Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(qb10): put the whole-suite numbers and the campaign state on the record

nirholasa8ee81ed27c4
diff --git a/prompts/finish/_context/quality-bar-README.md b/prompts/finish/_context/quality-bar-README.md
index aa8fcdabe..acb1e8e9c 100644
--- a/prompts/finish/_context/quality-bar-README.md
+++ b/prompts/finish/_context/quality-bar-README.md
@@ -22,7 +22,7 @@ special case.
| 04 | [04-pbr-texture-material-realism.md](../003-quality-bar-04-pbr-texture-material-realism.md) | Full PBR sets, measured presets, skin/eye/hair | Partial. Everything but the bench delta is verified: the channel matrix, the derive pass wired opt-out into `forge-store.js`, the presets rendered under all four shipped HDRIs (`docs/assets/material-presets/`), the avatar skin/eye/hair readback at 320/768/1440 (the check had been aimed at `/viewer`, which never runs that pass, and had been reporting zero meshes), and the mobile payload gate with measured 4K byte sizes (`npm run check:glb:payload`). The quality-bench comparison needs Vertex, and this Codespace's gcloud credentials have expired: `gcloud auth application-default login` unblocks it. |
| 07 | [07-design-system-sweep.md](../002-quality-bar-07-design-system-sweep.md) | Tokens, states, microinteractions sitewide | Open |
| 08 | [08-mobile-performance.md](../005-quality-bar-08-mobile-performance.md) | Mobile excellence, GLB delivery | Partial. A measured baseline exists in `_generated/08/`; the fixes and the re-measure are open. |
-| 10 | [10-avatar-likeness-irl-people.md](../007-quality-bar-10-avatar-likeness-irl-people.md) | IRL people: likeness, hands, rig, AR | Partial. Animation dignity is proven across ten rig conventions; one runtime-lane defect plus the likeness audit are open. |
+| 10 | [10-avatar-likeness-irl-people.md](../007-quality-bar-10-avatar-likeness-irl-people.md) | IRL people: likeness, hands, rig, AR | Partial. Rig half is DONE: animation dignity 10/10 across ten conventions (re-verified 2026-09-08), the clavicle-pivot defect is fixed, `audit:rig-coverage` is 204/204, and the rig-failure degrade path is kill-tested. The likeness audit is now MEASURED rather than open, with 44 rendered views and per-case bounding boxes in `prompts/quality-bar/_generated/10/pass-2026-09-08/` (read its `REPORT.md` first): 1 clean, 2 partial, 3 outright failures. Two causes were found and fixed (avatars were generated on the weaker TRELLIS lane despite the docs promising Hunyuan3D; text-to-3D skipped the rembg cutout so the reference's backdrop was reconstructed as a slab), plus two real production bugs in the vision chain that had left the quality gate fail-open passing everything. **What remains is a re-measure, not a diagnosis:** it needs the deploy plus `CRON_SECRET` for the high tier, neither available in that session. Hunyuan3D still has no matte support at all. |
Retired after verification (readable in git history): 01 photoreal reference pipeline, 02 the
Hunyuan3D flagship lane, 05 cinematic viewers, 06 the `/forge` generation flow (the pre-submit
diff --git a/prompts/quality-bar/_generated/10/pass-2026-09-08/REPORT.md b/prompts/quality-bar/_generated/10/pass-2026-09-08/REPORT.md
index b835f25c4..69936a49a 100644
--- a/prompts/quality-bar/_generated/10/pass-2026-09-08/REPORT.md
+++ b/prompts/quality-bar/_generated/10/pass-2026-09-08/REPORT.md
@@ -173,19 +173,29 @@ Green, run this session:
- `npm run audit:rig-coverage` clean
- `node scripts/animation-dignity-sweep.mjs` 10/10, re-run AFTER these changes
-A whole-suite `vitest run` in this worktree is **not** a clean signal right now:
-21 concurrent vitest processes from other agents share the same repo, and tests
-that touch shared files race. `tests/version-endpoint.test.js` ("trusts a
-pre-build snapshot ... then consumes it"), `tests/build-asset-paths.test.js` and
-`tests/server-404-routes.test.js` each failed in the shared run and each pass in
-isolation, verified individually.
-
-One genuine failure remains and is **not** from this work:
-`tests/audit-guards.test.js` reports that `data/guards.json` never describes the
-gate steps `check:skills-seed`, `audit:motion` and `audit:tour-global`, and that a
-`check-thing` fixture edits a `public/gone.json` that does not exist. That file and
-`package.json` are untouched by this pass; the recent commits there belong to
-another agent's in-flight guards registration.
+Whole-suite `vitest run`: **29,103 passed, 2 failed, 155 skipped (3 files red of
+2,018)**. None of the three is from this work, and all three are accounted for:
+
+- `tests/audit-guards.test.js`: genuine, and someone else's. `data/guards.json`
+ never describes the gate steps `check:skills-seed`, `audit:motion` and
+ `audit:tour-global`, and a `check-thing` fixture edits a `public/gone.json` that
+ does not exist. Neither that file nor `package.json` is touched by this pass; the
+ recent commits there belong to another agent's in-flight guards registration.
+- `tests/version-endpoint.test.js`: shared-file race. The case is literally
+ "trusts a pre-build snapshot ... then consumes it", and 21 concurrent vitest
+ processes from other agents were consuming the same snapshot. Passes in
+ isolation, verified.
+- `tests/extension-build.test.js`: shared-directory race:
+ `ENOENT ... chmod dist/extension/popup.html` on a file the build had just
+ written, i.e. a concurrent `vite build` (which runs with `emptyOutDir`) wiped
+ `dist/` mid-copy.
+
+`tests/build-asset-paths.test.js` and `tests/server-404-routes.test.js` also failed
+in an earlier shared run and both pass in isolation, verified individually.
+
+The Playwright stage was not run to completion: it needs the single shared dev
+server on port 3000, which concurrent agents were already using, and starting a
+second run kills the first one's server.
## What is still open