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: correct feature guides against the code they describe

nirholas69ce4b317fb8
diff --git a/docs/ALL.md b/docs/ALL.md
index da0a50d0b..fd429f5b8 100644
--- a/docs/ALL.md
+++ b/docs/ALL.md
@@ -33902,17 +33902,23 @@ The resolver returns three ready-to-run snippets, all pointed at the resolved
endpoint. The page shows them behind a `cURL` / `Node` / `Agent` tab strip with a
copy button.
-**cURL.** Step one shows the price by taking the `402`; step two pays and calls in
-one command:
+**cURL.** Step one shows the price by taking the `402`; step two replays the same
+call with the payment header:
```bash
# 1) See the price (returns HTTP 402 + payment requirements)
curl -i "https://three.ws/api/x402/token-intel?mint=<MINT_ADDRESS>"
-# 2) Pay & call in one step with the x402 CLI
-npx x402-fetch "https://three.ws/api/x402/token-intel?mint=<MINT_ADDRESS>"
+# 2) Pay: sign the requirement from step 1 into a base64 X-PAYMENT header,
+# then replay the same call with it. The Node tab below builds that
+# header for you; export it as X_PAYMENT to finish here in curl.
+curl -i -H "X-PAYMENT: $X_PAYMENT" "https://three.ws/api/x402/token-intel?mint=<MINT_ADDRESS>"
```
+There is no `x402-fetch` CLI to shell out to: the published package is a library
+(`wrapFetchWithPayment`, the Node tab), not a bin. Signing is what step two costs
+you, and the Node and Agent tabs are the two paths that do it for you.
+
**Node**, with `x402-fetch` wrapping a viem wallet client:
```js
diff --git a/docs/agent-identities.md b/docs/agent-identities.md
index 7e3906ebd..2f63d34e3 100644
--- a/docs/agent-identities.md
+++ b/docs/agent-identities.md
@@ -108,7 +108,9 @@ The endpoint is an A2MCP service: MCP Streamable HTTP, JSON-RPC `tools/call` ove
}
```
-`identity_status` is free, unauthenticated, and needs nothing but the job token (which is HMAC-signed, so it is the only capability required to read a job). Jobs persist for about 30 days. While running:
+`identity_status` is free, unauthenticated, and needs nothing but the job token (which is HMAC-signed, so it is the only capability required to read a job). The token carries no
+expiry, and the job document lives in object storage, so a handle stays pollable. While
+running:
```json
{
@@ -205,3 +207,4 @@ Two environment overrides help a batch run outlast the rig lane's rate-limit bac
- [Viewer](./viewer.md): the `/viewer` surface the `viewer_url` deep link opens.
- [MCP servers](./mcp.md): the wider set of three.ws MCP endpoints this one follows the transport conventions of.
- [Avatar Engines Atlas](./avatar-engines.md): what engine families exist for building human avatars, and which of them are wired into the commercial pipeline.
+- [Draft agent mint](./draft-agent-mint.md): the inactive on-chain identity minted for every finished reconstruction, Solana first with ERC-8004 alongside it.
diff --git a/docs/agent-runtime.md b/docs/agent-runtime.md
index fbedadfae..1f729483f 100644
--- a/docs/agent-runtime.md
+++ b/docs/agent-runtime.md
@@ -103,9 +103,12 @@ curl -s https://three.ws/api/agent/guard \
}'
```
-Response: one verdict per call, each with `decision`, `blockedBy`, a per-layer trace
-(`layers`), `warnings`, `blindSpots`, `coverageScore`, and `modifiedArguments` when the
-engine recommends safer parameters (the chat client applies these before executing).
+Response: `{ "engine": "@three-ws/agent-runtime", "verdicts": [...] }`, one entry per call,
+each `{ identifier, apiName, verdict }`. The verdict always carries `decision`, `reason`, a
+per-layer trace (`layers`), `warnings`, `blindSpots`, `coverageScore`, and `totalElapsedMs`.
+Two fields appear only when they apply: `blockedBy` (plus a `code`) names the layer that
+refused a `block`, and `modifiedArguments` carries safer parameters when the engine
+recommends them (the chat client applies those before executing).
Body fields: `calls[]` (each: `identifier`, `apiName`, `arguments`, optional `valueUsd`,
`protocol`, `destination`, `token`, `interventionConfig`, `executionPath`, `x402`), plus
diff --git a/docs/agent-symphony.md b/docs/agent-symphony.md
index 3be0c61ac..d31015786 100644
--- a/docs/agent-symphony.md
+++ b/docs/agent-symphony.md
@@ -47,14 +47,15 @@ visualization, and its own mute button in the legend.
|---|---|---|
| `money` | `payment`, `agora-earned` | A plucked triangle wave with a sine shimmer one octave up, 0.8s decay. |
| `bass` | `coin-buy` | A sine one octave below the note's pitch that glides down another octave over half a second. Whale buys are felt, not just heard. |
-| `bell` | `agent-deploy`, `agent-onchain`, `member-join`, `agora-registered` | An FM bell: sine carrier with an inharmonic partial at the classic 2.76 bell ratio, 1.4s ring. |
-| `arp` | `level-up`, `world-join`, `mission-complete`, `agora-task-posted`, `agora-hired`, `agora-task-claimed`, `agora-task-completed`, `agora-vouched` | The actor's stable three-note ascending motif, plucked at 90ms intervals. |
+| `bell` | `agent-deploy`, `agent-onchain`, `member-join`, `agora-registered`, `agora-arena-lost` | An FM bell: sine carrier with an inharmonic partial at the classic 2.76 bell ratio, 1.4s ring. |
+| `arp` | `level-up`, `world-join`, `mission-complete`, `agora-task-posted`, `agora-hired`, `agora-task-claimed`, `agora-task-completed`, `agora-vouched`, `agora-arena-entered`, `agora-guild-joined`, `agora-guild-contributed` | The actor's stable three-note ascending motif, plucked at 90ms intervals. |
| `alarm` | `agent-guard`, `agora-flagged` | Two short square waves a minor second apart: deliberate dissonance for safety refusals and flagged proofs. |
-| `jackpot` | `jackpot` | A sawtooth glissando through a resonant bandpass filter, rising two octaves, with sparkle plucks of the actor's motif on top. |
+| `jackpot` | `jackpot`, `agora-arena-won` | A sawtooth glissando through a resonant bandpass filter, rising two octaves, with sparkle plucks of the actor's motif on top. |
Any feed type without a mapping falls back to the `bell` voice, so a new
`ALLOWED_TYPES` entry is audible on day one; give it a proper voice mapping in
-`TYPE_TO_CATEGORY` when you add one.
+`TYPE_TO_CATEGORY` when you add one. `war-result` is the one allow-listed type
+riding that fallback today.
### Pitch, loudness, and identity
diff --git a/docs/airdrops.md b/docs/airdrops.md
index c1646dd4c..3b8df4423 100644
--- a/docs/airdrops.md
+++ b/docs/airdrops.md
@@ -60,7 +60,9 @@ Three layers behind it:
{
"address": "…",
"family": "solana",
+ "ts": "2026-08-06T20:00:00.000Z",
"activity": {
+ "family": "solana",
"tx_count": 342, "days_active": 61, "account_age_days": 410,
"last_active_days": 1, "unique_tokens": 12, "chains_active": 1,
"contract_interactions": null, "volume_usd": null,
diff --git a/docs/ar.md b/docs/ar.md
index 310c7ca1f..c9e3cece6 100644
--- a/docs/ar.md
+++ b/docs/ar.md
@@ -236,17 +236,23 @@ Poor AR performance almost always traces to model size or geometry complexity. A
**Optimization tools:**
```bash
-# Draco compress and optimize with gltf-transform (WebXR only — may break Quick Look/Scene Viewer)
-npx @gltf-transform/cli optimize model.glb optimized.glb --draco
+# Draco compress and optimize with gltf-transform (WebXR only; may break Quick Look/Scene Viewer)
+npx @gltf-transform/cli optimize model.glb optimized.glb --compress draco
-# Lossless optimization (safe for all three AR methods)
-npx @gltf-transform/cli optimize model.glb optimized.glb
+# Geometry left uncompressed and unsimplified (safe for all three AR methods)
+npx @gltf-transform/cli optimize model.glb optimized.glb --compress false --simplify false
# Resize textures
npx @gltf-transform/cli resize model.glb small.glb --width 1024 --height 1024
```
-> **Draco and Quick Look / Scene Viewer:** Draco-compressed GLBs require the Three.js Draco decoder — Quick Look and Scene Viewer don't include one, so they may refuse to load Draco GLBs. If you want AR across all three methods, compress with basis/KTX2 textures only, and leave geometry uncompressed.
