hall (#543)
diff --git a/changes/543-multiscatter.md b/changes/543-multiscatter.md
deleted file mode 100644
index 5f650a03..00000000
--- a/changes/543-multiscatter.md
+++ /dev/null
@@ -1 +0,0 @@
-- Volume fog: opt-in `scatterOctaves`, `multiScatter`, `multiExtinction`, `multiAnisotropy`, `keyTransmittance` and `surfaceTransmittance` approximate multiple scattering of the sun in the haze and dim the sun by the haze between it and a point or surface. Off by default; the scenes without them render the same (#543)
diff --git a/changes/543-multiscatter.size.md b/changes/543-multiscatter.size.md
deleted file mode 100644
index dbbc5ff7..00000000
--- a/changes/543-multiscatter.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- entry:fog-shader: the opt-in multiple scattering octaves and sun transmittance add one function set to the lazy fog shader (636 bytes gzip); a frame without volume fog never loads it.
-- scene:agent-scene+lazy: the fog shader's opt-in multiple scattering and sun transmittance, measured against Cycles in #543; off by default.
-- goal:agent-scene: six validated volume fog fields in agent-look (30 bytes gzip) so recipes can switch the opt-in fog scattering on.
diff --git a/changes/560-registry-wait.md b/changes/560-registry-wait.md
deleted file mode 100644
index 25edde5b..00000000
--- a/changes/560-registry-wait.md
+++ /dev/null
@@ -1 +0,0 @@
-- The publish workflow waits up to five minutes for the registry to serve a new version instead of two. The 0.16.0 publish succeeded, but the registry took longer than two minutes, so the check failed and skipped the release page.
diff --git a/changes/572-fast-path.md b/changes/572-fast-path.md
deleted file mode 100644
index e56d515c..00000000
--- a/changes/572-fast-path.md
+++ /dev/null
@@ -1 +0,0 @@
-- `applyAnimation(time, clip, { checks: false })` and `renderer.updateParts(parts, { checks: false })` skip the per-frame pose checks for clips you trust, and frames on the same clip no longer reset and snapshot every node. Eight animated characters: update CPU 1.6 ms to 1.3 ms with checks, 0.9 ms without.
diff --git a/changes/572-fast-path.size.md b/changes/572-fast-path.size.md
deleted file mode 100644
index 5f6f8fd2..00000000
--- a/changes/572-fast-path.size.md
+++ /dev/null
@@ -1,9 +0,0 @@
-- entry:core: the updateParts checks option, so trusted animation skips per-frame pose checks.
-- scene:lit-box: the same updateParts option, which lives in the core renderer.
-- goal:lit-box: 73 bytes for the checks option; accepted by the lead under the trade-off rule.
-- scene:gltf-box: the applyAnimation checks option and the same-clip path; skinned update CPU 1.6 ms to 0.9 ms trusted.
-- scene:gltf-box+lazy: same code as scene:gltf-box.
-- goal:gltf-box: 196 bytes for that speed-up; accepted on the same grounds.
-- scene:agent-scene+lazy: the same skinning and glTF pose code, which agent scenes import.
-- goal:agent-scene: about 100 bytes for the same animation speed-up; accepted on the same grounds.
-- scene:agent-scene: the same skinning and glTF pose code, which agent scenes import.
diff --git a/changes/572-frame.md b/changes/572-frame.md
deleted file mode 100644
index e5747767..00000000
--- a/changes/572-frame.md
+++ /dev/null
@@ -1 +0,0 @@
-Skinned character frames cost less CPU: palette math reads joint matrices in place and a normal pose allocates nothing, so eight animated characters update about 0.1 ms faster (#572)
diff --git a/changes/572-frame.size.md b/changes/572-frame.size.md
deleted file mode 100644
index 721c45c3..00000000
--- a/changes/572-frame.size.md
+++ /dev/null
@@ -1,7 +0,0 @@
-- entry:core: the offset matrix product and the joint check that remove per-joint copies and allocations from the skinned frame.
-- scene:gltf-box: the same skinning and glTF pose code; skinned update CPU about 0.1 ms less for eight characters.
-- scene:gltf-box+lazy: same code as scene:gltf-box.
-- goal:gltf-box: 485 bytes for a faster skinned frame, accepted by the lead under the trade-off rule (a measured speed-up, not a score).
-- scene:agent-scene: the same skinning code, which agent scenes import.
-- scene:agent-scene+lazy: same code as scene:agent-scene.
-- goal:agent-scene: about 200 bytes for the same skinned frame speed-up.
diff --git a/changes/572.md b/changes/572.md
deleted file mode 100644
index ceb6ef04..00000000
--- a/changes/572.md
+++ /dev/null
@@ -1 +0,0 @@
-- Animating an imported glTF character costs about a third of the CPU it did: `applyAnimation` prepares the pose once instead of twice, reuses its rollback snapshot, and skin validation checks each distinct set of joint weights once instead of every vertex. Eight RobotExpressive characters went from 4.5 ms to about 1.5 ms per frame.
diff --git a/changes/572.size.md b/changes/572.size.md
deleted file mode 100644
index 74f52a4c..00000000
--- a/changes/572.size.md
+++ /dev/null
@@ -1,7 +0,0 @@
-- entry:core: the distinct influence-set cache in skin validation, which cuts skinned animation CPU per frame by about three times.
-- scene:gltf-box: one-prepare applyAnimation, the reused rollback snapshot and the influence-set cache; 4.5 ms to about 1.5 ms per frame for 8 characters.
-- scene:gltf-box+lazy: same code as scene:gltf-box.
-- scene:agent-scene: the same skinning and glTF pose code, which agent scenes import.
-- scene:agent-scene+lazy: same code as scene:agent-scene.
-- goal:gltf-box: 346 bytes for a three times cheaper animated character is worth it; the lead accepts it under the trade-off rule.
-- goal:agent-scene: 231 bytes for the same skinning cost cut; accepted on the same grounds.
diff --git a/changes/573.md b/changes/573.md
deleted file mode 100644
index ecac55a7..00000000
--- a/changes/573.md
+++ /dev/null
@@ -1 +0,0 @@
-- Add a per-frame density and velocity `field` to volume fog (also `renderer.setFogField`), flow-advected noise (`flowStrength`, `flowPeriod`), a `profile` choice, and scattering from up to 8 point and spot lights in the fog (issue #573). A fog without a field or lights draws exactly as before.
diff --git a/changes/573.size.md b/changes/573.size.md
deleted file mode 100644
index 6f60c1c6..00000000
--- a/changes/573.size.md
+++ /dev/null
@@ -1,11 +0,0 @@
-- entry:core: the fog field upload, flow-advected noise and point/spot scattering are new agent-facing capability (fog that reacts and torches that light it); the fog shader and field code live in the core post pass, 1.8 KB gzip for the whole feature.
-- entry:replace-textured-parts: the same renderer and post shader growth, in a bundle that carries the core.
-- entry:texture-update: the same renderer and post shader growth, in a bundle that carries the core.
-- scene:lit-box: the same core growth; the fog code is in the post pass every scene carries.
-- scene:gltf-box: the same core growth.
-- scene:gltf-box+lazy: the same core growth.
-- scene:agent-scene: the same core growth.
-- scene:agent-scene+lazy: the same core growth.
-- goal:lit-box: functional fog capability from #573 (field, flow, local lights); moving the fog pass out of the core is a separate change.
-- goal:gltf-box: functional fog capability from #573; moving the fog pass out of the core is a separate change.
-- goal:agent-scene: functional fog capability from #573; moving the fog pass out of the core is a separate change.
diff --git a/changes/577.md b/changes/577.md
deleted file mode 100644
index e9fb7149..00000000
--- a/changes/577.md
+++ /dev/null
@@ -1 +0,0 @@
-- Breaking: a renderer that draws `environment.volumeFog` needs `import '@codepawlhq/67js/addons/fog-shader'` first (recipes, the film and the MCP server load it on their own), which takes about 2.5 KB gzip out of the core (#577)
diff --git a/changes/577.size.md b/changes/577.size.md
deleted file mode 100644
index d6b7ee78..00000000
--- a/changes/577.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: the fog code is a lazy chunk now, so the all-modules total pays about 0.8 KB gzip for a separate chunk and its loader; every scene without a fog saves 2.5 KB in the core.
diff --git a/changes/578.md b/changes/578.md
deleted file mode 100644
index 4d26c12c..00000000
--- a/changes/578.md
+++ /dev/null
@@ -1 +0,0 @@
-- Motion blur and bloom no longer leave a ghost of the previous frame after a seek (#578)
diff --git a/changes/578.size.md b/changes/578.size.md
deleted file mode 100644
index e32d52d5..00000000
--- a/changes/578.size.md
+++ /dev/null
@@ -1,5 +0,0 @@
-- goal:lit-box: 50 bytes gzip for the one-method fix that stops a ghost of the previous frame after a seek (#578)
-- goal:gltf-box: 58 bytes gzip for the same fix (#578)
-- goal:agent-scene: 63 bytes gzip for the same fix (#578)
-- scene:agent-scene: the mipmap filter reset before post stages (issue #578), same reason as the goal line.
-- scene:agent-scene+lazy: same reason.
diff --git a/changes/581.md b/changes/581.md
deleted file mode 100644
index bc6fbadc..00000000
--- a/changes/581.md
+++ /dev/null
@@ -1 +0,0 @@
-- A few mirrored skin vertices (a finger pinched between two joints) no longer fail the frame. They draw pinched, and `renderer.diagnostics` lists `skin-reflected-vertices` with the count and the joints and weights of the first one (#581)
diff --git a/changes/581.size.md b/changes/581.size.md
deleted file mode 100644
index c8682985..00000000
--- a/changes/581.size.md
+++ /dev/null
@@ -1,10 +0,0 @@
-- entry:core: the core entry exports skinning.js, which now counts mirrored weighted vertices and builds the warning text instead of throwing (issue #581). A frame that used to fail now draws.
-- entry:surface-attachment: carries skinning.js, same reason.
-- entry:retarget: carries skinning.js, same reason.
-- scene:gltf-box: loads skinning.js, same reason.
-- scene:gltf-box+lazy: loads skinning.js, same reason.
-- scene:agent-scene: loads skinning.js, same reason.
-- scene:agent-scene+lazy: loads skinning.js, same reason.
-- goal:lit-box: 23 bytes for the renderer to pass the skin warnings into renderer.diagnostics. Accepted by the lead under the trade-off rule: it turns a failed frame into a drawn one with a named warning.
-- goal:gltf-box: about 375 bytes of skinning.js for the same fix. Accepted by the lead: a skinned glTF frame that used to fail now draws.
-- goal:agent-scene: about 440 bytes of skinning.js for the same fix. Accepted by the lead for the same reason.
diff --git a/changes/586-additive.md b/changes/586-additive.md
deleted file mode 100644
index d5ab1a24..00000000
--- a/changes/586-additive.md
+++ /dev/null
@@ -1 +0,0 @@
-Materials accept `alphaMode: 'add'` for additive glow, and the opacity of blend and add parts changes each frame through `updateParts` without a new batch. (#586)
diff --git a/changes/586-additive.size.md b/changes/586-additive.size.md
deleted file mode 100644
index cc630904..00000000
--- a/changes/586-additive.size.md
+++ /dev/null
@@ -1,9 +0,0 @@
-- entry:core: additive blend mode and per-instance opacity (one more instance float), about 256 bytes gzip; sparks, trails and glow need both.
-- scene:lit-box: the same renderer code, which lives in the core renderer.
-- goal:lit-box: 279 bytes for additive blending and per-part opacity; an agent-facing VFX capability (issue 586).
-- scene:gltf-box: the same renderer code.
-- scene:gltf-box+lazy: same code as scene:gltf-box.
-- goal:gltf-box: 275 bytes for the same capability.
-- scene:agent-scene: the same renderer code plus the recipe alphaMode list.
-- scene:agent-scene+lazy: same code as scene:agent-scene.
