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.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
content(x): release seven reviewed stories by policy and record every review of the run
nirholasb72e29b77546
diff --git a/data/x-content/queue.json b/data/x-content/queue.jsonindex 7d3216317..7e444b101 100644--- a/data/x-content/queue.json+++ b/data/x-content/queue.json@@ -2824,12 +2824,12 @@},{"id": "splat-viewer",- "status": "review",+ "status": "approved","kind": "post","tier": 2,"lane": "community","pattern": "mechanism",- "notBefore": "2026-09-30T00:00:00Z",+ "notBefore": "2026-10-01T08:24:29.235Z","covers": ["/splat","/docs/splat"@@ -3094,16 +3094,17 @@]}],- "mentions": {}+ "mentions": {},+ "approvedBy": "policy"},{"id": "agent-monitor",- "status": "review",+ "status": "approved","kind": "post","tier": 2,"lane": "developer","pattern": "number",- "notBefore": "2026-09-30T00:00:00Z",+ "notBefore": "2026-10-01T08:27:22.393Z","covers": ["/monitor"],@@ -3370,7 +3371,8 @@]}],- "mentions": {}+ "mentions": {},+ "approvedBy": "policy"},{"id": "agent-spotlight",@@ -3673,12 +3675,12 @@},{"id": "avatar-cli",- "status": "review",+ "status": "approved","kind": "post","tier": 3,"lane": "developer","pattern": "mechanism",- "notBefore": "2026-09-30T00:00:00Z",+ "notBefore": "2026-10-01T08:40:54.050Z","covers": ["/avatar-cli","@three-ws/avatar-cli"@@ -3982,16 +3984,17 @@]}],- "mentions": {}+ "mentions": {},+ "approvedBy": "policy"},{"id": "choreograph",- "status": "review",+ "status": "approved","kind": "post","tier": 2,"lane": "community","pattern": "walkthrough",- "notBefore": "2026-09-30T00:00:00Z",+ "notBefore": "2026-10-01T08:45:48.935Z","covers": ["/choreograph"],@@ -4182,16 +4185,17 @@]}],- "mentions": {}+ "mentions": {},+ "approvedBy": "policy"},{"id": "docs-world",- "status": "review",+ "status": "approved","kind": "post","tier": 2,"lane": "community","pattern": "clip",- "notBefore": "2026-09-30T00:00:00Z",+ "notBefore": "2026-10-01T08:50:23.737Z","covers": ["/docs/world"],@@ -4465,7 +4469,8 @@]}],- "mentions": {}+ "mentions": {},+ "approvedBy": "policy"},{"id": "drive-copilot",@@ -6302,12 +6307,12 @@},{"id": "model-inspector",- "status": "draft",+ "status": "approved","kind": "post","tier": 2,"lane": "developer","pattern": "walkthrough",- "notBefore": "2026-09-30T00:00:00Z",+ "notBefore": "2026-10-01T09:09:34.408Z","covers": ["/inspect"],@@ -6618,16 +6623,17 @@]}],- "mentions": {}+ "mentions": {},+ "approvedBy": "policy"},{"id": "prompt-library",- "status": "draft",+ "status": "approved","kind": "post","tier": 2,"lane": "community","pattern": "walkthrough",- "notBefore": "2026-09-30T00:00:00Z",+ "notBefore": "2026-10-01T09:11:43.196Z","covers": ["/prompts","/docs/prompts",@@ -6910,7 +6916,8 @@]}],- "mentions": {}+ "mentions": {},+ "approvedBy": "policy"}]}diff --git a/data/x-content/reviews/agent-monitor.json b/data/x-content/reviews/agent-monitor.jsonindex 0f21121f5..831fa7f4d 100644--- a/data/x-content/reviews/agent-monitor.json+++ b/data/x-content/reviews/agent-monitor.json@@ -1,27 +1,23 @@{"id": "agent-monitor",- "contentHash": "e02df0329e0a2e1466cd00aeeea084afdb7d16d37e3476e35eb0028fdd42be68",- "reviewedAt": "2026-09-30T08:45:45.698Z",- "passed": false,- "blockers": [- "probe:scenario the filmed scenario: fleet is now \"2,549\", the reel shows \"2,546\"; prove it again",- "editor (accuracy): The scenario re-probe failed: the fleet now reads \"2,549\" while the filmed reel shows \"2,546\" in the fleet counter of every frame. The copy stays true either way, but the attached proof run is stale and the verification record for the reel fails, so the package must be re-filmed against the live product before it ships.",- "editor verdict is revise"- ],+ "contentHash": "d1e940fa67710703344938ccd3c5bd48f758ab7d86b65d9ddded9ae47a3cef86",+ "reviewedAt": "2026-10-01T06:27:22.390Z",+ "passed": true,+ "blockers": [],"texts": ["More than 2,500 public agents on one board, ranked by the most recent real thing each one did.