+> **`optimize` compresses by default.** With no `--compress` flag it writes
+> `EXT_meshopt_compression` + `KHR_mesh_quantization`, and `--simplify` is on by
+> default too, so a bare `optimize` is neither lossless nor loadable by Quick
+> Look and Scene Viewer. Pass `--compress false --simplify false` for the
+> all-methods-safe output above.
+
+> **Draco and Quick Look / Scene Viewer:** Draco-compressed GLBs require the Three.js Draco decoder. Quick Look and Scene Viewer don't include one, so they may refuse to load Draco GLBs. If you want AR across all three methods, compress with basis/KTX2 textures only, and leave geometry uncompressed.
---
@@ -456,7 +462,7 @@ How this link ships through ChatGPT end to end (the app connector, the custom GP
- [AR on the homepage](https://three.ws/#home-ar) — live demo with real Forge models
- [Blog: See Your 3D Avatar in the Real World](https://three.ws/blog/see-your-3d-in-ar) — full walkthrough
-- [Avatar AR page](/avatars/:id/ar) — the dedicated AR experience for any avatar
+- Avatar AR page at `/avatars/<id>/ar`: the dedicated AR experience for any saved avatar. Open it from any avatar's **AR** tab, or start from [AR Forge](/ar).
- [Walk feature](/features/walk) — WebXR immersive walk mode (different from placement AR)
- [Web component reference](/docs/web-component): full `<agent-3d>` attribute list
- [Embedding guide](/docs/embedding) — iframe setup with XR permissions
diff --git a/docs/atlas.md b/docs/atlas.md
index 0c22b71c9..a1e3d6a6a 100644
--- a/docs/atlas.md
+++ b/docs/atlas.md
@@ -1,6 +1,6 @@
# Atlas: one keystroke to anywhere
-three.ws has more than 650 public routes. Atlas is the layer that makes that
+three.ws has more than 700 public routes. Atlas is the layer that makes that
navigable instead of overwhelming.
Press **Cmd+K** (Ctrl+K on Windows and Linux) on any page of the site. A search
@@ -45,9 +45,9 @@ Query: "get paid"
Get paid per API call
Meter any endpoint with x402 and let other agents pay to use it.
- 1. Wire the paywall in x402 Studio. -> /x402/studio
- 2. Follow the worked example end to end. -> /tutorials/paid-x402-endpoint
- 3. Watch the receipts and revenue come in. -> /x402-revenue
+ 1. Wire the paywall in x402 Studio. -> /x402/studio
+ 2. Follow the worked example of a paid endpoint end to end. -> /tutorials/paid-x402-endpoint
+ 3. Watch the receipts and revenue come in. -> /x402-revenue
```
Tasks appear above page results and only on a real signal (an exact-ish match on
diff --git a/docs/avatar-reconstruction.md b/docs/avatar-reconstruction.md
index 4be9c7177..4be398d80 100644
--- a/docs/avatar-reconstruction.md
+++ b/docs/avatar-reconstruction.md
@@ -155,10 +155,11 @@ cleanly: a pose that will not solve logs and leaves the warp-only skin in place.
It runs *after* the geometry morph, since projecting onto the template and then
morphing would slide the texture off the features it was sampled from.
-> **Fixed in this change:** `precompute_uv.py` and `glb_ops` previously read the
-> interleaved head buffer stride-unaware, so `face_uv_map.json` was built on
-> corrupted vertex data. The reader is now stride-aware, which also repairs the
-> Phase-1 texture correspondence.
+> **Every buffer read is stride-aware.** The head's vertex attributes share one
+> interleaved bufferView, so `precompute_uv.py` and `glb_ops` honour
+> `bufferView.byteStride` on every accessor read and write. A stride-unaware read
+> silently interleaves POSITION with NORMAL, which builds `face_uv_map.json` on
+> corrupted vertex data and breaks the Phase-1 texture correspondence too.
## Licensing — why this stack
@@ -186,7 +187,7 @@ each link:
| **v1 (shipped)** | Real face-shape morph (sparse) | MediaPipe Face Mesh | Apache-2.0 — clean |
| **v2 (recommended)** | Dense identity geometry, fused via `register_head_to_target` | **MICA + FLAME**, commercial licence from MPI | Paid MPI commercial licence — clean once signed. FLAME's fixed topology + expression basis map straight to ARKit |
| **v2 (fallback)** | Dense identity, no licence fee | HRN re-based on **FLAME-2023-Open** (CC-BY-4.0) | Clean, but weeks of GPU R&D to retrain the identity regressor |
-| **v2 (texture)** SHIPPED | Projective texturing off the morphed mesh: fills ears, jawline and neck from the *same* photo (10.4% to 37.5% of the head) | none, geometry we already have | Clean; no new model or licence |
+| **v2 (texture)** SHIPPED | Projective texturing off the morphed mesh: fills ears, jawline and neck from the *same* photo (10.5% to 41.1% of the head) | none, geometry we already have | Clean; no new model or licence |
| **v3 (texture)** | Inpaint what no camera saw (scalp, far cheek) | **Imagen** inpaint on Vertex AI | GCP, pre-approved |
| **v3** | Drop the RPM-template dependency for a fully-owned body | **Anny** (parametric body) + **ICT-FaceKit** (ARKit-52) + deformation transfer | Anny Apache-2.0 + CC0; ICT-FaceKit + DT MIT — clean |
diff --git a/docs/ca2x402.md b/docs/ca2x402.md
index 0d5a815a6..1affc2547 100644
--- a/docs/ca2x402.md
+++ b/docs/ca2x402.md
@@ -255,17 +255,23 @@ The resolver returns three ready-to-run snippets, all pointed at the resolved
endpoint. The page shows them behind a `cURL` / `Node` / `Agent` tab strip with a
copy button.
-**cURL.** Step one shows the price by taking the `402`; step two pays and calls in
-one command:
+**cURL.** Step one shows the price by taking the `402`; step two replays the same
+call with the payment header:
```bash
# 1) See the price (returns HTTP 402 + payment requirements)
curl -i "https://three.ws/api/x402/token-intel?mint=<MINT_ADDRESS>"
-# 2) Pay & call in one step with the x402 CLI
-npx x402-fetch "https://three.ws/api/x402/token-intel?mint=<MINT_ADDRESS>"
+# 2) Pay: sign the requirement from step 1 into a base64 X-PAYMENT header,
+# then replay the same call with it. The Node tab below builds that
+# header for you; export it as X_PAYMENT to finish here in curl.
+curl -i -H "X-PAYMENT: $X_PAYMENT" "https://three.ws/api/x402/token-intel?mint=<MINT_ADDRESS>"
```
+There is no `x402-fetch` CLI to shell out to: the published package is a library
+(`wrapFetchWithPayment`, the Node tab), not a bin. Signing is what step two costs
+you, and the Node and Agent tabs are the two paths that do it for you.
+
**Node**, with `x402-fetch` wrapping a viem wallet client:
```js
diff --git a/docs/character-library.md b/docs/character-library.md
index 76e9402eb..fc3d46142 100644
--- a/docs/character-library.md
+++ b/docs/character-library.md
@@ -34,7 +34,7 @@ Gallery controls:
## API: `GET /api/avatars/library`
-Public, no authentication, CORS-open to GET from web origins. The endpoint proxies a small manifest JSON from the R2 CDN (`avatars/library/manifest.json`) with an edge cache (`Cache-Control: public, s-maxage=300, stale-while-revalidate=3600`). The GLB and thumbnail bytes never pass through the API: every entry carries absolute CDN URLs the browser loads directly.