-- goal:agent-scene: 301 bytes for the same capability.
diff --git a/changes/586-fog-detail.md b/changes/586-fog-detail.md
deleted file mode 100644
index 9504ebe8..00000000
--- a/changes/586-fog-detail.md
+++ /dev/null
@@ -1 +0,0 @@
-- Volume fog takes `detailOctaves`, `detailAmount`, `warp` and `contrast` for wispier mist, `profileSoftness` for where a layer top starts to fade, and `keyScattering` to scale only the key light shafts (#586)
diff --git a/changes/586-fog-detail.size.md b/changes/586-fog-detail.size.md
deleted file mode 100644
index dffd2646..00000000
--- a/changes/586-fog-detail.size.md
+++ /dev/null
@@ -1,8 +0,0 @@
-- entry:fog-shader: the six detail fields' validation and defaults, the warp, extra octave, contrast, layer top and key scattering GLSL, which a detail-free fog compiles out.
-- scene:agent-scene: the same fog-shader code, which this scene bundles, plus the six recipe field names.
-- goal:agent-scene: about 460 bytes, almost all the fog-shader code above; accepted under the trade-off rule.
-- scene:agent-scene+lazy: the same fog-shader code in the lazy build.
-- goal:lit-box: 49 bytes in the core (a spread of unknown fog fields and one post step line); the detail code itself lives in the fog addon.
-- goal:gltf-box: 48 bytes for the same core change.
-- tool:scene_create: six bounded volumeFog detail fields (detailOctaves, detailAmount, warp, contrast, profileSoftness, keyScattering) so agents can shape ground mist (issue #586)
-- tool-total: same six fields
diff --git a/changes/588.md b/changes/588.md
deleted file mode 100644
index 3d66d73f..00000000
--- a/changes/588.md
+++ /dev/null
@@ -1 +0,0 @@
-Metal with detailed normal and roughness maps no longer sparkles from frame to frame at grazing angles; `environment.specularAntiAliasing` sets the strength (#588)
diff --git a/changes/588.size.md b/changes/588.size.md
deleted file mode 100644
index 07ed6b07..00000000
--- a/changes/588.size.md
+++ /dev/null
@@ -1,11 +0,0 @@
-- entry:core: the Kaplanyan-Hoffman and Toksvig terms in the lit fragment shader plus the strength uniform and its validation; fixes visible frame-to-frame sparkle on normal-mapped metal (flips over 40/255 fell by 73 percent), accepted under the trade-off rule.
-- scene:lit-box: the same shader and uniform code, about 215 bytes.
-- scene:gltf-box: the same shader and uniform code, about 217 bytes.
-- scene:gltf-box+lazy: same code as scene:gltf-box.
-- scene:agent-scene: the same shader code plus the environment field, about 242 bytes.
-- scene:agent-scene+lazy: same code as scene:agent-scene.
-- goal:lit-box: about 215 bytes for a visible specular anti-aliasing fix, accepted by the lead under the trade-off rule.
-- goal:gltf-box: about 217 bytes for the same fix.
-- goal:agent-scene: about 242 bytes for the same fix and its recipe field.
-- tool:scene_create: one bounded environment.specularAntiAliasing field so agents can tune or turn off the specular anti-aliasing (issue #588)
-- tool-total: same field
diff --git a/changes/593-false-pairs.md b/changes/593-false-pairs.md
deleted file mode 100644
index ba65b498..00000000
--- a/changes/593-false-pairs.md
+++ /dev/null
@@ -1 +0,0 @@
-- resolveCoplanar: resolveFlushParts no longer layers pairs that lie on one plane to within a micrometre (the depth buffer already settles them) and settles chains of equal-sized parts, so it stops adding flicker on dense scenes (#593)
diff --git a/changes/593-false-pairs.size.md b/changes/593-false-pairs.size.md
deleted file mode 100644
index 861de34b..00000000
--- a/changes/593-false-pairs.size.md
+++ /dev/null
@@ -1,4 +0,0 @@
-- goal:lit-box: 10 bytes for passing the exact-flush gap to resolveCoplanar so reversedDepth false keeps layering exact-flush pairs (#593)
-- goal:gltf-box: 16 bytes, same renderer line (#593)
-- goal:agent-scene: 13 bytes, same renderer line (#593)
-- scene:agent-scene+lazy: the exact-flush gap rule and layer settling in the coplanar addon; fixes flicker that resolveCoplanar caused (issue #593).
diff --git a/changes/593.md b/changes/593.md
deleted file mode 100644
index e4b25c17..00000000
--- a/changes/593.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- A film with a very long lens (far / near over 10,000, such as near 0.08 and far 9000) now switches to reversed-Z float depth on its own: `Renderer67` loads `reversed-depth.js` as a separate chunk on the first such frame and every frame after it draws with it (`await renderer.reversedDepthReady` to have it from frame one; `reversedDepth: false` keeps the 24-bit buffer). On the metro film, 24-bit depth could not tell surfaces 2 micrometres apart at 2.6 m, and stairs, canopy panels, handrails and steel flickered between frames; those pairs went from thousands of flipping pixels to under 60 (#593)
-- `resolveCoplanar` is now a function, `resolveFlushParts` from `67js/addons/coplanar-overlap`, so the plane-keyed search stays out of the core bundle. Its depth layers also work under reversed depth: the slope part of the offset is 16 times larger there, because the constant part moves a surface by under a micrometre in a float depth buffer (#593)
diff --git a/changes/593.size.md b/changes/593.size.md
deleted file mode 100644
index e3a46c81..00000000
--- a/changes/593.size.md
+++ /dev/null
@@ -1,14 +0,0 @@
-- entry:core: the lens check and the lazy import of reversed depth; the coplanar search is no longer in the core.
-- entry:reversed-depth: the 16 times slope scale and the false-mode flag, about 1 KB gzip more than main with the earlier branch work counted.
-- entry:coplanar-overlap: the plane-keyed search that replaces the capped triangle product (issue #593), 1 KB gzip.
-- entry:gltf+lazy: renderer imports counted again for the depth-layer code from the earlier commits of this branch; needs the lead's trade-off call.
-- entry:replace-textured-parts: the same renderer imports; needs the lead's trade-off call.
-- entry:texture-update: the same renderer imports; needs the lead's trade-off call.
-- scene:lit-box: the same renderer imports.
-- scene:gltf-box: the same renderer imports.
-- scene:gltf-box+lazy: the same renderer imports.
-- scene:agent-scene: the same renderer imports.
-- scene:agent-scene+lazy: the same renderer imports.
-- goal:lit-box: 545 bytes for reversed depth on a stretched lens and flush-layer support; a fixed glitch, not a score gain.
-- goal:gltf-box: the same bytes.
-- goal:agent-scene: the same bytes.
diff --git a/changes/595-more-constraints.md b/changes/595-more-constraints.md
deleted file mode 100644
index 26d10eef..00000000
--- a/changes/595-more-constraints.md
+++ /dev/null
@@ -1 +0,0 @@
-- motion.constraints: adds copy_transforms, limit_distance, limit_scale, stretch_to, locked_track, transformation and action, each one scene_motion call (#595)
diff --git a/changes/595-more-constraints.size.md b/changes/595-more-constraints.size.md
deleted file mode 100644
index d0cc3c3b..00000000
--- a/changes/595-more-constraints.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- scene:agent-scene+lazy: seven constraint operators and their validation load only with motion (#595)
-- tool:scene_create: the seven new type names join the floor entry's type enum and one args record carries their fields, validated in node-constraints with errors that list the allowed fields (#595)
-- tool-total: the same enum and args record, counted once (#595)
diff --git a/changes/596.md b/changes/596.md
deleted file mode 100644
index eba621b7..00000000
--- a/changes/596.md
+++ /dev/null
@@ -1 +0,0 @@
-- Recipes gain `cone`, `icoSphere`, `circle` (filled or ring) and `line` geometries, and `plane` is accepted as an alias of `grid` (#596)
diff --git a/changes/596.size.md b/changes/596.size.md
deleted file mode 100644
index be4089fa..00000000
--- a/changes/596.size.md
+++ /dev/null
@@ -1,6 +0,0 @@
-- scene:agent-scene: cone, icoSphere and circle generators plus the line and plane recipe types, 1129 bytes gzip, close the most common primitive gaps that agents otherwise build from cylinders and lathes (#596)
-- scene:agent-scene+lazy: the same generators, 1120 bytes gzip
-- goal:agent-scene: the four primitives are an agent-facing capability; the generators share the existing surface helper
-- tool-total: four new geometry records (cone, icoSphere, circle, line) and the plane name in the recipe schema of two tools, 3124 bytes in all (#596)
-- tool:scene_create: the four geometry records, 1562 bytes
-- tool:scene_apply: the same recipe schema, 1562 bytes
diff --git a/changes/597-field-surface-sampling.md b/changes/597-field-surface-sampling.md
deleted file mode 100644
index 9a60c5a2..00000000
--- a/changes/597-field-surface-sampling.md
+++ /dev/null
@@ -1 +0,0 @@
-- Fields can read other surfaces: `nearest` (distance, point or normal of the closest point on a surface) and `raycast` (hit, distance, point or normal along a direction, with a `miss` value), plus `attr: group` to read a vertex group back. In `instances` the surface is a geometry id, so a scatter can avoid a path mesh and follow terrain; in `set_position`, `displace`, `vertex_group` and `vertex_color` it is a geometry record, and `vertex_group` stores the result by name (#597)
diff --git a/changes/597-field-surface-sampling.size.md b/changes/597-field-surface-sampling.size.md
deleted file mode 100644
index aae9320b..00000000
--- a/changes/597-field-surface-sampling.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: the surface inputs and the BVH query wrapper load only with the mesh chunk and buy scatter that avoids paths and follows terrain, plus density maps stored as vertex groups; the core bundle shrinks by 4 bytes because new capability text is paid for by shorter notes.
diff --git a/changes/598-force-fields.md b/changes/598-force-fields.md
deleted file mode 100644
index d05da4f7..00000000
--- a/changes/598-force-fields.md
+++ /dev/null
@@ -1 +0,0 @@
-- Force fields: a `fields` list on the recipe `particles` and `cloth` sections accelerates particles, hair and cloth through one shared evaluation, with wind, turbulence, vortex, drag and magnet types, `strength`, `falloff`, `radius` and `seed`. The same seed gives the same frames (#598).
diff --git a/changes/598-force-fields.size.md b/changes/598-force-fields.size.md
deleted file mode 100644
index 1f3407fd..00000000
--- a/changes/598-force-fields.size.md
+++ /dev/null
@@ -1,4 +0,0 @@
-- scene:agent-scene+lazy: force-fields.js (about 2 KB gzip), the cloth-fields.js wrapper and the fielded particle integrator load only with the cloth and particles modules, and the fields list reaches cloth and hair as well as particles.
-- entry:cloth: one optional hook call in the substep loop so the field code stays out of the core.
-- tool-total: the cloth section of scene_create accepts the `fields` list (138 bytes); field details live in scene_capabilities.
-- tool:scene_create: the `fields` list on the cloth section.
diff --git a/changes/599-ocean.md b/changes/599-ocean.md
deleted file mode 100644
index 7a13a03c..00000000
--- a/changes/599-ocean.md
+++ /dev/null
@@ -1 +0,0 @@
-- Ocean modifier: the `ocean` mesh edit and stack node turn a flat grid into a seeded sea (Phillips spectrum for the wind, `choppiness`, `size`, `seed`, `time`) and write the vertex group `foam` where the surface folds. The same seed and time give identical vertices. Part of #599.
diff --git a/changes/599-ocean.size.md b/changes/599-ocean.size.md
deleted file mode 100644
index a397b37e..00000000
--- a/changes/599-ocean.size.md
+++ /dev/null
@@ -1,4 +0,0 @@
-- entry:mesh: the ocean modifier is about 1.2 KB gzip of mesh code that agents cannot get any other way.
-- scene:agent-scene+lazy: the ocean modifier lives in the lazy part, which holds the whole mesh engine.
-- goal:agent-scene: the ocean modifier adds its validated field list and note to the agent-scene core (55 bytes gzip); the wave code is in the lazy mesh part.
-- scene:agent-scene: 112 bytes for the ocean modifier's validated field list and its note, once main's note trims overlapped the ones this pull request had made; the ocean is the only way for an agent to make an open sea.
diff --git a/changes/600-motion-blur.md b/changes/600-motion-blur.md
deleted file mode 100644
index 91f8fdc2..00000000
--- a/changes/600-motion-blur.md
+++ /dev/null
@@ -1 +0,0 @@
-- Per-object motion blur: `effects.motionBlur.objects` (0 to 1, off by default) smears each part that moves during the shutter along its own screen path, while the still background stays sharp; it replaces camera motion blur on that frame and loads on demand (#600)
diff --git a/changes/600-motion-blur.size.md b/changes/600-motion-blur.size.md
deleted file mode 100644
index decb50e0..00000000
--- a/changes/600-motion-blur.size.md
+++ /dev/null
@@ -1,8 +0,0 @@
-- scene:agent-scene+lazy: the object blur stage and its packing load on demand and only for a recipe that sets motionBlur.objects; nothing runs for any other scene
-- goal:lit-box: 31 bytes gzip in the shared effect resolver (the optional objects field) and the pass texture getter, which the lazy stage needs
-- goal:gltf-box: the same 36 bytes gzip of shared effect and pass code, in a scene that never loads the stage
-- goal:agent-scene: the same 32 bytes gzip of shared effect and pass code, in a scene that never loads the stage
-- tool-total: the shared effects schema gains the opt-in `motionBlur.objects` field (122 bytes in each of the three tools that embed it).
-- tool:scene_create: the `motionBlur.objects` field of the shared effects schema.
-- tool:scene_apply: the `motionBlur.objects` field of the shared effects schema.
-- tool:scene_create_asset: the `motionBlur.objects` field of the shared effects schema.
diff --git a/changes/601-lighting-passes.md b/changes/601-lighting-passes.md
deleted file mode 100644
index 91bbafb1..00000000
--- a/changes/601-lighting-passes.md
+++ /dev/null
@@ -1 +0,0 @@
-- render jobs: passes now include `diffuse`, `glossy`, `shadow`, `ao` and `mist` next to albedo, normal, emission, depth and objectId; diffuse + glossy + emission add back to the lit image, and `exr` keeps values above 1 (#601)
diff --git a/changes/601-lighting-passes.size.md b/changes/601-lighting-passes.size.md
deleted file mode 100644
index aef13eb9..00000000
--- a/changes/601-lighting-passes.size.md
+++ /dev/null
@@ -1,4 +0,0 @@
-- entry:render-passes: the lighting passes (diffuse, glossy, shadow, ao, mist) bring their GLSL (lighting-pass-shader.js) into the lazy render-passes addon, which only a render job with passes loads; 524 bytes gzip for five new outputs.
-- goal:agent-scene: 22 bytes gzip for the five new pass names in the render job schema and the pass mode range in the main shader; the lighting GLSL stays in the lazy addon.
-- goal:lit-box: 10 bytes gzip for the pass mode range in the main shader (renderPassMode < 5.5), which lets the lighting passes reuse the lit shader; their GLSL stays in the lazy render-passes addon.
-- goal:gltf-box: the same 10 bytes gzip for the pass mode range in the main shader.
diff --git a/changes/602-post-steps.md b/changes/602-post-steps.md
deleted file mode 100644
index 2341ba33..00000000
--- a/changes/602-post-steps.md
+++ /dev/null
@@ -1 +0,0 @@
-- Three named post steps join defocus (`effects.depthOfField`) in the look: `effects.colorBalance` (`lift`, `gamma`, `gain`, each three numbers), `effects.glare` (`streaks`, `count`, `angle`, `fogGlow`, `threshold`, `size`) and `effects.lensDistortion` (`distortion`, `dispersion`). Each is off until a recipe names it and draws nothing at its neutral values, so a recipe without them renders the same pixels as before. Their code is a lazy module that loads only for a recipe that names one; `installPostSteps(renderer)` from `src/post-steps.js` adds the stages (#602)
diff --git a/changes/602-post-steps.size.md b/changes/602-post-steps.size.md
deleted file mode 100644
index f94e05b9..00000000
--- a/changes/602-post-steps.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- scene:agent-scene+lazy: the lazy post-steps module (three GLSL stages, the resolver and the stage installer), loaded only for a recipe that names colorBalance, glare or lensDistortion; it brings the camera-effects stage code with it.
-- goal:agent-scene: a module marker, a loader line and two reads of the registry that carry the three field names (43 bytes gzip); every stage and shader is lazy.
-- tool-total: three bare group fields (49 bytes in each of scene_create, scene_apply and scene_create_asset); the fields and ranges are in docs/agent-look.md and the engine's errors.
diff --git a/changes/603-bake-stale.md b/changes/603-bake-stale.md
deleted file mode 100644
index 537b3983..00000000
--- a/changes/603-bake-stale.md
+++ /dev/null
@@ -1 +0,0 @@
-- `scene_inspect` reports a `bake-stale` warning when motion has moved a light, the sun or a part off the pose a light bake was traced at; a scene without motion or without a bake pays nothing. `bake.rebake: true` bakes each pose a render or seek samples (the last 4 are kept, about 0.7 s each on a small room), so the baked light matches the frame and the warning does not apply. Material animation is not checked, see docs/light-bake.md (#603)
diff --git a/changes/603-bake-stale.size.md b/changes/603-bake-stale.size.md
deleted file mode 100644
index 33dfeeca..00000000
--- a/changes/603-bake-stale.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: the stale check, the bake key, the pose-keyed rebake cache (last 4 poses) and `bake.rebake` live in the lazy bake chunk, which imports the motion look code it needs to sample lights (+1027 bytes gzip). The core scene:agent-scene bundle shrinks by 10 bytes: the small rebake hook in `sampleMotion` is paid for by shorter capability notes.
diff --git a/changes/604-607.size.md b/changes/604-607.size.md
deleted file mode 100644
index 18386363..00000000
--- a/changes/604-607.size.md
+++ /dev/null
@@ -1,19 +0,0 @@
-- entry:core: orthoScale, focal length with sensor width and lens shift in applyProjection (about 245 bytes gzip); the renderer already drew orthographic, and agents could not reach it.
-- entry:core+lazy: the same projection code.
-- entry:camera-effects: effectsWithFStop maps a camera f-stop onto the existing depth of field blur and focus, so no second depth of field path exists.
-- scene:lit-box: the projection code in the core bundle.
-- scene:lit-box+lazy: the projection code in the core bundle.
-- scene:gltf-box: the projection code in the core bundle.
-- scene:gltf-box+lazy: the projection code in the core bundle.
-- scene:agent-scene: camera lens validation (projection, orthoScale, focalLength, sensorWidth, shift, fStop) and the fov or focalLength rule; validated in the scene so a bad lens fails at create and set-camera.
-- scene:agent-scene+lazy: the same validation.
-- goal:lit-box: 265 bytes for the agent-reachable lens; the scene gains orthographic and shifted cameras, which agents could not set before.
-- goal:gltf-box: 246 bytes for the same lens code.
-- goal:agent-scene: 1.3 KB for lens validation and motion that keeps the lens; an agent-facing capability.
-- tool:scene_create: six camera fields (about 990 bytes) shared by every tool that takes a camera.
-- tool:scene_apply: the same six camera fields on set-camera.
-- tool:scene_create_asset: the same six camera fields.
-- tool:scene_review: the same six camera fields on shot cameras.
-- tool-total: four tools carry the six new camera fields, with short descriptions; the 16 point and 7 area light caps add no bytes.
-- result:scene_capabilities: one camera note that holds the rules the field descriptions leave out.
-- entry:agent-scene: the same camera lens validation as scene:agent-scene; an agent-facing capability (ortho, focal length, shift, f-stop).
diff --git a/changes/604.md b/changes/604.md
deleted file mode 100644
index 9588bf7c..00000000
--- a/changes/604.md
+++ /dev/null
@@ -1,2 +0,0 @@
----
-- MCP scene_create and set-environment now accept the 16 point, 8 spot and 7 area lights the engine renders, read from one limits file, not 8 point and 4 area (#604)
diff --git a/changes/605.md b/changes/605.md
deleted file mode 100644
index f523336c..00000000
--- a/changes/605.md
+++ /dev/null
@@ -1 +0,0 @@
-- Motion tracks animate lights, the sun, sky and fog, exposure and focus distance, so a sunset, a lamp flicker or a rack focus plays from a recipe (#605)
diff --git a/changes/605.size.md b/changes/605.size.md
deleted file mode 100644
index aa093d27..00000000
--- a/changes/605.size.md
+++ /dev/null
@@ -1,6 +0,0 @@
-- scene:agent-scene: light, sun, environment and effects tracks need target checks, ranges, a patch per kind and two getters on the scene (about 330 bytes gzip); the code that applies them loads with the motion module (#605).
-- scene:agent-scene+lazy: same code, in the lazy motion module (#605).
-- goal:agent-scene: a sunset, lamp flicker or rack focus could not be authored before; the core change is the two getters and the pose publish line, the rest sits in the lazy motion module (#605).
-- tool-total: one track schema covers four kinds, with the properties of each kind in scene_capabilities motionLook instead of the tool text.
-- tool:scene_create: the single lookTrack schema in the motion tracks union.
-- result:scene_capabilities: motionLook lists the kinds, ids and properties once, so the tool schemas stay small.
diff --git a/changes/606.md b/changes/606.md
deleted file mode 100644
index 38bc2e67..00000000
--- a/changes/606.md
+++ /dev/null
@@ -1 +0,0 @@
-- Nodes take visible (keyable in a motion track), castShadow, receiveShadow, holdout and shadowCatcher, so films hide parts without parking them and renders composite over footage (#606)
diff --git a/changes/606.size.md b/changes/606.size.md
deleted file mode 100644
index 8c8520f8..00000000
--- a/changes/606.size.md
+++ /dev/null
@@ -1,15 +0,0 @@
-- entry:core: per-part receiveShadow, holdout and shadowCatcher need batch state, two uniforms and three shader lines, the first renderer support for hiding a part and compositing over footage (#606).
-- entry:core+lazy: same renderer and shader lines as entry:core.
-- scene:lit-box: same renderer and shader lines as entry:core.
-- scene:lit-box+lazy: same renderer and shader lines as entry:core.
-- scene:gltf-box: same renderer and shader lines as entry:core.
-- scene:gltf-box+lazy: same renderer and shader lines as entry:core.
-- scene:agent-scene: five node fields, a visible motion track, hidden-subtree resolution and part flags (about 650 bytes beyond the renderer): films stop parking hidden parts in the batch list (#423).
-- scene:agent-scene+lazy: same agent-scene lines.
-- goal:lit-box: the renderer lines are the only way to give a part its own shadow and composite behaviour, and hiding or holding out a mesh for a composite is a visible capability.
-- goal:gltf-box: same renderer lines as goal:lit-box.
-- goal:agent-scene: five node fields and the keyed visible track replace a workaround that films used (#423).
-- tool-total: the five node fields appear in every scene tool that takes nodes; the descriptions are one line and the detail sits in scene_capabilities renderFlags.
-- tool:scene_create: the five node fields in the node schema.
-- tool:scene_apply: the five node fields in the node schema, repeated for add-node and set-node.
-- result:scene_capabilities: the renderFlags entry that carries the detail the tool schemas leave out.
diff --git a/changes/607.md b/changes/607.md
deleted file mode 100644
index 4958378a..00000000
--- a/changes/607.md
+++ /dev/null
@@ -1,2 +0,0 @@
----
-- Scene cameras take projection orthographic with orthoScale, focalLength with sensorWidth (instead of fov), shift [x, y] and fStop for depth of field, in recipes and set-camera (#607)
diff --git a/changes/608.md b/changes/608.md
deleted file mode 100644
index a8c15c11..00000000
--- a/changes/608.md
+++ /dev/null
@@ -1 +0,0 @@
-- environment.hdri and a .cube environment.look: shipped HDRIs (sky-clouds, studio-softbox) and looks (teal-orange, noir), or catalog files by ID, work from recipes and MCP (#608)
diff --git a/changes/608.size.md b/changes/608.size.md
deleted file mode 100644
index bb6cc282..00000000
--- a/changes/608.size.md
+++ /dev/null
@@ -1,7 +0,0 @@
-- entry:view-transform-shader: a catalog .cube look is looked up by name in the renderer's look resolver (164 bytes gzip), so a recipe's look can be a shipped or catalog .cube file.