+Public, no authentication, `GET` and `OPTIONS` only (`HEAD` is answered like `GET`). Browser `fetch` from a third-party origin is not enabled: the endpoint sends `Access-Control-Allow-Origin` only for three.ws and its partner origins, exactly like the [Object Library API](./object-library.md). Call it server-side, or from a three.ws page. The endpoint proxies a small manifest JSON from the R2 CDN (`avatars/library/manifest.json`) with an edge cache (`Cache-Control: public, s-maxage=300, stale-while-revalidate=3600`). The GLB and thumbnail bytes never pass through the API: every entry carries absolute CDN URLs the browser loads directly.
### Query parameters
diff --git a/docs/choreography.md b/docs/choreography.md
index 5a07fdc07..cc17a4658 100644
--- a/docs/choreography.md
+++ b/docs/choreography.md
@@ -34,7 +34,7 @@ Compose one at **[three.ws/choreograph](https://three.ws/choreograph)**. This pa
| `loop` | `true` restarts the routine instead of ending it. Use it for waiting states, not greetings. |
| `steps` | 1 to 24 beats, played back to back in order. |
| `steps[].slot` | One of the 19 gesture slots ([the vocabulary](animations.md#agent-slots)). Required. |
-| `steps[].clip` | Optional. Pins this step to one specific clip instead of letting the slot resolve. `null` is the normal case. |
+| `steps[].clip` | Optional. Pins this step to one specific clip instead of letting the slot resolve. `null` is the normal case. Naming the slot's own platform default is stored as `null`, because pinning today's default would freeze the step against a later remap. |
| `steps[].hold` | How long the beat occupies the timeline, 0.2 to 20 seconds. Defaults to 1.6. |
| `steps[].speed` | Playback rate, 0.25 to 3. A step at `2` occupies half its `hold` on the timeline. |
@@ -198,7 +198,7 @@ Name|slot:hold[*speed][@clip],slot:hold,…
- `Welcome|wave:2,nod:1.2` gives two beats
- `Slow%20clap|celebrate:3*0.5` plays at half speed
-- `Encore|dance:4@rumba` pins a specific clip
+- `Encore|dance:4@av-offabean-dance` pins a specific clip (naming the slot's default instead, `@rumba`, stores `null` and leaves the slot resolving)
- `Waiting~loop|patience:3,fidget:2.4` sets `loop` via the `~loop` suffix on the name
`encodeRoutine()` and `decodeRoutine()` in the runtime module are the canonical implementations. A malformed link throws with the reason rather than dropping steps quietly, and the studio shows that reason instead of silently loading something else.
diff --git a/docs/cookbook/text-to-3d-cli.md b/docs/cookbook/text-to-3d-cli.md
index 2e83bde3b..63fe123a4 100644
--- a/docs/cookbook/text-to-3d-cli.md
+++ b/docs/cookbook/text-to-3d-cli.md
@@ -17,7 +17,7 @@ saved 1651 KB to a-wooden-treasure-chest-with-iron-bands.glb
![The treasure chest that command produced, rendered by the platform renderer](figure:img:/cookbook/posters/text-to-3d-cli.png)
That is the whole recipe. No API key, no account, no `pip install`. The file is
-about 160 lines of standard-library Python, and roughly a third of it is there
+about 170 lines of standard-library Python, and roughly a third of it is there
to handle the two things that actually go wrong in production.
## What you get
diff --git a/docs/demo-routes.md b/docs/demo-routes.md
index d68701548..9feddb262 100644
--- a/docs/demo-routes.md
+++ b/docs/demo-routes.md
@@ -1,7 +1,7 @@
# Demo routes
The canonical map of every `/demo/*` and `/demos/*` route as of
-2026-08-05. Update whenever a route is added, removed, or moved.
+2026-08-14. Update whenever a route is added, removed, or moved.
The demo namespace splits in two:
@@ -18,8 +18,9 @@ Vite dev-server middleware is in [vite.config.js](../vite.config.js).
Both must be kept in sync.
Every row below was verified against production (`https://three.ws`)
-with `curl -sIL`, not read from config alone.
-See [Verification notes](#verification-notes-2026-08-05) at the end.
+with `curl -sIL`, not read from config alone. Rows added after that
+sweep are marked in the [Verification notes](#verification-notes-2026-08-05)
+at the end, which also records what each later pass actually checked.
## Routes
@@ -34,6 +35,7 @@ See [Verification notes](#verification-notes-2026-08-05) at the end.
| `/demo/avatar-os/combined.html` | `public/demo/avatar-os/combined.html` | Studio + selfie shown side-by-side. |
| `/demos/` | `public/demos/index.html` | Index of all lab demos. |
| `/demos/3d-home` | `public/demos/3d-home.html` | "Give your AI a body" homepage demo. |
+| `/demos/404.html` | `public/demos/404.html` | Lost-avatar 404 page: she falls in from off-screen, lands on the `0` with a dust splash, and runs offscreen before the redirect fires. Carries its own card in the `/demos/` index and doubles as the lab's not-found page; `noindex`. |
| `/demos/audio2face` | `public/demos/audio2face.html` | NVIDIA Audio2Face-3D lipsync: type text, hear it in NVIDIA Magpie's voice, watch the avatar's face animate via ARKit blendshapes. |
| `/demos/avatar-sdk` | `public/demos/avatar-sdk.html` | `@three-ws/avatar` SDK walkthrough. |
| `/demos/bonding-curve` | `public/demos/bonding-curve.html` | Bonding curve simulator. |
@@ -61,9 +63,13 @@ See [Verification notes](#verification-notes-2026-08-05) at the end.
| `/demos/voice-clone` | `public/demos/voice-clone.html` | Voice cloning demo. |
| `/demos/walk-embed-sdk` | `public/demos/walk-embed-sdk.html` | Walk animation embed via SDK. |
-`public/demos/404.html` (`/demos/404`) is the lab's designed not-found
-page, not a content demo — the `/demos/` index also embeds it as a
-hidden empty-state. It is intentionally omitted from the table above.
+`public/demos/404.html` does double duty: it is the lab's designed
+not-found page (the `/demos/` index embeds it as a hidden empty-state)
+and a demo in its own right, carrying its own card in the index grid.
+That card is why it is listed in the table above. The extensionless
+`/demos/404` resolves to the same file through the generic rewrite, but
+nothing links it that way and the page is `noindex`, so prefer the
+explicit `.html` form the index card uses.
The legacy `/lipsync`, `/lipsync/mic`, and `/audio2face` shortcuts in
`vercel.json` still resolve to the `/demos/lipsync-*.html` and
@@ -78,13 +84,16 @@ Single-purpose demos of an avatar reacting to the page. Index hub is
`/demos/` index. Each page lives at `public/demos/agents/<slug>.html` and
is served straight from the filesystem at that `.html` path; the
extensionless `/demos/agents/<slug>` form is a dev-only convenience
-(`vite.config.js` middleware). `vercel.json` has no `/demos/agents/*`
-rewrite, so in production the `.html` form (the one the hub links) is the
-only one that resolves; extensionless agent URLs 404. The routes below
-use the canonical `.html` form.
+(`vite.config.js` middleware). The only `/demos/agents` rule in
+`vercel.json` is the hub rewrite (`/demos/agents` and `/demos/agents/`
+both serve `public/demos/agents/index.html`), so in production the
+`.html` form is the only one that resolves for a sub-demo, which is the
+form the hub links. Extensionless agent slugs 404. The routes below use
+that canonical `.html` form.
| Route | Page file | What it does |
|---|---|---|
+| `/demos/agents` | `public/demos/agents/index.html` | Hub for the lab below, linked from the `/demos/` index. `/demos/agents/` reaches it too, via the sitewide trailing-slash 301. |
| `/demos/agents/auto-rig.html` | `public/demos/agents/auto-rig.html` | Auto-rigging an imported mesh. |
| `/demos/agents/builds-button.html` | `public/demos/agents/builds-button.html` | Agent assembles a CTA button. |
| `/demos/agents/climb-title.html` | `public/demos/agents/climb-title.html` | Agent climbs the page title. |
@@ -170,14 +179,16 @@ Re-verified against production (`https://three.ws`) with `curl -sIL`:
`vercel.json` rewrite) and the `.html` form; in-page links use `.html`.