-- scene:agent-scene+lazy: HDRI and .cube catalog entries add the name lists, the registry, two validated fields (hdri, look) and a capabilities entry (about 640 bytes gzip); the decoding runs in the preview page, not in the recipe code.
-- goal:agent-scene: about 100 KB of shipped files in assets/looks (four files: two 128 x 64 HDRIs, two 9-point looks) sit outside every bundle; a bigger HDRI comes from the asset catalog.
-- tool-total: the new hdri field and the wider look field in the one shared environment schema (343 bytes), the only MCP surface for catalog HDRIs and .cube looks.
-- tool:scene_create: the hdri field and the wider look field (343 bytes), with names listed as an enum so an agent cannot guess a path.
-- result:scene_capabilities: the catalogLooks entry lists the shipped HDRI and look names (308 bytes), so an agent finds them without a call to the docs.
-- scene:agent-scene: HDRI and catalog .cube look support in the agent scene (issue #608); opt-in, default renders unchanged.
diff --git a/changes/609-material-graph-outputs.md b/changes/609-material-graph-outputs.md
deleted file mode 100644
index 04514fc6..00000000
--- a/changes/609-material-graph-outputs.md
+++ /dev/null
@@ -1 +0,0 @@
-- Material graphs: new outputs `opacity` (base texture alpha), `clearcoat` and `transmission` (per-pixel masks), and mixing nodes `layer` (mask-blended layers such as paint over rust), `hue` and `curve`. The rasterizer reads opacity and clearcoat, the CPU path tracer all three. Graphs without them bake the same (#609)
diff --git a/changes/609-material-graph-outputs.size.md b/changes/609-material-graph-outputs.size.md
deleted file mode 100644
index bf8b66cd..00000000
--- a/changes/609-material-graph-outputs.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: the material graph gains opacity, clearcoat and transmission outputs and the layer, hue and curve nodes (about 1.3 KB gzip in the lazily loaded graph baker), so an agent can paint over rust or fade a part in one recipe.
diff --git a/changes/610.md b/changes/610.md
deleted file mode 100644
index baab1fec..00000000
--- a/changes/610.md
+++ /dev/null
@@ -1 +0,0 @@
-- A recipe mesh takes `shapeKeys` (named offsets or move-only edits), animated by the existing morph tracks and written as morph targets in a GLB (#610)
diff --git a/changes/610.size.md b/changes/610.size.md
deleted file mode 100644
index ae08c23c..00000000
--- a/changes/610.size.md
+++ /dev/null
@@ -1,7 +0,0 @@
-- scene:agent-scene: the shape key part handling (morph pose per part, weight publish, validation call), 481 bytes gzip (#610)
-- scene:agent-scene+lazy: shape key normalizing and offset targets in the mesh engine, 1204 bytes gzip
-- goal:agent-scene: shape keys let agents animate faces, blinks and squash on built models; the engine code loads only with meshes
-- tool-total: one shapeKeys field on the mesh geometry in two tools, 500 bytes; the field details sit in scene_capabilities meshShapeKeys
-- tool:scene_create: the shapeKeys field, 250 bytes
-- tool:scene_apply: the same recipe schema, 250 bytes
-- result:scene_capabilities: meshShapeKeys explains keys, offsets, edits, value and the morph track in under 1 KB so the tool schemas stay small
diff --git a/changes/611-render-output.md b/changes/611-render-output.md
deleted file mode 100644
index f3862cd4..00000000
--- a/changes/611-render-output.md
+++ /dev/null
@@ -1 +0,0 @@
-- Render jobs gain four optional output settings, all off by default so a plain job writes the same bytes: `transparent: true` (PNG with alpha, for `image` and `stills`), `format: "png16"` or `"exr"` (16-bit PNG and OpenEXR from a half-float frame, passes included: depth in metres as float32 `Z`, object IDs as float32 `id`), `region: { x, y, width, height }` (a crop of the whole frame, for `image`, `stills` and `video`) and `fps` up to 60. The encoders are small hand-written Node code with no new dependency. `docs/agent-render-jobs.md` has the details.
diff --git a/changes/611-render-output.size.md b/changes/611-render-output.size.md
deleted file mode 100644
index 21b2b94c..00000000
--- a/changes/611-render-output.size.md
+++ /dev/null
@@ -1,6 +0,0 @@
-- entry:render-passes: a pass target can be half float (128 bytes gzip, a static addon, not lazy), which 16-bit PNG and EXR passes read back instead of 8-bit bytes.
-- goal:lit-box: 9 bytes gzip for one `frame.outputFramebuffer ?? null` in the final pass, the only core change that lets a render job draw the finished frame into a float target.
-- goal:gltf-box: the same 11 bytes in the core renderer.
-- goal:agent-scene: 25 bytes gzip for the transparent flag and the lazy hook; the float target and its readback load on demand (`src/float-output.js`), and the encoders run in Node.
-- tool-total: scene_render_submit gains the transparent, format, region and fps options (771 bytes), read only by agents that submit render jobs.
-- tool:scene_render_submit: the transparent, format (png, png16, exr), region and fps options.
diff --git a/changes/612-lights.md b/changes/612-lights.md
deleted file mode 100644
index cb5c6057..00000000
--- a/changes/612-lights.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- Area lights can cast soft shadows: `castShadow: true` on an `areaLights` entry takes one tile of the local shadow atlas, and the penumbra widens with the rectangle and with the gap between blocker and receiver. Cost is 16 blocker taps and 32 filter taps per lit pixel, only for a shadowed area light.
-- `environment.sunAngularRadius` (radians, 0.0005 to 0.2) softens the sun shadow in the raster through the cascades (at least one slice is used) and sets the sun disk of the path tracer, which was fixed at 0.0087. Without the field nothing changes.
-- IES profiles: `ies` on spot and area lights takes a bundled name (`wide-flood`, `downlight`, `narrow-spot`, `batwing`) or `{ angles, candela }`; `parseIesPhotometry` reads an LM-63 file into that shape. The raster and the CPU path tracer read the same 16 samples. The GPU path tracer and the light bake ignore profiles.
diff --git a/changes/612-lights.size.md b/changes/612-lights.size.md
deleted file mode 100644
index 36f96aba..00000000
--- a/changes/612-lights.size.md
+++ /dev/null
@@ -1,10 +0,0 @@
-- entry:path-tracer: the sun disk radius and the 16-sample IES profile math (light-profile.js) so the tracer matches the raster.
-- entry:path-tracer-progressive: same tracer addition.
-- entry:cascaded-shadows: the soft sun (blocker search plus a 16 tap filter) lives in the lazy cascade module and runs only when sunAngularRadius is set.
-- scene:agent-scene: the recipe hook for sunAngularRadius, area castShadow and ies, three field names and a module marker (70 bytes gzip); the code is in the lazy light-shapes module.
-- scene:agent-scene+lazy: the lazy light-shapes and light-profile modules, loaded only for a recipe that names the new fields.
-- tool:scene_create: three bare schema fields (sunAngularRadius, ies on spot and area lights, castShadow on area lights), 76 bytes; the rules and names are in the docs and the recipe errors.
-- tool-total: the same three fields.
-- goal:lit-box: three one-line hooks in lighting, renderer and shaders (56 bytes gzip) that the lazy light-shapes module fills.
-- goal:gltf-box: the same three hooks (46 bytes gzip).
-- goal:agent-scene: the hooks plus the recipe fields (70 bytes gzip); everything else is lazy.
diff --git a/changes/613.md b/changes/613.md
deleted file mode 100644
index 4d7cb8af..00000000
--- a/changes/613.md
+++ /dev/null
@@ -1 +0,0 @@
-- `vertex_group` and `vertex_color` mesh edits write named weights and vertex colors; `displace` and cloth `pinGroup` read a group, and colors export as `COLOR_0` (#613)
diff --git a/changes/613.size.md b/changes/613.size.md
deleted file mode 100644
index 6213524a..00000000
--- a/changes/613.size.md
+++ /dev/null
@@ -1,6 +0,0 @@
-- tool-total: the two new mesh_edit ops (vertex_group, vertex_color) and the group and pinGroup fields add 1496 bytes of schema; the field details sit in scene_capabilities meshAttributes (#613)
-- tool:mesh_edit: vertex_group and vertex_color are the two ops that let an agent weight and paint a mesh, 1411 bytes; the select and field types are shortened to a one-line description
-- result:scene_capabilities: meshAttributes explains both ops, group and pinGroup in about 1 KB, so the tool schemas do not carry that text
-- scene:agent-scene: capability notes and the cloth pinGroup check in the core bundle, 406 bytes gzip (#613)
-- scene:agent-scene+lazy: the vertex attribute edits, group-aware displace and per-vertex group arrays in the mesh engine, 2121 bytes gzip
-- goal:agent-scene: weights and vertex colors are an agent-facing capability that Blender users expect; the code itself loads only with meshes
diff --git a/changes/614-particles.md b/changes/614-particles.md
deleted file mode 100644
index 0b2f4ecc..00000000
--- a/changes/614-particles.md
+++ /dev/null
@@ -1 +0,0 @@
-- Particle collision and look (#614): an emitter system takes `colliders` (static root nodes, bounced off by triangle), `bounce` and `friction` (also for the floor), `spin`, and `color` and `alpha` over life. A fast particle cannot tunnel through a thin mesh, a slow bounce settles, and the same seed gives the same frames after any seek. Alpha multiplies the vertex colors, a fade to black, because the renderer has no per-vertex alpha.
diff --git a/changes/614-particles.size.md b/changes/614-particles.size.md
deleted file mode 100644
index cad87cb7..00000000
--- a/changes/614-particles.size.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- scene:agent-scene+lazy: particle collision, bounce, friction, spin, and color and alpha over life load only with the particles module (about 2.1 KB gzip: the triangle path test, the contact response, the per-particle step cache, the new validators); a scene without particles never loads it.
-- goal:agent-scene: 29 bytes gzip for the one-line collision entry in scene_capabilities, which an agent needs to find the new fields; no tool schema grows because the particles section is already an open record.
diff --git a/changes/615-field-math.md b/changes/615-field-math.md
deleted file mode 100644
index adb3c4a2..00000000
--- a/changes/615-field-math.md
+++ /dev/null
@@ -1 +0,0 @@
-- Field math and curves: field expressions gain `rotate`, `align` and `transform` for rotation and matrix math. Instances take a `curve` source (`points`, `count`, `closed`, `fillet`) that places points at equal arc length, rounds corners and, with `alignToNormal`, turns each instance along the curve. The `bounding_box` and `convex_hull` modifiers replace a mesh with its box or hull (#615).
diff --git a/changes/615-field-math.size.md b/changes/615-field-math.size.md
deleted file mode 100644
index bc040876..00000000
--- a/changes/615-field-math.size.md
+++ /dev/null
@@ -1,6 +0,0 @@
-- entry:mesh: rotation operators for fields, curve resample and fillet for instances, and the bounding box and convex hull modifiers are about 1.8 KB gzip of mesh code that agents cannot get any other way.
-- scene:agent-scene+lazy: the new mesh operators live in the lazy part, which holds the whole mesh engine.
-- goal:agent-scene: the core grows under 64 bytes for the operator names; the notes were shortened elsewhere to pay for the rest.
-- tool:scene_create: the instance spec gains `curve` (points, count, closed, fillet), the only way for an agent to place instances along a path; the instance spec is inlined in each recipe field that carries nodes, so one field costs about 400 bytes.
-- tool:scene_apply: same `curve` instance field as scene_create, inlined through the shared node record.
-- tool-total: the `curve` instance field in scene_create and scene_apply, as above.
diff --git a/changes/622.md b/changes/622.md
deleted file mode 100644
index 697f93bd..00000000
--- a/changes/622.md
+++ /dev/null
@@ -1 +0,0 @@
-- Render jobs take `settings.renderer.backend: "gpu"` to path trace on the GPU, film and stills jobs can path trace every frame with one kernel compile, and recipe bakes run on the GPU when the page can, with `scene_inspect` naming the backend (#622)
diff --git a/changes/622.size.md b/changes/622.size.md
deleted file mode 100644
index 5cd24b0f..00000000
--- a/changes/622.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: the GPU bake page kernel; recipe bakes run on the GPU (68 bytes gzip)
diff --git a/changes/623-tree-2.md b/changes/623-tree-2.md
deleted file mode 100644
index 43563209..00000000
--- a/changes/623-tree-2.md
+++ /dev/null
@@ -1 +0,0 @@
-- The light tree splits by the surface area orientation heuristic, draws 90 percent by the angle-aware measure, runs on the WebGPU tracer (megakernel and wavefront), and `lightSelection` defaults to `"auto"`: the tree for a scene with more than 8 local lights, uniform otherwise (#623)
diff --git a/changes/623-tree-2.size.md b/changes/623-tree-2.size.md
deleted file mode 100644
index d3a66055..00000000
--- a/changes/623-tree-2.size.md
+++ /dev/null
@@ -1,5 +0,0 @@
-- entry:path-tracer: the SAOH light tree builder and the "auto" light selection; many-lights FLIP against Cycles goes from 0.1471 (uniform) to 0.0593 at 64 samples.
-- entry:path-tracer+lazy: the same light tree, now also packed for the WebGPU megakernel and wavefront kernels.
-- entry:path-tracer-progressive+lazy: the progressive preview resolves "auto" per scene and carries the WebGPU tree.
-- entry:path-tracer-realtime+lazy: shares the WebGPU tracer chunk that gained the tree.
-- entry:light-bake-gpu+lazy: shares the path tracer chunk that gained the tree.
diff --git a/changes/623-tree.md b/changes/623-tree.md
deleted file mode 100644
index 37cc88ab..00000000
--- a/changes/623-tree.md
+++ /dev/null
@@ -1 +0,0 @@
-- Path tracer `lightSelection: "tree"`: a light tree after Conty and Kulla (2018) as Cycles builds it, on the CPU reference and the WebGL2 kernel. On the new 24 lamp `many-lights` scene it scores 0.0617 against Cycles at 64 samples (uniform 0.1471) and takes 38 ms per full size sample against 72 ms; it stays off by default because it scores worse than uniform on `brick-room` and `light-linking` (issue #623)
diff --git a/changes/623-tree.size.md b/changes/623-tree.size.md
deleted file mode 100644
index 5667b305..00000000
--- a/changes/623-tree.size.md
+++ /dev/null
@@ -1,7 +0,0 @@
-- entry:path-tracer: the light tree builder and walk (about 1.7 KB gzip) let the CPU reference match the GPU light tree; it is the only way to draw many lights with fewer shadow rays
-- entry:path-tracer+lazy: the light tree builder and walk, plus the GLSL walk in the lazy kernel
-- entry:path-tracer-progressive: the light tree builder and the GLSL walk of the GPU kernel (about 2.2 KB gzip)
-- entry:path-tracer-progressive+lazy: the light tree builder and the GLSL walk of the GPU kernel
-- entry:path-tracer-realtime: the shared GPU kernel source carries the light tree walk, a branch the real-time path leaves off
-- entry:path-tracer-realtime+lazy: the shared GPU kernel source carries the light tree walk
-- entry:light-bake-gpu+lazy: the shared GPU kernel source carries the light tree walk, a branch the bake leaves off
diff --git a/changes/623.md b/changes/623.md
deleted file mode 100644
index 2a2fe06c..00000000
--- a/changes/623.md
+++ /dev/null
@@ -1 +0,0 @@
-- `67js/addons/path-tracer-progressive` draws the GPU path tracer one sample per pixel per step, starting at a quarter of the size, and `samplingPattern: "blue-noise-first"` and `lightSelection: "importance"` are new GPU options, both off by default (#623)
diff --git a/changes/623.size.md b/changes/623.size.md
deleted file mode 100644
index 34b41b4f..00000000
--- a/changes/623.size.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- entry:path-tracer+lazy: the blue noise first sample and the importance light choice are new GPU kernel paths plus the tile maker (loaded only on demand); the core bundle does not grow.
-- entry:light-bake-gpu: the shared GPU kernel text carries the two new options (compiled out unless asked); 19 bytes over the allowance.
diff --git a/changes/624-realtime-2.md b/changes/624-realtime-2.md
deleted file mode 100644
index 67d6d41a..00000000
--- a/changes/624-realtime-2.md
+++ /dev/null
@@ -1 +0,0 @@
-- Real-time WebGPU path tracer (#624): `update({ parts })` with the same part count keeps the history and the radiance cache, follows a part that only moved or turned (its matrix changed) through a per-part motion table, and rejects history on a part whose geometry changed, so a moving object leaves no trail; `followPartMotion: false` turns the following off for comparison
diff --git a/changes/624-webgpu.md b/changes/624-webgpu.md
deleted file mode 100644
index 1cd9e527..00000000
--- a/changes/624-webgpu.md
+++ /dev/null
@@ -1 +0,0 @@
-- `createRealtimeTracer(scene, { backend: 'webgpu' })` runs the real-time path tracer (#624) as WebGPU compute: the WebGPU megakernel with the radiance cache lookup and record built in, a cache update with atomic fixed point sums, and the temporal denoiser, a-trous filter and upsample as compute passes in one command buffer, with the picture copied to `canvas`. On the RTX 3060 the demo room at its own camera takes about 20 ms a frame at 1920 by 1080 (traced at 960 by 540) against 36 ms on WebGL2, and the kernel compiles in about 0.5 s against 11 s. WebGL2 stays the default and the fallback: without WebGPU the tracer's `backend` is `"gpu"` and `fallbackReason` says why. On the WebGPU backend `frame(camera, { wait: true }).done` and `readFrame()` are promises. The denoiser also rejects history where the material at the first hit differs (WebGPU backend only)
diff --git a/changes/624-webgpu.size.md b/changes/624-webgpu.size.md
deleted file mode 100644
index 9ebe4a18..00000000
--- a/changes/624-webgpu.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- entry:path-tracer-realtime+lazy: the WebGPU backend of the real-time tracer (WGSL kernel pieces, cache update and denoiser passes, about 9 KB gzip) loads only for `backend: 'webgpu'`; the demo room takes 20 ms a frame on it against 36 ms on WebGL2, and the kernel compiles in under a second.
diff --git a/changes/624.md b/changes/624.md
deleted file mode 100644
index 091c396b..00000000
--- a/changes/624.md
+++ /dev/null
@@ -1 +0,0 @@
-- `67js/addons/path-tracer-realtime` (`createRealtimeTracer(scene, options)`, then `frame(camera)`) is a real-time view of the WebGL2 path tracer: one path per pixel and frame at half the output size, two bounces whose last hit reads a spatial hash radiance cache, and a temporal denoiser (reprojection, variance, a-trous passes) that upsamples to the output size and draws to a canvas; the scene is static, motion vectors come from the camera, and primary hits are traced rather than rasterized (#624)
diff --git a/changes/624.size.md b/changes/624.size.md
deleted file mode 100644
index 8e3a7a5c..00000000
--- a/changes/624.size.md
+++ /dev/null
@@ -1,5 +0,0 @@
-- entry:path-tracer+lazy: the GPU kernel builder carries the radiance cache shader text (about 1 KB gzip) that only the new real-time addon compiles in; the GPU tracer's own pixels and speed do not change.
-- entry:path-tracer-progressive: same 1 KB of cache shader text in the kernel builder the progressive addon shares; it is only compiled in for the real-time addon.
-- entry:path-tracer-progressive+lazy: same 1 KB of cache shader text in the kernel builder.
-- entry:light-bake-gpu: same 1 KB of cache shader text in the kernel builder the GPU light bake shares.
-- entry:light-bake-gpu+lazy: same 1 KB of cache shader text in the kernel builder.
diff --git a/changes/625-parity.md b/changes/625-parity.md
deleted file mode 100644
index 14f99b28..00000000
--- a/changes/625-parity.md
+++ /dev/null
@@ -1 +0,0 @@
-- The WebGPU path tracer (`backend: 'webgpu'`) now traces volume fog and subsurface scattering, compiled into the megakernel only for a scene that uses them, and `bakeLightmapsAsync(..., { backend: 'webgpu' })` bakes lightmap texels on WebGPU with the same report fields. On the `fog` and `subsurface` reference scenes it is within 0.0039 FLIP of WebGL2 (480 by 270, 32 samples, no denoiser), and the `lit-interior` bake gives the same light as WebGL2 in 0.54 s against 0.76 s. The wavefront kernel still refuses fog and subsurface with a clear error. No default changes (#625)
diff --git a/changes/625-parity.size.md b/changes/625-parity.size.md
deleted file mode 100644
index 4669b538..00000000
--- a/changes/625-parity.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- entry:light-bake-gpu+lazy: the WGSL texel bake chunk (loaded only for backend 'webgpu') holds the WebGPU tracer with fog and the subsurface walk, so a bake on WebGPU is possible without WebGL2; the WebGL2 bake loads none of it.
-- entry:path-tracer+lazy: the WGSL fog and subsurface walk (3.9 KB gzip, in the lazy chunk loaded only for backend 'webgpu') so scenes with fog or subsurface no longer need the WebGL2 backend.
-- entry:path-tracer-progressive+lazy: the same lazy WebGPU chunk (3.9 KB gzip), shared with the progressive tracer.
diff --git a/changes/625.md b/changes/625.md
deleted file mode 100644
index 0b8b7417..00000000
--- a/changes/625.md
+++ /dev/null
@@ -1 +0,0 @@
-- The path tracer has a third backend, `backend: 'webgpu'`: the WebGL2 kernel written as WGSL compute (`src/path-tracer-webgpu.js`, loaded only for this backend, so the core bundle does not grow). It compiles in about a second instead of 8 to 20, matches the WebGL2 backend within 0.002 FLIP on the 16 reference scenes, and traces the demo room faster. It throws a clear error without `navigator.gpu`, and names volume fog and subsurface scattering (not ported yet) so callers fall back to `gpu`. No default changes (#625)
diff --git a/changes/625.size.md b/changes/625.size.md
deleted file mode 100644
index 0a729eed..00000000
--- a/changes/625.size.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- entry:path-tracer+lazy: the WebGPU path tracer backend adds two WGSL kernels (the megakernel and the wavefront, which share one prelude) and about 300 lines of host code, loaded only by `backend: 'webgpu'`; the core bundle is unchanged and it buys a kernel that compiles in about a second instead of 8 to 20
-- entry:path-tracer-progressive+lazy: the same WebGPU backend chunk, reachable from the progressive addon through the lazy backend import in path-tracer.js; loaded only by `backend: 'webgpu'`, the core bundle is unchanged.
diff --git a/changes/626-bake.md b/changes/626-bake.md
deleted file mode 100644
index d060123e..00000000
--- a/changes/626-bake.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- Recipe light bakes run on the native ray query tracer when `packages/render-native` is built: `bake.backend` is `native`, with the page's GPU kernel as the fallback (`bake.backendNote` says why). `bakeLightmapsAsync` takes `backend: "native"` with a `texelTracer`.
-- `renderer.maxSeconds` is honoured by `renderer.backend: "native"`: a trace that passes it fails the job instead of falling back.
diff --git a/changes/626-cutouts.md b/changes/626-cutouts.md
deleted file mode 100644
index 464412d4..00000000
--- a/changes/626-cutouts.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- The native tracer asks the hardware about alpha cutouts itself: cutout triangles are their own geometry and the kernel confirms a candidate hit when its colour map alpha reaches the cutoff, so rays no longer step through transparent texels (8 steps at most before).