- All 17 `/demos/agents/<slug>.html` pages return 200. The extensionless
form 404s in production (dev-only, see the agents section above).
-- **Known production gap:** `/demos/agents` and `/demos/agents/` both
- 404 in production. The generic `/demos/([^/.]+)` rewrite maps
- `/demos/agents` to the nonexistent `/demos/agents.html`, and the
- trailing-slash form is 301-normalized to the slashless one before the
- directory index can resolve. Only `/demos/agents/index.html` serves the
- hub, yet the `/demos/` index links `/demos/agents/`. The dev server
- serves both forms (explicit `fileMap` entries in `vite.config.js`), so
- the breakage is invisible in dev.
+- **The `/demos/agents` hub 404 recorded here is fixed** (2026-08-14).
+ It was a real production gap for as long as the hub existed: the
+ generic `/demos/([^/.]+)` rewrite mapped `/demos/agents` to the
+ nonexistent `/demos/agents.html`, and the trailing-slash form is
+ 301-normalized to the slashless one before any directory index can
+ resolve, so the `/demos/` index linked a URL that could not be served.
+ It stayed invisible because the dev server has explicit `fileMap`
+ entries for both forms in `vite.config.js`. `vercel.json` now carries
+ a `/demos/agents` rewrite to `/demos/agents/index.html`, placed above
+ the generic slug rule so it wins. It ships with the next deploy.
- Directory URLs with a trailing slash (`/demos/`, `/demo/`,
`/demo/avatar-os/`) return **301** to the slashless form, which serves
the index with 200.
@@ -187,3 +198,23 @@ Re-verified against production (`https://three.ws`) with `curl -sIL`:
sub-pages.
- `/app-demo` and `/avatar-studio-demo` have been removed (pages and
routes are gone; both 404) and no longer appear in the tables above.
+
+## Later pass (2026-08-14)
+
+Not a production sweep. Every route in the tables was resolved through
+the same route table the server runs (`scripts/lib/page-routing.mjs`
+loads `vercel.json` and `server/index.mjs` serves from it), against the
+files on disk. That catches a route pointing at nothing, which is what
+the agents-hub gap was; it does not catch CDN or revision problems, so
+the 200s above still come from the 2026-08-05 curl sweep.
+
+- All 62 rows resolve to a built file, with two deliberate exceptions:
+ the `/coin` and `/coin/` rows resolve to a 301, and their second
+ column names the redirect target rather than a file.
+- Every `.html` file under `public/demos/` and `public/demo/` now has a
+ row. Two had none: `public/demos/agents/index.html`, covered in prose
+ but absent from the agents table, and `public/demos/404.html`, which
+ was deliberately excluded as "not a content demo" despite carrying its
+ own card in the `/demos/` index grid.
+- The `/demos/<slug>` count is unchanged at 27. The two rows added here
+ are the agents hub and the 404 page, neither of which is a lab slug.
diff --git a/docs/forged.md b/docs/forged.md
index fdb100385..8fc738169 100644
--- a/docs/forged.md
+++ b/docs/forged.md
@@ -46,7 +46,9 @@ runs as a registry entry inside that loop, and on each run it:
5,000 distinct prompts, and a per-hour hash walk selects one. Two runs in the
same hour pick the same prop, which the paid endpoint's idempotency guard
collapses instead of double-charging.
-2. **Pays `POST /api/x402/forge` for a standard-tier generation.** The payment
+2. **Pays `POST /api/x402/forge` for a standard-tier generation** ($0.15 USDC,
+ the standard-tier price in
+ [api/_lib/forge-tiers.js](../api/_lib/forge-tiers.js)). The payment
is a real on-chain USDC transfer from the loop's seeder wallet, settled by
the platform's self-hosted facilitator on Solana mainnet. The same seeder
wallet is the `x402-ring-payer` signer in the
@@ -92,7 +94,7 @@ The response carries the props plus aggregate stats:
"glb_url": "https://pub-2534e921bf9c4314addcd4d8a6e98b7b.r2.dev/forge/nvidia/5ab878cd-513d-4fb6-94b8-d389f7d7e433.glb",
"novelty": 0.41,
"cluster_id": 3,
- "price_usdc": 0.05,
+ "price_usdc": 0.15,
"payer": "7sk…full base58 wallet…",
"payer_short": "7skQxs…9fWn",
"tx_sig": "5Yw…settlement signature…",
@@ -104,7 +106,7 @@ The response carries the props plus aggregate stats:
"total": 120,
"done": 112,
"queued": 5,
- "spent_usdc": 5.6,
+ "spent_usdc": 18,
"categories": { "container": 18, "furniture": 15 },
"latest_ts": "2026-07-29T14:00:11.000Z"
}
diff --git a/docs/ghost-copy.md b/docs/ghost-copy.md
index a305bf3c8..9b99c6211 100644
--- a/docs/ghost-copy.md
+++ b/docs/ghost-copy.md
@@ -107,13 +107,16 @@ The replay response:
{
"paper": true,
"custody": "none",
- "leader": { "agent_id": "…", "name": "…", "settled": 41, "profile_url": "/trader/…" },
+ "network": "mainnet",
+ "leader": { "agent_id": "…", "name": "…", "avatar": "…", "settled": 41,
+ "profile_url": "/trader/…", "first_trade_at": "…", "last_close_at": "…" },
"window": "7d",
+ "window_start": "2026-07-26T…",
"budget_sol": 1,
"sizing": { "sizing_rule": "fixed", "fixed_sol": 0.1, "per_trade_cap_sol": 0.25, "…": "…" },
- "fills": [ { "mint": "…", "symbol": "TICKER", "order_sol": 0.1, "pnl_sol": 0.062,
+ "fills": [ { "mint": "…", "symbol": "TICKER", "name": "…", "order_sol": 0.1, "pnl_sol": 0.062,
"pnl_pct": 62.4, "multiple": 1.62, "hold_seconds": 812,
- "buy_sig": "…", "sell_sig": "…" } ],
+ "buy_sig": "…", "sell_sig": "…", "…": "…" } ],
"skipped": [ { "mint": "…", "reason": "ghost_cash_locked", "detail": "Your ghost wallet had …" } ],
"still_open": [ { "mint": "…", "order_sol": 0.1, "mark_pct": 12.4,
"unrealized_sol": 0.0124, "marked": "leader_last_quote" } ],
@@ -122,7 +125,9 @@ The replay response:
"start_sol": 1, "end_sol": 1.34, "realized_pnl_sol": 0.34, "realized_pnl_pct": 34,
"unrealized_pnl_sol": 0.012, "copied": 9, "wins": 5, "losses": 4, "win_rate_pct": 55.56,
"max_drawdown_pct": 8.2, "avg_hold_seconds": 940,
- "leader_trades": 14, "skipped_count": 5, "still_open_count": 1
+ "leader_trades": 14, "skipped_count": 5, "still_open_count": 1,
+ "deployed_sol": 0.9, "idle_sol": 0.1, "mark_to_market_sol": 1.352,
+ "best": { "…": "…" }, "worst": { "…": "…" }
},
"honesty": [ "Paper only. No wallet was connected…", "…" ]
}
diff --git a/docs/kol-tracker.md b/docs/kol-tracker.md
index 4293fa459..2da91e146 100644
--- a/docs/kol-tracker.md
+++ b/docs/kol-tracker.md
@@ -4,7 +4,7 @@ The KOL Tracker ranks Solana KOLs by realized P&L computed from their own wallet
Page: [/tracker](https://three.ws/tracker)
-API: `/api/kol/tracker` (the board), with sibling reads `/api/kol/leaderboard`, `/api/kol/trades`, and `/api/kol/wallets` from the same dispatcher.