-- Shadow rays on the CPU, WebGL2 and WebGPU tracers no longer leak light through dense alpha cutouts. Their any-hit walk took the first hit traversal met as if it were the nearest, so a transparent texel far along the ray let it skip closer leaves. A solid any-hit still ends the walk; a transparent one now retraces for nearest hits. On 300 masked leaf cards the four backends agree within FLIP 0.0004, down from 0.043 between native and the rest.
diff --git a/changes/626-finish.md b/changes/626-finish.md
deleted file mode 100644
index 2c664e1e..00000000
--- a/changes/626-finish.md
+++ /dev/null
@@ -1 +0,0 @@
-- The native tracer (renderer.backend "native", recipe bakes) now traces volume fog and subsurface scattering with the same walk as the WebGPU megakernel, compiled in only for a scene that uses them, and honours `noiseThreshold` on the GPU: after 16 samples, and then every 8, a converged pixel draws no more, and `metrics.pathTracer.averageSamples` reports the mean a pixel drew. `JS67_NATIVE_TRACER=off` keeps a job on the gpu or cpu backend whatever is installed; a native job without the package gives the cpu backend's bytes and report fields (#626)
diff --git a/changes/626-foliage-gap.md b/changes/626-foliage-gap.md
deleted file mode 100644
index 396c02d0..00000000
--- a/changes/626-foliage-gap.md
+++ /dev/null
@@ -1 +0,0 @@
-- Fixed: the CPU, WebGL2 and WebGPU path tracers gave up after 8 transparent cutout texels on one ray, so dense foliage showed holes and lost shadows; the limit is now 64, and the foliage reference scene matches the native tracer to FLIP 0.001
diff --git a/changes/626-jobs.md b/changes/626-jobs.md
deleted file mode 100644
index 97c9822a..00000000
--- a/changes/626-jobs.md
+++ /dev/null
@@ -1 +0,0 @@
-- renderer.backend "native" on render jobs (image, stills, video): hardware ray queries through the optional @codepawlhq/67js-render-native package, then the usual wavelet or OIDN denoise and PNG. Without the package, its binary or a Vulkan ray query adapter the job runs on gpu (cpu on a cpu runtime) and metrics.pathTracer.backendNote says why (#626)
diff --git a/changes/626.md b/changes/626.md
deleted file mode 100644
index 8c9ff131..00000000
--- a/changes/626.md
+++ /dev/null
@@ -1 +0,0 @@
-- `@codepawlhq/67js-render-native` is an optional Rust addon that traces a scene on the GPU with Vulkan ray queries, with a bench against the browser WGSL kernel (#626)
diff --git a/changes/635.md b/changes/635.md
deleted file mode 100644
index 6ccf53ba..00000000
--- a/changes/635.md
+++ /dev/null
@@ -1 +0,0 @@
-- The path tracer draws the textures and cutouts of loaded glTF files, which it used to trace untextured (#635)
diff --git a/changes/647-baseline.md b/changes/647-baseline.md
deleted file mode 100644
index 68ccb586..00000000
--- a/changes/647-baseline.md
+++ /dev/null
@@ -1 +0,0 @@
-The agent bench has a committed baseline for main: 28 tasks, 3 runs each, on Haiku (pass@1 99%) and Sonnet (pass@1 95%).
diff --git a/changes/647.md b/changes/647.md
deleted file mode 100644
index 370c2312..00000000
--- a/changes/647.md
+++ /dev/null
@@ -1 +0,0 @@
-- bench: `bench/agent-modeling/bench.mjs` runs a 28-task agent bench (modeling, revision, scene, look, light, camera, animation and film tasks, each with an automatic grader) k times per model through headless `claude -p`, and prints pass@1, pass^k, calls, tokens and time per model; `compare-bench.mjs` diffs a branch against a baseline file (#647)
diff --git a/changes/648.md b/changes/648.md
deleted file mode 100644
index a652f4da..00000000
--- a/changes/648.md
+++ /dev/null
@@ -1 +0,0 @@
-- MCP tool schemas shrink from 196,826 to about 66,000 bytes (#648). The tools list common fields and one line naming the rest, the strict schema still checks every call, and scene_capabilities answers with a short index and serves each recipe part by topic (motion, environment, materialFields, meshEdit, meshEdit.<op> and more). CI now has absolute ceilings for the schema total and each tool.
diff --git a/changes/648.size.md b/changes/648.size.md
deleted file mode 100644
index 5943de85..00000000
--- a/changes/648.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- tool:scene_capabilities: the new topic field is what lets every other tool shed its long field lists; the tool grows by about 200 bytes and the other tools shrink by over 130,000.
-- scene:agent-scene: set-environment now accepts the opt-in volume fog fields (#661), 32 bytes gzip.
-- goal:agent-scene: the fix for #661 reuses the field list agent-look already holds; 32 bytes gzip for a validation path that rejected valid fog fields.
diff --git a/changes/649.md b/changes/649.md
deleted file mode 100644
index e27a4153..00000000
--- a/changes/649.md
+++ /dev/null
@@ -1 +0,0 @@
-- The `build-67js-scene` and `render-67js-media` skills are rewritten for agents that use 67js through MCP: short entry files, references for look, light, camera, film and animation, render jobs and error recovery, and a working tool call for lens and depth of field, light caps, animated lights, sun and sky, catalog HDRIs and looks, `bake.gridFog`, fog multiple scattering, depth fixes and render flags. Evals under `skills/*/evals` run through `claude -p` with `scripts/run-skill-evals.mjs` (issue #649)
diff --git a/changes/650.md b/changes/650.md
deleted file mode 100644
index 8314cbfc..00000000
--- a/changes/650.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- A Claude Code plugin: `/plugin marketplace add codepawl/67.js` then `/plugin install 67js@67js` adds the 67js MCP server (started with `npx` from the npm package) and the build-67js-scene and render-67js-media skills (#650)
-- `npm run plugin:sync` copies the package.json version into the plugin manifests, and a test fails when they differ (#650)
diff --git a/changes/651.md b/changes/651.md
deleted file mode 100644
index 8bced77f..00000000
--- a/changes/651.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene_apply and mesh_edit answer with one entry per operation (index, op, target, what changed, a warning when nothing changed) and a diff of node, material and geometry IDs added, removed and changed; entries past 20 are counted, not listed (#651)
diff --git a/changes/651.size.md b/changes/651.size.md
deleted file mode 100644
index 370c24af..00000000
--- a/changes/651.size.md
+++ /dev/null
@@ -1,4 +0,0 @@
-- scene:agent-scene: describeApplication is exported from the agent-scene entry so library users get the same per-operation answer as MCP clients (77 bytes gzip).
-- scene:agent-scene+lazy: same export, 88 bytes gzip.
-- goal:agent-scene: the per-operation report saves agents a review call after every batch; it is 77 bytes gzip in the core entry.
-- tool:scene_apply: the description names the new results and diff fields, so an agent reads them instead of re-inspecting the scene (128 bytes).
diff --git a/changes/652.md b/changes/652.md
deleted file mode 100644
index 78e2fd7f..00000000
--- a/changes/652.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene_review, scene_look and the render result warn, with a field and a value to try, on an empty frame (`empty-frame`), a subject out of frame or too small (`subject-out-of-frame`, `subject-too-small`) and a black or clipped frame (`exposure-black`, `exposure-clipped`). The `bake-stale` warning of #603 now also arrives in a look, a review shot and a render job's metrics, not only in scene_inspect. They are in the frame's `diagnostics`, and a render job keeps one entry per warning (#652)
diff --git a/changes/652.size.md b/changes/652.size.md
deleted file mode 100644
index 29afeaac..00000000
--- a/changes/652.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- scene:agent-scene: worldBounds and a bakeStaleness alias (183 bytes gzip) let the preview tell an agent that a frame is empty or framed wrong, which it otherwise finds only by reading the image.
-- goal:agent-scene: 183 bytes for one small scene read that the empty-frame and out-of-frame warnings need; the sphere and luminance math sits outside the scene bundle.
-- scene:agent-scene+lazy: the measured growth against the recorded base (3556 bytes gzip) is mostly the merged main work of #603 and #543; this change adds the framing and exposure checks (about 1.1 KB, outside the scene bundle) and the 183 bytes above.
diff --git a/changes/653-mesh-edits.md b/changes/653-mesh-edits.md
deleted file mode 100644
index 24ee649d..00000000
--- a/changes/653-mesh-edits.md
+++ /dev/null
@@ -1 +0,0 @@
-- Mesh edit validation errors (cut, deform, generator, modifier, normal and attribute edits, and the parameter checks of the edit log) now read `edits[2].divisions: got 9; expected a whole number from 2 to 8, such as 2.`: the field path, the value received and a range, a list of choices or an example. Inside a recipe the path starts at the geometry (`geometries[0].edits[2].levels`). Size-limit errors for subdivide, array and subdivide_edges now name the field to lower (#653)
diff --git a/changes/653-mesh-edits.size.md b/changes/653-mesh-edits.size.md
deleted file mode 100644
index e859e994..00000000
--- a/changes/653-mesh-edits.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: 555 bytes gzip of message text and the small mesh-edit-field helper, so that a rejected mesh edit names the edit field, the value received and an example; nothing new runs unless an edit is rejected.
diff --git a/changes/653-motion.md b/changes/653-motion.md
deleted file mode 100644
index 643e8fe2..00000000
--- a/changes/653-motion.md
+++ /dev/null
@@ -1 +0,0 @@
-- Errors for motion tracks and IK, clip layers, part relations, branches, path queries and crowds now name the field, the value and an example: `nodes[1].relations[0].side: got "+w"; expected one of +x, -x, +y, -y, +z, -z, the face of "a" it sits against.` The shared motion helpers (`bounded`, `denseArray`, `knownKeys`) keep their wording for now (#653)
diff --git a/changes/653-motion.size.md b/changes/653-motion.size.md
deleted file mode 100644
index 644197f4..00000000
--- a/changes/653-motion.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: 955 bytes gzip of message text in the motion, clip layer, relation, branch, path query and crowd validators; nothing new runs unless a recipe is rejected, and an agent fixes the input in one retry.
diff --git a/changes/653-nature.md b/changes/653-nature.md
deleted file mode 100644
index e7531953..00000000
--- a/changes/653-nature.md
+++ /dev/null
@@ -1 +0,0 @@
-- The same readable errors now cover vegetation (`nodes[3].vegetation.scatter.density: got 9; expected a number from 0.0001 to 4, such as 1.`), `environment.wind`, particles and hair, cloth sheets and colliders, and image sequence maps: each names the full field path, the value received and an accepted example or range (#653)
diff --git a/changes/653-nature.size.md b/changes/653-nature.size.md
deleted file mode 100644
index ac70da3b..00000000
--- a/changes/653-nature.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: about 700 bytes gzip of message text in the vegetation, particle, hair, cloth and image sequence validators; nothing new runs unless a recipe is rejected, and an agent fixes the input in one retry.
diff --git a/changes/653-rest.md b/changes/653-rest.md
deleted file mode 100644
index be0a092e..00000000
--- a/changes/653-rest.md
+++ /dev/null
@@ -1 +0,0 @@
-- The same readable errors now cover physics bodies, joints and ragdolls, vehicle drives (`motion.drive`, wheels, path, speed, lights), asset recipes and operations (root, overrides, animation, inspect, checkpoints), placed assets and the shrinkwrap edit: `motion.drive.wheels[1].node: got "Nope"; expected the name of a node in the asset, such as one scene_inspect lists.` The mesh-edit parameter checks that still say `Mesh edit 2 (...)` are not converted yet (#653)
diff --git a/changes/653-rest.size.md b/changes/653-rest.size.md
deleted file mode 100644
index cf831cdf..00000000
--- a/changes/653-rest.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- entry:vehicle: 592 bytes gzip of message text in the vehicle kit (wheel, path, speed and light errors now name the field, the value and an example); nothing new runs unless a drive is rejected.
-- scene:agent-scene+lazy: 1.5 KB gzip of message text in the physics, drive, asset, placement and shrinkwrap validators; nothing new runs unless a recipe is rejected, and an agent fixes the input in one retry.
-- goal:agent-scene: 11 bytes gzip in the core bundle (message text only); the rest of the new text sits in lazy modules.
diff --git a/changes/653-scene-core.md b/changes/653-scene-core.md
deleted file mode 100644
index 2fc809f0..00000000
--- a/changes/653-scene-core.md
+++ /dev/null
@@ -1 +0,0 @@
-- Validation errors in the scene core now read `nodes[1].decal: got "yes"; expected true or false.`: scene operations (remove, set, inspect, checkpoints, revision), camera lens fields, geometry fields (points, holes, base, cutter), armatures, cloth and particle checks (`Cloth gravity: got [0,1]; expected three numbers, such as [0, 1, 0].`), terrain (`terrain.size: got 0; expected a number from 1 to 2000, such as 40.`) and mesh regions. Size limits and engine invariants keep their older wording (#653)
diff --git a/changes/653-scene-core.size.md b/changes/653-scene-core.size.md
deleted file mode 100644
index 87246b8a..00000000
--- a/changes/653-scene-core.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- scene:agent-scene: error text grows by 388 bytes gzip in the core bundle and 595 bytes with the lazy modules (field path, value received and an example in the scene, operation, camera, geometry, cloth, particle and terrain messages); nothing new runs unless a recipe is rejected, and an agent fixes the input in one retry.