+API: `/api/kol/tracker` (the board), with sibling reads `/api/kol/leaderboard`, `/api/kol/wallets`, and `/api/kol/trades`.
## Why it exists
@@ -46,7 +46,7 @@ Each integration degrades honestly: a missing key or an upstream outage yields a
## API
-All reads are public, CORS-open, and IP rate-limited (600 requests per minute per IP; over that returns 429). The dispatcher is `api/kol/[action].js`.
+All reads are public, CORS-open, and IP rate-limited (600 requests per minute per IP; over that returns 429). `tracker`, `leaderboard`, `wallets`, and the admin import share one dispatcher, `api/kol/[action].js`; `/api/kol/trades` is served by its own exact file, `api/kol/trades.js`, which wins on filesystem precedence.
### `GET /api/kol/tracker`
diff --git a/docs/media-api.md b/docs/media-api.md
index 4d25d047a..ef502f234 100644
--- a/docs/media-api.md
+++ b/docs/media-api.md
@@ -501,7 +501,7 @@ Pagination is opt-in and backward compatible:
"license": "Mixamo"
}
],
- "total": 106,
+ "total": 107,
"generated_at": "2026-07-01T00:00:00.000Z"
}
```
@@ -552,7 +552,7 @@ for the full parameter table.
"source": "polyhaven"
}
],
- "total": 0,
+ "total": 511,
"generated_at": "2026-07-01T00:00:00.000Z"
}
```
diff --git a/docs/notifications.md b/docs/notifications.md
index b8abfb6fa..3fbdaf989 100644
--- a/docs/notifications.md
+++ b/docs/notifications.md
@@ -66,7 +66,7 @@ mapping (`TYPE_CATEGORY`) lives in
|---|---|---|
| Sales & earnings | `sales` | `skill_purchased`, `asset_purchased`, `sale`, `payment-earned`, `payment_received`, `referral_earned`, `referral_signup`, `referral_reward`, `pump_launch_filled`, `royalty_paid` |
| Purchases | `purchases` | `skill_purchase_confirmed`, `asset_purchase_confirmed`, `skill_gift_received`, `skill_gift_sent` |
-| Social | `social` | `remix`, `reply`, `embed`, `mention`, `fork`, `follow`, `dm_received`, `agent_review`, `quest_complete` |
+| Social | `social` | `remix`, `reply`, `comment`, `embed`, `mention`, `fork`, `follow`, `dm_received`, `agent_review`, `quest_complete` |
| In person | `irl` | `irl_interaction`, `irl_reply` |
| Market alerts | `alerts` | `pump_alert` |
| Creations | `creations` | `forge_complete`, `forge_failed` |
diff --git a/docs/nvidia-nemotron-spotlight.md b/docs/nvidia-nemotron-spotlight.md
index 76f79c972..c8d866f0a 100644
--- a/docs/nvidia-nemotron-spotlight.md
+++ b/docs/nvidia-nemotron-spotlight.md
@@ -53,10 +53,16 @@ usable reference: it should show ONE clear physical subject (an object, characte
creature, or person) that could plausibly be turned into a 3D model.
Reply ONLY with compact JSON, no prose, in exactly this shape:
-{"usable":true|false,"subject":"<2-5 word description>","issue":"none"|"text_screenshot"
-|"multiple_subjects"|"no_clear_subject"|"too_dark_or_blurry"|"abstract_or_diagram"}
-
-When in doubt, mark usable=true. A borderline photo still reconstructs.
+{"usable":true|false,"subject":"<2-5 word description of the main subject, or empty>",
+"issue":"none"|"text_screenshot"|"multiple_subjects"|"no_clear_subject"
+|"too_dark_or_blurry"|"abstract_or_diagram"}
+
+Mark usable=false ONLY when the image is genuinely unsuitable: a screenshot of
+text/UI (text_screenshot), a busy scene with no single dominant subject
+(multiple_subjects), no recognizable object at all (no_clear_subject), too
+dark/blurry to make out (too_dark_or_blurry), or an abstract pattern/chart/diagram
+(abstract_or_diagram). When in doubt, mark usable=true. A borderline photo still
+reconstructs.
```
The model answering that question is **`nvidia/nemotron-nano-12b-v2-vl`**. It returns in **1-2 seconds**. It costs us nothing. And it turned "your 3D model is garbage" into this:
@@ -73,14 +79,16 @@ For a small reference image, `nemotron-nano-12b-v2-vl` consumes roughly **281 pr
Nemotron Nano won on the axis that actually mattered. The 12B VL model is not competing with a frontier model on essay quality. It is being asked whether there is one clear object in a photograph, and that is a question it answers reliably, in 1-2 seconds, for nothing.
-Our vision chain, in order:
+Our free vision lanes, in order (from [`api/_lib/vision.js`](../api/_lib/vision.js), which
+the validator above calls):
```js
const NVIDIA_VISION_MODELS = [
'nvidia/nemotron-nano-12b-v2-vl', // leads: smallest image token footprint
'meta/llama-3.2-11b-vision-instruct', // different family = independent failure modes
];
-// paid vision-capable backstop appended last, and only if its key is set
+// then our own GCP-credits Vertex Gemini anchor, and a paid vision-capable
+// backstop last, only if its key is set
```
Note the comment on line two, because it is the load-bearing design decision in this file. The second lane is a **different model family**, not a bigger checkpoint of the first. A retry against the same family is a re-roll, not a fallback. If Nemotron Nano's failure mode is triggered by an input, Nemotron Super's probably is too. Llama 3.2 Vision fails differently. That is the whole point of having it.
diff --git a/docs/okx-marketplace.md b/docs/okx-marketplace.md
index 5a4ab4b5d..419bcf4f5 100644
--- a/docs/okx-marketplace.md
+++ b/docs/okx-marketplace.md
@@ -3,7 +3,10 @@
three.ws sells 3D services to other AI agents on [OKX.AI](https://web3.okx.com), OKX's
on-chain agent marketplace. Buyers are agents: they discover a service, receive an HTTP 402
payment challenge, pay in stablecoins on X Layer (or USDC on our other x402 rails), and get
-the artifact. Our marketplace listing is agent **#2632 "three.ws 3D Studio"**.
+the artifact. Our marketplace entry is agent **#2632 "three.ws 3D Studio"**. That listing is
+not live yet: it was rejected in review and a resubmission is pending, so the services below
+are reachable directly at the endpoints documented here rather than through the marketplace.
+Current listing state is tracked in [`prompts/okx-ai/PROGRESS.md`](../prompts/okx-ai/PROGRESS.md).
> Related: [REST API Reference](./api-reference.md) · [MCP servers](./mcp.md) ·
> [x402 payments](./x402.md) · seller-side protocol spec: [`specs/okx-agent-payments.md`](../specs/okx-agent-payments.md)
@@ -75,7 +78,7 @@ Arguments:
| `style_hints` | no | palette / materials / era / mood |
| `reference_image_url` | no | public image guiding the look, validated **before** any charge |
-Success returns `{ job_id, status: "queued", poll_tool: "identity_status", eta_seconds }`.
+Success returns `{ job_id, status: "running", stage: "generate", poll_tool: "identity_status", eta_seconds }`.