-- scene:agent-scene+lazy: the same message text, 595 bytes gzip with the shared recipe checks, terrain and mesh modules; nothing new runs unless a recipe is rejected.
-- goal:agent-scene: 388 bytes gzip of message text only, so that agent errors name the field, the value and an example; this is the issue's goal and nothing new runs unless a recipe is rejected.
diff --git a/changes/653.md b/changes/653.md
deleted file mode 100644
index c442c999..00000000
--- a/changes/653.md
+++ /dev/null
@@ -1 +0,0 @@
-- Validation errors for materials, nodes, the camera, lights, fog, bake, mesh edits and motion now read `materials[2].emissiveColor: got [-1,0,0]; expected three numbers of 0 or more, such as [1, 0.5, 0.2].`: the full field path, the value received (cut at 40 characters) and a range, a list of choices or an example. Schema errors from the MCP tools add the received value too (`recipe.materials.0.color.2: Too big: expected number to be <=1 (got 2)`). Mesh edit errors keep the position of the edit (`edits[1].segments`). Not every message in the engine is converted yet; the rest keep their older wording (#653)
diff --git a/changes/653.size.md b/changes/653.size.md
deleted file mode 100644
index 1b723f76..00000000
--- a/changes/653.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- scene:agent-scene: error text grows by about 0.9 KB gzip (field path, received value and an example in each converted message, plus the 30-line field-error module); an agent that reads "emissiveColor: got [-1,0,0]; expected three numbers of 0 or more, such as [1, 0.5, 0.2]" fixes the input in one retry instead of guessing.
-- goal:agent-scene: 888 bytes gzip of message text only, so that agent errors name the field, the value and an example; this is the issue's goal and nothing new runs unless a recipe is rejected.
-- scene:agent-scene+lazy: the same message text, 1.5 KB gzip with the lazy mesh, motion and bake modules; nothing new runs unless a recipe is rejected.
diff --git a/changes/654.md b/changes/654.md
deleted file mode 100644
index 27baf60f..00000000
--- a/changes/654.md
+++ /dev/null
@@ -1 +0,0 @@
-- llms.txt is now a task map; its tool list is generated from the MCP registry (`npm run llms`) and a test fails when it is stale (#654)
diff --git a/changes/655.md b/changes/655.md
deleted file mode 100644
index be37b9f2..00000000
--- a/changes/655.md
+++ /dev/null
@@ -1 +0,0 @@
-- mcp: every tool has a title and annotations (only scene_dispose, scene_render_release and scene_release_checkpoint are destructive), six tools declare a small outputSchema with structuredContent, and scene_render_result returns the PNG as image content plus a resource link to each file (#655)
diff --git a/changes/661-volume-fog-optional-fields.md b/changes/661-volume-fog-optional-fields.md
deleted file mode 100644
index 33ffd632..00000000
--- a/changes/661-volume-fog-optional-fields.md
+++ /dev/null
@@ -1 +0,0 @@
-- set-environment accepts the optional volumeFog fields (scatterOctaves, multiScatter, multiExtinction, ...) after creation, and the MCP schema names scatterOctaves and multiScatter (#661).
\ No newline at end of file
diff --git a/changes/661.md b/changes/661.md
deleted file mode 100644
index fc36c9b3..00000000
--- a/changes/661.md
+++ /dev/null
@@ -1 +0,0 @@
-- set-environment accepts the opt-in volume fog fields (scatterOctaves, multiScatter, multiExtinction, multiAnisotropy, keyTransmittance, surfaceTransmittance and the rest) on a scene that did not set them at creation, and the MCP schema lists them (#661).
diff --git a/changes/664.md b/changes/664.md
deleted file mode 100644
index add3a742..00000000
--- a/changes/664.md
+++ /dev/null
@@ -1 +0,0 @@
-- MCP: shipped HDRI and .cube look names (sky-clouds, teal-orange and the rest) load in the packaged server built by npm run build:lib; they failed with "The URL must be of scheme file" (#664)
diff --git a/changes/680-mirrored-winding.md b/changes/680-mirrored-winding.md
deleted file mode 100644
index 49fc3923..00000000
--- a/changes/680-mirrored-winding.md
+++ /dev/null
@@ -1 +0,0 @@
-- Parts with a mirrored (negative-determinant) matrix, such as a flipped billboard, now render with the right side lit and culled instead of ghosted or pale; the renderer batches them apart and glTF no longer flips winding itself (#680, #681)
diff --git a/changes/680-mirrored-winding.size.md b/changes/680-mirrored-winding.size.md
deleted file mode 100644
index 30e812f9..00000000
--- a/changes/680-mirrored-winding.size.md
+++ /dev/null
@@ -1,11 +0,0 @@
-- entry:core: mirrored parts batch apart and draw with clockwise front faces, so the visible side of a flipped quad is lit and culled correctly
-- entry:core+lazy: same renderer change as entry:core
-- scene:lit-box: the renderer owns mirrored winding (a determinant sign test, a batch key and a frontFace call), which fixes the washed-out flipped quads
-- scene:lit-box+lazy: same renderer change as scene:lit-box
-- scene:gltf-box: the same renderer change, minus the loader's winding flip that this replaces
-- scene:gltf-box+lazy: same renderer change as scene:gltf-box
-- scene:agent-scene: the same renderer change as scene:lit-box
-- scene:agent-scene+lazy: same renderer change as scene:agent-scene
-- goal:lit-box: fixes a visible glitch (flipped billboards render ghosted or pale); about 140 B
-- goal:gltf-box: fixes a visible glitch with mirrored glTF and flipped quads sharing one path; about 110 B
-- goal:agent-scene: fixes a visible glitch (flipped billboards render ghosted or pale); about 140 B
diff --git a/changes/689.md b/changes/689.md
deleted file mode 100644
index 916edd3c..00000000
--- a/changes/689.md
+++ /dev/null
@@ -1 +0,0 @@
-- Add multi-bone `chain` IK (with `pole`, `poleAngle`, `stretch`, `iterations` and per-joint `limits`) and `spline` IK along Catmull-Rom curves to `addons/skeleton-pose` and recipe `motion.ik` (#689).
diff --git a/changes/689.size.md b/changes/689.size.md
deleted file mode 100644
index ccb0e002..00000000
--- a/changes/689.size.md
+++ /dev/null
@@ -1,5 +0,0 @@
-- entry:skeleton-pose: multi-bone CCD `chain` IK solver (with pole plane, poleAngle, bone stretch and joint limits) and Catmull-Rom `spline` IK solver (#689).
-- scene:agent-scene: `motionIk` capability names (`chain`, `spline`) in `agent-scene-capabilities.js` (#689).
-- scene:agent-scene+lazy: recipe `motion.ik` validation, staged stretch translation bindings and curve node evaluation for `chain` and `spline` solvers (#689).
-- entry:agent-asset: multi-bone chain IK and Spline IK solvers in imported asset motion (#689).
-- goal:agent-scene: 10 B gzip for `motionIk` capability names (`chain`, `spline`) in `agent-scene-capabilities.js` (#689).
diff --git a/changes/690-fcurve-modifiers.md b/changes/690-fcurve-modifiers.md
deleted file mode 100644
index bd986ac8..00000000
--- a/changes/690-fcurve-modifiers.md
+++ /dev/null
@@ -1 +0,0 @@
-- Motion tracks take `extrapolation` (`constant` or `linear`) and `modifiers`: `cycles` (repeat, offset, mirror, with counts), `stepped`, `noise` (seeded), `generator`, `limits` and `envelope`, sampled the same in sampleMotion, camera paths and render jobs (#690)
diff --git a/changes/690-fcurve-modifiers.size.md b/changes/690-fcurve-modifiers.size.md
deleted file mode 100644
index e389732e..00000000
--- a/changes/690-fcurve-modifiers.size.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- entry:agent-asset: track modifiers and extrapolation (noise, cycles, envelope and four more) load with the motion code, which agents use for camera shake and looping walks
-- scene:agent-scene+lazy: the same lazy motion code; the eager agent-scene bundle stays within its allowance
diff --git a/changes/691.md b/changes/691.md
deleted file mode 100644
index 5e2e4cb3..00000000
--- a/changes/691.md
+++ /dev/null
@@ -1 +0,0 @@
-- Add `maintain_volume`, `damped_track`, `pivot`, `follow_path`, `clamp_to`, `shrinkwrap` and `armature` motion constraints, plus `floor` `useRotation` and `copy_transforms`/`transformation` `mix` modes (#691).
diff --git a/changes/691.size.md b/changes/691.size.md
deleted file mode 100644
index 5872ef36..00000000
--- a/changes/691.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- scene:agent-scene+lazy: `maintain_volume`, `damped_track`, `pivot`, `follow_path`, `clamp_to`, `shrinkwrap` and `armature` constraints in `src/node-constraints.js` (#691).
-- entry:agent-asset: `maintain_volume`, `damped_track`, `pivot`, `follow_path`, `clamp_to`, `shrinkwrap` and `armature` constraints in `src/node-constraints.js` (#691).
-- goal:agent-scene: 38 B gzip for the 7 new constraint names in `motionConstraints` in `agent-scene-capabilities.js` (#691).
diff --git a/changes/692.md b/changes/692.md
deleted file mode 100644
index 4af754cf..00000000
--- a/changes/692.md
+++ /dev/null
@@ -1 +0,0 @@
-- Add `cylinder`, `cone` and `compound` collider shapes, `spring` joints, joint `motor` controls, timed `events` (`impulse`, `kinematicPose`, `jointMotor`) and `fields` (force fields on dynamic rigid bodies) to physics definitions and recipe `physics` sections (#692).
diff --git a/changes/692.size.md b/changes/692.size.md
deleted file mode 100644
index 1ed8f2da..00000000
--- a/changes/692.size.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- entry:physics: `cylinder` and `cone` colliders, `spring` joints and force-field evaluation on dynamic rigid bodies in `src/physics.js` (#692).
-- scene:agent-scene+lazy: `cylinder`, `cone` and `compound` colliders, `spring` joints, joint motors, timed events and force fields in `src/agent-physics.js` (#692).
diff --git a/changes/698-instancing.md b/changes/698-instancing.md
deleted file mode 100644
index 2201e447..00000000
--- a/changes/698-instancing.md
+++ /dev/null
@@ -1 +0,0 @@
-- Instancing: `instances` takes a `place` record with Poisson-disk `spacing`, a `density` field, `variants` of geometry and material picked at random or by a `variant` field, per-point `rotation` and `size` fields, and `at: "vertices"` or `"faces"` on a surface source to place on a mesh's own vertices and face centres. All of it is seeded (#698).
diff --git a/changes/698-instancing.size.md b/changes/698-instancing.size.md
deleted file mode 100644
index 2523863a..00000000
--- a/changes/698-instancing.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- scene:agent-scene+lazy: the placement module (Poisson spacing, density, variants, rotation and size fields, vertex and face points) loads only with instances, so scenes without them do not pay.
-- entry:agent-asset: the asset entry carries the agent scene's lazy mesh engine, so the Poisson spacing, density, variant and per-point placement code rides in it; it is the only way for an agent to scatter mixed variants without overlaps.