### 2 · Poll (free)
diff --git a/docs/partners.md b/docs/partners.md
index c46450226..527ec8c6b 100644
--- a/docs/partners.md
+++ b/docs/partners.md
@@ -17,7 +17,7 @@ Page source: [`pages/partners.html`](../pages/partners.html). Partnership enquir
| Amazon Web Services | Cloud | `/aws` | [AWS Marketplace](./aws-marketplace.md), [listing kit](./aws-marketplace-listing-kit.md), [partner spotlight](./aws-partner-spotlight.md) |
| Google Cloud | Cloud | `/blog/three-ws-google-cloud-partnership` | [Production runbook](./ops/gcp-production.md), [credits plan](./ops/gcp-credits-plan.md), [model workers](./ops/gcp-model-workers.md) |
| Alibaba Cloud | Cloud | `/blog/three-ws-alibaba-cloud-partnership` | [Listings and distribution](./listings.md), [the model router](./brain.md) |
-| NVIDIA | Inception | `/docs/nvidia-inception` | [NVIDIA Inception](./nvidia-inception.md), [NVIDIA models](./nvidia-models.md), [Nemotron spotlight](./nvidia-nemotron-spotlight.md) |
+| NVIDIA | Inception | `/nvidia` | [NVIDIA Inception](./nvidia-inception.md), [NVIDIA models](./nvidia-models.md), [Nemotron spotlight](./nvidia-nemotron-spotlight.md) |
| HackerNoon | Media | `/blog/three-ws-hackernoon-partnership` | [Syndication](./syndication.md), [Listings and distribution](./listings.md) |
| Quicknode | Infrastructure | `/blog/three-ws-quicknode-startup-program` | [Solana integration](./solana.md), [Listings and distribution](./listings.md) |
diff --git a/docs/portfolio.md b/docs/portfolio.md
index 9a628afd8..1644f48d8 100644
--- a/docs/portfolio.md
+++ b/docs/portfolio.md
@@ -80,10 +80,10 @@ The same contract as the rest of the platform's money surfaces:
"other": { "usd": 234.56, "pct": 19, "count": 5 }
},
"topAssets": [
- { "symbol": "SOL", "usd": 800, "pct": 64.8, "slot": 1, "logo": "…", "id": "native" }
+ { "symbol": "SOL", "name": "Solana", "usd": 800, "pct": 64.8, "slot": 1, "logo": "…", "id": "native" }
],
"rows": [
- { "id": "…", "symbol": "SOL", "kind": "native", "class": "major", "amount": 4.2, "price": 190.4, "usd": 800, "change24h": 2.1, "sharePct": 64.8 }
+ { "id": "native", "symbol": "SOL", "name": "Solana", "kind": "native", "class": "major", "amount": 4.2, "price": 190.4, "usd": 800, "change24h": 2.1, "logo": "…", "sharePct": 64.8 }
],
"tokenCount": 8,
"truncated": false,
diff --git a/docs/selfie-to-avatar.md b/docs/selfie-to-avatar.md
index 4a384e85f..869cdf92b 100644
--- a/docs/selfie-to-avatar.md
+++ b/docs/selfie-to-avatar.md
@@ -25,7 +25,7 @@ Two of those numbers are measured rather than reasoned about. The blur floor (`B
6. **Reconstruct, server-side.** The reconstruct handler (`api/avatars/_actions.js`) selects a provider through [`api/_lib/regen-provider.js`](../api/_lib/regen-provider.js). The production lane is **GCP**: it POSTs `{ images, body_type }` to the Cloud Run service behind `GCP_RECONSTRUCTION_URL`, today the [`workers/avatar-reconstruction/`](../workers/avatar-reconstruction/) FastAPI app deployed as `avatar-reconstruction`. (The multi-backend [`workers/avatar-pipeline-controller/`](../workers/avatar-pipeline-controller/) speaks the identical contract and can take over that env var, but is not deployed today.) That pipeline (`face_texture_transfer_v2`) uses **MediaPipe FaceLandmarker** (468 landmarks, Apache-2.0) to transfer your face texture and geometry onto a fixed-topology, pre-rigged Wolf3D/Ready-Player-Me-style humanoid template carrying a Mixamo-compatible skeleton, 52 ARKit blendshapes, and 15 visemes. Full engine detail is in [avatar-reconstruction.md](./avatar-reconstruction.md).
7. **Auto-rig fallback.** The template is born rigged. If a generic mesh comes back unrigged and a rig model is configured, `reconstruct-finalize.js` chains a `rerig` job (the `model-rig` worker, `workers/rig`) and moves the job through a `rigging` status before surfacing the avatar. If rigging is impossible it delivers the static mesh tagged `unrigged` rather than nothing.
8. **Poll and finish.** The client polls `GET /api/avatars/regenerate-status?jobId=...` every ~3 seconds (backing off on errors). On `status: 'done'` it fetches the finished avatar from `GET /api/avatars/:id` and shows it in a `<model-viewer>`. The poll uses the adapter for the provider recorded on the job row, not whichever provider currently leads the precedence order, because the stored job id is provider-shaped and only that backend can read it.
-9. **Register and mint.** Materializing the avatar also registers it in the Forge store (`forge_creations`, `path: 'reconstruct'`) so galleries, share and embed pages, and provenance treat a capture like any other creation, and fires a best-effort draft agent identity mint. Both are best-effort by design: the avatar is already delivered, so neither can fail the job.
+9. **Register and mint.** Materializing the avatar also registers it in the Forge store (`forge_creations`, `path: 'reconstruct'`) so galleries, share and embed pages, and provenance treat a capture like any other creation, and fires a best-effort [draft agent identity mint](./draft-agent-mint.md): a Metaplex Core asset on Solana, with an ERC-8004 registration behind its own flag. Both are best-effort by design: the avatar is already delivered, so neither can fail the job.
10. **Backstop for a closed tab.** The browser poll is what drives finalization, so closing the tab mid-build would otherwise strand the job even though the worker finished. [`api/cron/reconstruct-sweep.js`](../api/cron/reconstruct-sweep.js) runs the same finalize stages server-side every few minutes, so an abandoned capture still lands in your library. It rescues jobs up to 30 days old, and fails anything older with an honest reason rather than leaving it open forever.
### The text-to-avatar variant
@@ -85,6 +85,7 @@ Request body: `name` (1-120, required), `photos` (1-6 image URLs or data URIs) *
- [Avatar reconstruction pipeline](./avatar-reconstruction.md): the full engine: MediaPipe face-texture transfer, geometry morph, the fixed pre-rigged head template, and the Cloud Run worker.
- [Avatar pipeline](./avatar-pipeline.md): the distinct Forge text-to-3D generation lane.
+- [Draft agent mint](./draft-agent-mint.md): the on-chain identity every finished reconstruction gets, the network flags, and the devnet proof script.
- [Trait-based avatar builder](./character-studio.md): customize and dress the avatar after you scan it.
- [Talking Avatar Video](./talking-avatar-video.md) and [Lipsync](./lipsync.md): drive the finished avatar's mouth.
- Pages: [/create/selfie](https://three.ws/create/selfie), [/create](https://three.ws/create), [/features/scan](https://three.ws/features/scan).
diff --git a/docs/sniper-autonomy.md b/docs/sniper-autonomy.md
index 719353395..abbecbf35 100644
--- a/docs/sniper-autonomy.md
+++ b/docs/sniper-autonomy.md
@@ -58,19 +58,21 @@ so anything that does not pass a tier is unchanged.
### 2. Which fields the optimizer may touch at all
Every tier can tune stops, take-profit, hold time, the quality and oracle floors,
-and position size. Reaching `trusted` unlocks four more:
+and position size. Reaching `trusted` unlocks five more (`TIER_UNLOCKS` in
+[`api/_lib/sniper-autonomy.js`](../api/_lib/sniper-autonomy.js)):
- `llm_min_confidence`: a profitable judge earns a lower bar, so it acts on more
of what it sees instead of passing on launches it would have won.
- `min_market_cap_usd` / `max_market_cap_usd`: a wider hunting ground, widened
outward only. An arm with no band set is already unrestricted and nothing here
will restrict it.
-- `initials_out_multiple`, the proactive
+- `initials_out_multiple` / `moonbag_min_pct`, the proactive
[take-initials ladder](./trading-experiment.md): recover the stake as soon as
- the position hits 2x rather than waiting for an exit to trigger. Turned on only
- for an arm whose winners demonstrably run past its average exit. Note this is
- only the *early* recovery: never fully selling a winner is fleet-wide policy
- that applies at every tier, ladder or no ladder.
+ the position hits 2x rather than waiting for an exit to trigger, and size the
+ moon bag that keeps riding. Turned on only for an arm whose winners
+ demonstrably run past its average exit. Note this is only the *early*
+ recovery: never fully selling a winner is fleet-wide policy that applies at
+ every tier, ladder or no ladder.
`autonomous` adds `max_creator_launches`, so a proven arm can test whether a
prolific creator is really a bad one.
diff --git a/docs/start-here.md b/docs/start-here.md
index 5692590cf..60df591cb 100644
--- a/docs/start-here.md
+++ b/docs/start-here.md
@@ -94,6 +94,7 @@ Curious what happens between the prompt and the model? [How the Forge works](./h
- **Curious how the trading fleet improves itself?** → [Earned autonomy](./sniper-autonomy.md), where each arm's own realized record decides how much rope it gets: wider tuning bounds, unlocked settings, a bigger share of the fleet budget, and a deeper evidence pack in front of its LLM judge, all re-earned continuously and none of it able to touch a safety rail
- **Trusting an agent with money?** → [Custody you can verify](./custody.md) — spend limits, freeze, Merkle proof-of-custody, and social recovery; [claim your wallet](./trader-card.md) turns any pump.fun track record into a public, provable Trader Card. Every real-funds feature sits behind the [risk acknowledgment](./risk-acknowledgment.md) — read the [Risk Disclosure](https://three.ws/legal/risk) before committing anything
- **Need a track record another app can verify?** → [Trader Passport](./trader-passport.md): the daily on-chain attestation of a trader's score, served as a public credential any terminal can fetch, pin, and re-check against Solana without trusting us. Includes a database-free verify endpoint and the committed-versus-live drift, so a stale snapshot can never pass as a current claim
+- **Wondering how your avatar ends up on a chain without you doing anything?** → [Draft agent mint](./draft-agent-mint.md): every finished reconstruction is written to durable storage and minted as an inactive on-chain identity in the background, a Metaplex Core asset on Solana with an ERC-8004 registration behind its own flag. No wallet prompt, no fee, nothing to remember. Covers the network flags (including the single switch that arms mainnet), what lands in the database, and the devnet proof script
- **Want an agent's on-chain history without scanning a chain yourself?** → [The agent index](./agent-index.md): the platform's own cross-chain index of agent registrations, token launches, reputation attestations, transfers and delegations, Solana first and EVM alongside it. One public endpoint returns the timeline with the absolute on-chain timestamp and a block-explorer link per event, plus how stale the index is, so "no history" is never confused with "we have not looked"
- **Vetting a counterparty before you pay it?** → [Trust primitives](./trust-primitives.md) — the cross-chain Agent Reputation endpoint scores ANY wallet, mint, or agent id (Solana or EVM) 0–100 from real on-chain evidence, in one paid call, before your agent transacts
- **Wondering where the platform's money goes?** → [The autonomous economy](./autonomous-economy.md): how the on-chain treasury funds itself, the funding-root → engines → sweepback loop, the locks that keep every dollar inside platform-owned wallets, and the two leak scanners that audit every wallet on-chain every few minutes (as of 2026-07-12: 44,122 transactions scanned, zero leaks; the live totals stay re-derivable from each scanner's own cursor)
diff --git a/docs/trader-passport.md b/docs/trader-passport.md
index 6141d0f49..889cfae0c 100644
--- a/docs/trader-passport.md
+++ b/docs/trader-passport.md
@@ -50,12 +50,17 @@ The response:
"kind": "threews.tradescore.v1",
"issuer": {
"name": "three.ws",
+ "kind": "threews.tradescore.v1",
"attester": "<ISSUER_PUBKEY>",
+ "attester_url": "https://solscan.io/account/<ISSUER_PUBKEY>",
"memo_program": "MemoSq4gqABAXKb96qnH8TysNcWxMyWCqXgDLGmfcHr",
- "cadence": "daily"
+ "cadence": "daily",
+ "note": "Scores are re-derived from on-chain fills and committed …"
},
"status": "attested",
+ "unattested_reason": null,
"credential": {
+ "kind": "threews.tradescore.v1",
"signature": "<TX_SIGNATURE>",
"slot": 300123456,
"block_time": "2026-06-14T09:00:00.000Z",
@@ -63,16 +68,22 @@ The response:
"window": "all",
"attester": "<ISSUER_PUBKEY>",
"subject": "<BASE58_WALLET>",
+ "agent_id": "<uuid>",
+ "revoked": false,
"well_formed": true,
"schema_problems": [],
- "snapshot": { "score": 78, "closed": 41, "win_rate": 0.61, "realized_pnl_sol": 12.5, "self_dealing_excluded": 2 },
+ "snapshot": { "score": 78, "closed": 41, "win_rate": 0.61, "realized_pnl_sol": 12.5, "self_dealing_excluded": 2, "…": "…" },
"explorer_url": "https://solscan.io/tx/<TX_SIGNATURE>"
},
"credential_age_days": 1,
"history": [{ "day": "2026-06-14", "score": 78, "signature": "<TX_SIGNATURE>" }],
"live": { "score": 81, "closed": 44, "win_rate": 0.58, "realized_pnl_sol": 10.25 },
"drift": { "moved": true, "fields": { "score": { "attested": 78, "live": 81, "delta": 3 } } },
- "verify": { "url": "https://three.ws/api/trader-passport/verify?signature=<TX_SIGNATURE>&network=mainnet" }
+ "verify": {
+ "url": "https://three.ws/api/trader-passport/verify?signature=<TX_SIGNATURE>&network=mainnet",
+ "how": "Fetch the attestation transaction from any Solana RPC, read the SPL-Memo payload, …"
+ },
+ "generated_at": "2026-06-15T09:02:11.418Z"
}
```
@@ -93,8 +104,14 @@ This endpoint reads **no three.ws database**. It fetches the transaction from a
5. the signer matches the `attester` you pinned, and the subject matches the `wallet` you passed.
```json
-{ "valid": true, "found": true, "attester": "<ISSUER_PUBKEY>", "subject": "<BASE58_WALLET>",
- "slot": 300123456, "block_time": "2026-06-14T09:00:00.000Z", "payload": { "score": 78 }, "reasons": [] }
+{ "valid": true, "found": true, "network": "mainnet", "signature": "<TX_SIGNATURE>",
+ "attester": "<ISSUER_PUBKEY>", "subject": "<BASE58_WALLET>",
+ "slot": 300123456, "block_time": "2026-06-14T09:00:00.000Z",
+ "payload": { "score": 78, "…": "…" }, "reasons": [],
+ "explorer_url": "https://solscan.io/tx/<TX_SIGNATURE>",
+ "kind": "threews.tradescore.v1",
+ "memo_program": "MemoSq4gqABAXKb96qnH8TysNcWxMyWCqXgDLGmfcHr",
+ "checked_at": "2026-06-15T09:02:11.418Z" }
```
Every failed check appears in `reasons[]`, so `valid: false` always says why. `wallet` and `attester` are optional; omit them and you get the on-chain facts without the equality checks. An RPC that is unreachable returns a `502 rpc_failed` error rather than a false `valid: false`: an unanswered question is not a negative verdict.
diff --git a/docs/tutorials/agent-spending-envelope.md b/docs/tutorials/agent-spending-envelope.md
index c6af158f8..83aadba06 100644
--- a/docs/tutorials/agent-spending-envelope.md
+++ b/docs/tutorials/agent-spending-envelope.md
@@ -141,6 +141,12 @@ curl -s $BASE/api/x402/rate-limit-probe | jq '{
"amount": "1000",
"asset": "<the Solana USDC mint>",
"feePayer": true
+ },
+ {
+ "network": "solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp",
+ "amount": "10000000",
+ "asset": "FeMbDoX7R1Psc4GEcvJdsbNbZA3bfztcyDCatJVJpump",
+ "feePayer": true
}
],
"what": "Rate-Limit Capacity Probe - pay $0.001 USDC to learn how many more calls the x402 autonomous loop can make to a target endpoint..."