-- goal:agent-scene: 32 bytes for the `place` capability note after main's note trims overlapped the ones this pull request had made to pay for it.
diff --git a/changes/698.md b/changes/698.md
deleted file mode 100644
index 2befa2d8..00000000
--- a/changes/698.md
+++ /dev/null
@@ -1 +0,0 @@
-- Add `curve.length` spacing and `realize` (into one mesh per material) for recipe `instances`.
diff --git a/changes/698.size.md b/changes/698.size.md
deleted file mode 100644
index 3cae559f..00000000
--- a/changes/698.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: curve.length spacing and realize instances (merging transformed geometries per material) in src/mesh-instances.js and src/mesh-instance-placement.js.
diff --git a/changes/705-physical-sky.md b/changes/705-physical-sky.md
deleted file mode 100644
index a5fbbeda..00000000
--- a/changes/705-physical-sky.md
+++ /dev/null
@@ -1 +0,0 @@
-- `environment.sky: { type: 'physical', sunElevation, sunRotation, turbidity, groundAlbedo, aerialPerspective }` drives the key light direction and colour, the sky map, the background gradient and the distance haze from one sun, so a time of day is one parameter. Recipes without `sky` render byte for byte as before. (#705)
diff --git a/changes/705-physical-sky.size.md b/changes/705-physical-sky.size.md
deleted file mode 100644
index 18e415f7..00000000
--- a/changes/705-physical-sky.size.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- scene:agent-scene+lazy: the physical sky model, its sky map and the recipe fields load only for a recipe that names environment.sky; the core entry shrank by 20 bytes.
-- goal:agent-scene: 58 bytes for the physical sky hook and its capability note, after main's note trims overlapped the ones this pull request had made to pay for it.
diff --git a/changes/705.md b/changes/705.md
deleted file mode 100644
index 84229703..00000000
--- a/changes/705.md
+++ /dev/null
@@ -1 +0,0 @@
-- Add `ozone`, `sunDisc`, and time-of-day `environment` motion tracks for `environment.sky` (`type: "physical"`).
diff --git a/changes/705.size.md b/changes/705.size.md
deleted file mode 100644
index 184550ef..00000000
--- a/changes/705.size.md
+++ /dev/null
@@ -1 +0,0 @@
-- scene:agent-scene+lazy: ozone absorption, solar disc, and physical-sky environment motion tracks in src/physical-sky.js and src/agent-motion.js.
diff --git a/changes/706.md b/changes/706.md
deleted file mode 100644
index 569ee154..00000000
--- a/changes/706.md
+++ /dev/null
@@ -1 +0,0 @@
-- Add `position`, `velocity`, `uv`, `transmission`, `volume`, `environment`, `cryptomatte` and `aov` render passes (`material.aov` and material graph `aov` outputs), plus `multilayer: true` OpenEXR render jobs (#706).
diff --git a/changes/706.size.md b/changes/706.size.md
deleted file mode 100644
index 164313a6..00000000
--- a/changes/706.size.md
+++ /dev/null
@@ -1,12 +0,0 @@
-- entry:core: 69 B gzip for passSupport.bindFrame, passSupport.drawsInPass and passSupport.drawBatch hooks plus aov/aovMap/transmissionMap batch separation in groupParts (#706).
-- entry:core+lazy: 69 B gzip for the same passSupport hooks and batch separation in renderer.js (#706).
-- entry:render-passes: 8118 B gzip for position, velocity, UV, transmission, volume, environment, cryptomatte (FNV-1a manifest + float IDs) and AOV render passes (#706).
-- scene:gltf-box: 73 B gzip for the passSupport hooks and batch separation in renderer.js (#706).
-- scene:gltf-box+lazy: 73 B gzip for the passSupport hooks and batch separation in renderer.js (#706).
-- scene:agent-scene: 230 B gzip for material.aov validation, aovMap pattern field, material graph aov output and RENDER_PASS_NAMES (#706).
-- scene:agent-scene+lazy: 292 B gzip for material.aov validation, aovMap pattern field, material graph aov output and RENDER_PASS_NAMES (#706).
-- goal:lit-box: 63 B gzip for the passSupport hooks and batch separation in renderer.js (#706).
-- goal:gltf-box: 73 B gzip for the passSupport hooks and batch separation in renderer.js (#706).
-- goal:agent-scene: 230 B gzip for material.aov validation, aovMap pattern field, material graph aov output and RENDER_PASS_NAMES (#706).
-- tool-total: 214 B for material.aov and render settings passes/multilayer schemas on scene_render_submit (#706).
-- tool:scene_render_submit: 214 B for the 8 new pass names and multilayer EXR option (#706).
diff --git a/changes/713.md b/changes/713.md
deleted file mode 100644
index 8b3b2927..00000000
--- a/changes/713.md
+++ /dev/null
@@ -1 +0,0 @@
-- Unprojected ambient occlusion samples in view space rather than world space so cameras far from the origin do not quantize distant wall normals into vertical stripes (#713).
diff --git a/changes/718.md b/changes/718.md
deleted file mode 100644
index b7df432d..00000000
--- a/changes/718.md
+++ /dev/null
@@ -1 +0,0 @@
-- `scene_apply`, `scene_restore` and `mesh_edit` answers no longer carry the node inspection: they return the revision, the render status, the scene diagnostics when there are any, a result per operation and the diff. The results and diff of a 20-operation batch stay under 2 KB (the render status adds about 1 KB on top) instead of about 8 KB for the whole answer. Call `scene_inspect` to read the nodes (#718)
diff --git a/changes/718.size.md b/changes/718.size.md
deleted file mode 100644
index 73e6b6bb..00000000
--- a/changes/718.size.md
+++ /dev/null
@@ -1,2 +0,0 @@
-- goal:agent-scene: 9 bytes gzip with no change to any file the bundle imports (this issue only edits the preview script, docs and tests), so the difference looks like build noise between the two measurement checkouts; the issue itself shrinks the agent answer by about 6 KB per edit.
-- scene:agent-scene+lazy: the same noise, 15 bytes gzip, no imported file changed.
diff --git a/changes/726.md b/changes/726.md
deleted file mode 100644
index 812cdf1d..00000000
--- a/changes/726.md
+++ /dev/null
@@ -1 +0,0 @@
-- Keep mesh-leaf tree and blade grass vertex and material colors within `[0, 1]` while preserving their rendered `baseColor * vertexTint` product so `exportGlb` accepts mesh vegetation directly.
diff --git a/changes/731.md b/changes/731.md
deleted file mode 100644
index 18bca4d6..00000000
--- a/changes/731.md
+++ /dev/null
@@ -1 +0,0 @@
-- Report interior vertex valence (regular and poles), quad quality metrics and a normalized `topologyScore` in `lintMesh` and `mesh_inspect` (#731).
diff --git a/changes/731.size.md b/changes/731.size.md
deleted file mode 100644
index f12c07d8..00000000
--- a/changes/731.size.md
+++ /dev/null
@@ -1,3 +0,0 @@
-- scene:agent-scene+lazy: valence, quadQuality, topologyScore and quad distortion lint in mesh_inspect (#731)
-- result:mesh_inspect: valence, quadQuality and topologyScore fields in mesh reports (#731)
-- result:mesh_edit: valence, quadQuality and topologyScore fields in the returned mesh report (#731)
diff --git a/changes/733-734.md b/changes/733-734.md
deleted file mode 100644
index a7bc37fa..00000000
--- a/changes/733-734.md
+++ /dev/null
@@ -1 +0,0 @@
-- Fixed: automatic `objectReflections` skip with a reported reason when a scene has more than 256 other parts instead of failing `createAgentScene`, and `Renderer67.dispose()` deletes `liquidProgram` so transmissive scenes release every WebGL program (#733, #734).
diff --git a/changes/bistro-hero.md b/changes/bistro-hero.md
deleted file mode 100644
index 3e61158a..00000000
--- a/changes/bistro-hero.md
+++ /dev/null
@@ -1 +0,0 @@
-- The README opens on a Bistro street render under a Poly Haven sky, with two more Bistro views and the three showcase worlds below it. The Bistro scene is credited under CC BY 4.0.
diff --git a/changes/blender-inventory.md b/changes/blender-inventory.md
deleted file mode 100644
index a44237c9..00000000
--- a/changes/blender-inventory.md
+++ /dev/null
@@ -1 +0,0 @@
-- Add a Blender 5.2 inventory script and a report that compares it with the ledger and agent surface, with a ranked list of missing features
diff --git a/changes/film-checklist.md b/changes/film-checklist.md
deleted file mode 100644
index 54b5a466..00000000
--- a/changes/film-checklist.md
+++ /dev/null
@@ -1 +0,0 @@
-The scene skill lists the film glitches to check before a film goes out: specular sparkle, matte rails, depth of field motes, lens ghosts, sun shaft sources and dot-matrix signs.
diff --git a/changes/goal-doc.md b/changes/goal-doc.md
deleted file mode 100644
index 2760f0cf..00000000
--- a/changes/goal-doc.md
+++ /dev/null
@@ -1 +0,0 @@
-- The goal and its scope now live in docs/goal.md instead of a tracking issue
diff --git a/changes/readme-banner-b.md b/changes/readme-banner-b.md
deleted file mode 100644
index 3dc31bc9..00000000
--- a/changes/readme-banner-b.md
+++ /dev/null
@@ -1 +0,0 @@
-- The README banner is a full-width Bistro street render with the wordmark on the left, and the showcase table lists the airport, data center and port worlds again.
diff --git a/changes/readme-house-style.md b/changes/readme-house-style.md
deleted file mode 100644
index 9e553868..00000000
--- a/changes/readme-house-style.md
+++ /dev/null
@@ -1 +0,0 @@
-- The README follows the same layout as the other CodePawl repos: a banner with the wordmark and three renders, flat-square badges, a showcase table, an install section that starts with a prompt to paste into an agent, and a License heading.
diff --git a/changes/short-changelog.md b/changes/short-changelog.md
deleted file mode 100644
index 443e509e..00000000
--- a/changes/short-changelog.md
+++ /dev/null
@@ -1 +0,0 @@
-- The changelog groups its entries under headings, and 0.16.0 is rewritten as a short summary
diff --git a/changes/skills-capability-topics.md b/changes/skills-capability-topics.md
deleted file mode 100644
index 951a65f6..00000000
--- a/changes/skills-capability-topics.md
+++ /dev/null
@@ -1 +0,0 @@
-The scene skill points agents at the scene_capabilities topics for field detail, drops the notes for the fixed fog schema (#661) and packaged catalog (#664) gaps, and its HDRI, look and fog eval checks the shipped names and 3 scatter octaves again.
diff --git a/changes/tagline.md b/changes/tagline.md
deleted file mode 100644
index f954b7d8..00000000
--- a/changes/tagline.md
+++ /dev/null
@@ -1 +0,0 @@
-- The tagline is now "3D engine that AI agents drive" on the site, the package and the README banner
diff --git a/package.json b/package.json
index 9e5725ea..2212d836 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
{
"name": "@codepawlhq/67js",
- "version": "0.16.0",
+ "version": "0.17.0",
"private": false,
"type": "module",
"description": "3D engine that AI agents drive: an MCP server, PNG previews and CPU or GPU export to images and video",
diff --git a/scripts/smoke-package.mjs b/scripts/smoke-package.mjs
index 768d4b39..a5a2ceb7 100644
--- a/scripts/smoke-package.mjs
+++ b/scripts/smoke-package.mjs
@@ -16,7 +16,7 @@ assert.equal(release.version, metadata.version);
assert.equal(release.packageName, metadata.name);
assert.equal(release.packageAlias, '67js');
assert.equal(release.integrity, `sha512-${createHash('sha512').update(binary).digest('base64')}`);
-assert.ok(binary.length < 1024 * 1024, 'Library and optional addon wrappers must fit within a 1 MiB archive.');
+assert.ok(binary.length <= sizeBudgets.archiveBytes, `Library and optional addon wrappers must fit within the archive budget of ${sizeBudgets.archiveBytes} bytes.`);
assert.ok((await readFile('build/67.module.js')).length < 150000, 'The core must exclude the optional asset decoder.');
// The archive is measured against the base branch like a bundle. The release ceiling in budgets.json is reported, and
// npm run smoke:package -- --absolute enforces it.
diff --git a/src/version.js b/src/version.js
index 5e6a22a7..8d262655 100644
--- a/src/version.js
+++ b/src/version.js
@@ -1,2 +1,2 @@
// Kept as a literal so bundles do not carry all of package.json. tests/version.test.js keeps it in step.
-export const VERSION = '0.16.0';
+export const VERSION = '0.17.0';