@@ -149,8 +155,9 @@ curl -s $BASE/api/x402/rate-limit-probe | jq '{
Three things to read off that:
-- **`amount` is in atomic units**, and USDC has 6 decimals, so `1000` is $0.001. A session's `max_per_tx_usd` is expressed in dollars, so the cap you want here is anything above `0.001`.
-- **The accept must be on a `solana...` network** and must carry `extra.feePayer`. `POST /api/pay/execute` settles on Solana only. An endpoint that offers no Solana option fails with `no_solana_accept` and your budget is never touched.
+- **`amount` is in atomic units**, and USDC has 6 decimals, so `1000` is $0.001. A session's `max_per_tx_usd` is expressed in dollars, so the cap you want here is anything above `0.001`. Do not size the cap from the second entry: that is the same endpoint priced in $THREE, and a session never pays it.
+- **The accept must be on a `solana...` network** and must carry `extra.feePayer`. `POST /api/pay/execute` settles on Solana only. When an endpoint advertises more than one Solana accept, execute picks the one whose `asset` is the Solana USDC mint, so the USDC price is the one your budget is charged. An endpoint that offers no Solana option fails with `no_solana_accept` and your budget is never touched.
+- **A Solana accept can disappear for a while.** An endpoint drops its Solana entries whenever the platform's fee-payer sponsor cannot co-sign (for example while its wallet sits under its SOL settle floor), leaving only Base accepts. That is a transient operational state, not a change in the endpoint's price: retry the probe rather than re-plumbing the tutorial onto another chain.
- **The service fee payer covers the Solana transaction fee.** Your budget is charged the payment amount and nothing else.
This same endpoint answers a genuinely useful question (how much x402 capacity is left before a rate limit bites), which is why it makes a good first purchase instead of a throwaway. Browse more at [/bazaar](/bazaar), or read the [x402 protocol doc](/docs/x402.md) for how the challenge is constructed.
diff --git a/docs/tutorials/build-a-scene.md b/docs/tutorials/build-a-scene.md
index 011bf49b6..1b368af5c 100644
--- a/docs/tutorials/build-a-scene.md
+++ b/docs/tutorials/build-a-scene.md
@@ -159,7 +159,7 @@ When the scene looks right, open **File → Export** and choose a format. For a
The Export submenu also offers single-object/whole-scene formats for specific pipelines: **OBJ** and **DRC** (these export the *selected* object — select a mesh first), and **PLY**, **STL** (ASCII and binary variants), and **USDZ** for the whole scene. Reach for these only when a downstream tool requires them; for everything on three.ws, **GLB** is the right choice.
-The file downloads through your browser. That's your finished scene — re-import it anywhere, [view it in AR](/docs/tutorials/view-in-ar), or hand it to another agent.
+The file downloads through your browser. That's your finished scene: re-import it anywhere, [view it in AR](/tutorials/view-in-ar), or hand it to another agent.
> **Save the editable project too.** Export bakes the scene into a delivery format. To keep working on it later with full editor state, use **File → Save**, which writes a `project.json` you can reopen with **File → Open**. The autosaved browser copy persists between visits, but `project.json` is your portable, backup-able source of truth.
@@ -205,8 +205,8 @@ The leverage of [Scene Studio](/scene) is that it's a *real* editor producing a
## See also
-- [Generate a 3D model from text](/docs/tutorials/text-to-3d) — make GLBs to import here.
-- [Image to 3D](/docs/tutorials/image-to-3d) — turn a photo into a model.
-- [Upload a custom GLB](/docs/tutorials/upload-custom-glb) — bring your own assets in.
-- [View in AR](/docs/tutorials/view-in-ar) — open your exported scene on a phone.
-- [Swap an avatar in the studio](/docs/tutorials/swap-avatar-in-studio) — work with avatars across the editors.
+- [Generate a 3D model from text](/tutorials/text-to-3d): make GLBs to import here.
+- [Image to 3D](/tutorials/image-to-3d): turn a photo into a model.
+- [Upload a custom GLB](/tutorials/upload-custom-glb): bring your own assets in.
+- [View in AR](/tutorials/view-in-ar): open your exported scene on a phone.
+- [Swap an avatar in the studio](/tutorials/swap-avatar-in-studio): work with avatars across the editors.
diff --git a/docs/tutorials/character-library-to-embed.md b/docs/tutorials/character-library-to-embed.md
index fe21cd4ce..038343516 100644
--- a/docs/tutorials/character-library-to-embed.md
+++ b/docs/tutorials/character-library-to-embed.md
@@ -205,7 +205,7 @@ One script tag, one element. Create `index.html`:
<script
type="module"
src="https://three.ws/agent-3d/1.5.2/agent-3d.js"
- integrity="sha384-qsd2ZiZi75y0AgsaG8CdbA7MwEkqQI8/s04Eiw8ti0NFNbSD+0CpcU331csU01fn"
+ integrity="sha384-qCG5gH4q2+k2Gsf98zs9RFXyN8iezoklCWt63pA2Xk2YF7Onae4rfUwu+oZSqRzN"
crossorigin="anonymous"
></script>
diff --git a/docs/tutorials/render-avatar-images.md b/docs/tutorials/render-avatar-images.md
index 210998de7..cd139b0b4 100644
--- a/docs/tutorials/render-avatar-images.md
+++ b/docs/tutorials/render-avatar-images.md
@@ -35,9 +35,11 @@ curl -s https://three.ws/api/avatar/render | jq 'keys'
```
```json
-["endpoint", "description", "parameters", "poses", "scenes", "example"]
+["description", "endpoint", "example", "parameters", "poses", "scenes"]
```
+(`jq 'keys'` sorts alphabetically. Use `jq 'keys_unsorted'` if you want the order the endpoint actually emits.)
+
That response is the machine-readable source of truth, which matters more than it sounds: it means a code generator, an LLM tool definition, or a build script never has to hardcode a list that can drift. Look at the camera presets:
```bash
@@ -90,11 +92,16 @@ curl -s https://three.ws/api/avatars/featured \
```
81a076b6-55ff-49a2-b007-1d88e7dce2aa ansem-with-animation
bea9f0b9-3442-4544-84df-912989e86211 Boss Vernington
+4f00c120-b3af-4f2b-9e0c-f34a71df8d37 Boss Vernington
13f259c7-7024-4d68-b1f0-dbbf52c06209 Michelle
-9c192f82-059f-44b3-b21d-100583c743fd LittlestTokyo
a4bad2f5-8a07-43cf-82e5-b6ba1314441e Selfie Girl
+9c192f82-059f-44b3-b21d-100583c743fd LittlestTokyo
+3cbc5b19-bb36-4658-a569-d0c2f7cb6dab Horse
+72bc0b6d-7888-4923-a532-16b0b4d7b58b CZ
```
+That list is live and its order changes as avatars are featured, so read it rather than pasting it. Two names can repeat with different IDs, because a name is a label and the UUID is the identity.
+
Two of those get used throughout this tutorial, and the difference between them is the single most important practical fact on this page:
| Avatar | ID | Rig | Morph targets |
diff --git a/docs/world-lines.md b/docs/world-lines.md
index 6ae5de354..2959622d5 100644
--- a/docs/world-lines.md
+++ b/docs/world-lines.md
@@ -129,7 +129,7 @@ Server-side, `complete` runs eight steps in this order, and each one can end the
| Farm a quest for many collectibles | `(world_line_id, completer_hash)` is unique. One proof per visitor per quest, whether identified by account or device token. |
| Lift someone else's proof and claim it | The signed bytes include the completer's salted hash, so a proof only ever attests to one visitor. |
| Mint proofs without the agent | The signing key is the agent's custodial wallet, and the verifier checks against the pubkey captured at creation. |
-| Script the endpoints | Per-IP limits: 15 creates per 10 min, 30 challenges per 5 min, 20 completes per 5 min, 240 public reads per min, 30 fix mints per min. |
+| Script the endpoints | Per-IP limits: 15 creates per 10 min, 30 challenges per 5 min, 20 completes per 5 min, 240 public reads per min, 60 nearby lookups per min, 30 fix mints per min. |
---