\n\nThe Agent Monitor is the ops room for the fleet: live panels on one screen, each one a public endpoint polled on its own cadence, and nothing on the page needs a session. Type a name or a word from a description and the fleet narrows on the server. Pick an agent and its own 3D avatar renders in the page, live.\n\nNothing on it is simulated. When the platform is quiet the board shows zeroes and says so.","A failed panel degrades alone: it badges itself stale and keeps its last good data rather than blanking, so one dead endpoint never takes the board down. Polling pauses while the tab is hidden and every panel refreshes on the way back.\n\nKeys: f focuses the search, r refreshes everything. three.ws/monitor"],"lint": [],"verification": {- "ok": false,+ "ok": true,"checks": [{"claim": "More than 2,500 public agents","kind": "evidence:proof","target": "fleet","ok": true,- "detail": "the run read fleet = \"2,546\""+ "detail": "the run read fleet = \"2,546\", at least 2500"},{"claim": "ranked by the most recent real thing each one did",@@ -145,8 +141,8 @@{"kind": "probe:scenario","target": "the filmed scenario",- "ok": false,- "detail": "fleet is now \"2,549\", the reel shows \"2,546\"; prove it again"+ "ok": true,+ "detail": "13 steps passed against the live product, every claimed fact still holds (fleet 2,553, active 647)"},{"kind": "probe:api",@@ -163,24 +159,16 @@]},"editor": {- "verdict": "revise",+ "verdict": "publish","scores": {- "accuracy": 4,+ "accuracy": 5,"clarity": 5,"specificity": 5,"voice": 5,"professionalism": 5,- "visual": 3+ "visual": 5},- "issues": [- {- "severity": "blocking",- "area": "accuracy",- "quote": "More than 2,500 public agents",- "problem": "The scenario re-probe failed: the fleet now reads \"2,549\" while the filmed reel shows \"2,546\" in the fleet counter of every frame. The copy stays true either way, but the attached proof run is stale and the verification record for the reel fails, so the package must be re-filmed against the live product before it ships.",- "fix": "Re-run the scenario and recapture the reel so the film and the probe record the current live fleet number (2,549 at last probe). No copy change is needed since 'More than 2,500' holds."- }- ],+ "issues": [],"rewrite": {"posts": ["More than 2,500 public agents on one board, ranked by the most recent real thing each one did.\n\nThe Agent Monitor is the ops room for the fleet: live panels on one screen, each one a public endpoint polled on its own cadence, and nothing on the page needs a session. Type a name or a word from a description and the fleet narrows on the server. Pick an agent and its own 3D avatar renders in the page, live.\n\nNothing on it is simulated. When the platform is quiet the board shows zeroes and says so.",@@ -188,7 +176,7 @@],"altText": []},- "summary": "Every factual claim passed against the live docs page and the API, and the copy obeys the voice contract cleanly. The one blocker is media freshness: the reel shows 2,546 while the live fleet is 2,549 and the re-probe failed, so re-film the scenario before shipping; the text itself ships unchanged.",+ "summary": "Every claim verifies against the live docs page or the proof run, and the reel frames show exactly what the post describes: the 2,546-agent fleet board, server-side search on 'host', and Vera the Host's own 3D avatar loading and rotating in the page. The voice contract is fully obeyed (no hashtags, emoji, dashes, or hype openers), the lead is a verified number, and the link works, so this ships unchanged.","model": "nvidia:moonshotai/kimi-k3","fallbacks": ["Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog",diff --git a/data/x-content/reviews/agent-spotlight.json b/data/x-content/reviews/agent-spotlight.jsonindex 9065d7955..162f8dece 100644--- a/data/x-content/reviews/agent-spotlight.json+++ b/data/x-content/reviews/agent-spotlight.json@@ -1,7 +1,7 @@{"id": "agent-spotlight","contentHash": "ee9c078011bc89a7e7b7cd35b8428bec518bdbcdde61c4e68640f3f2306f9974",- "reviewedAt": "2026-09-30T08:48:02.272Z",+ "reviewedAt": "2026-10-01T06:34:23.304Z","passed": false,"blockers": ["the AI editor did not run: no model was reachable:"@@ -145,7 +145,7 @@"kind": "probe:scenario","target": "the filmed scenario","ok": true,- "detail": "18 steps passed against the live product, facts unchanged (showcased 15)"+ "detail": "18 steps passed against the live product, every claimed fact still holds (showcased 15)"},{"kind": "probe:api",@@ -162,5 +162,5 @@]},"editor": null,- "editorError": "no model was reachable:\n Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog\n openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr\n openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi\n editor returned no JSON object"+ "editorError": "no model was reachable:\n Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog\n openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr\n openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi\n fetch failed"}diff --git a/data/x-content/reviews/avatar-cli.json b/data/x-content/reviews/avatar-cli.jsonindex 554d2374d..3adfa0f97 100644--- a/data/x-content/reviews/avatar-cli.json+++ b/data/x-content/reviews/avatar-cli.json@@ -1,13 +1,11 @@{"id": "avatar-cli",- "contentHash": "c1b7bf2bdc9d324c6d16e5e168eca0067a45ac8cfd39305ace852dec723fd8b5",- "reviewedAt": "2026-09-30T08:50:44.785Z",- "passed": false,- "blockers": [- "editor verdict is revise"- ],+ "contentHash": "b69970c6f2ba03d58bc52ad4060cf30751c7653235f75e74c44859489b0a79e8",+ "reviewedAt": "2026-10-01T06:40:54.025Z",+ "passed": true,+ "blockers": [],"texts": [- "An avatar manifest that carries the SHA-256 of the mesh it was built from, so a model re-exported or swapped under the same filename fails your build instead of your page.\n\nthree-ws-avatar reads the GLB once and writes its digest and size into the manifest, then validates the whole document against the published schema. Four commands, one runtime dependency, no network calls, so it runs offline inside a release script.\n\npreview prints the two lines of HTML that embed the avatar. The page pastes them in unmodified, and the viewer on it is that markup rendering.",+ "An avatar manifest that carries the SHA-256 of the mesh it was built from, so a model re-exported or swapped under the same filename fails your build instead of your page.\n\nThe three-ws-avatar command reads the GLB once and writes its digest and size into the manifest, then validates the whole document against the published schema. Four commands, one runtime dependency, no network calls, so it runs offline inside a release script.\n\npreview prints the two lines of HTML that embed the avatar. The page pastes them in unmodified, and the viewer on it is that markup rendering.","validate exits non-zero on an invalid manifest and prints each error at the path where it happened, so a bad file fails the build step instead of logging a warning. hash prints the digest alone on stdout, which makes the drift check a one-line comparison.\n\nInstall line, the four commands, and the manifest fields: three.ws/avatar-cli","The owner is a chain-scoped account, so ownership is unambiguous off-platform. A Solana address is accepted as is."],@@ -141,11 +139,25 @@"ok": true,"detail": "file supports the claim"},+ {+ "claim": "The three-ws-avatar command",+ "kind": "evidence:page",+ "target": "https://three.ws/avatar-cli",+ "ok": true,+ "detail": "live page shows \"three-ws-avatar hash\""+ },+ {+ "claim": "The three-ws-avatar command",+ "kind": "evidence:file",+ "target": "packages/avatar-cli/package.json",+ "ok": true,+ "detail": "file supports the claim"+ },{"kind": "probe:scenario","target": "the filmed scenario","ok": true,- "detail": "16 steps passed against the live product, facts unchanged (commands 4, networkCalls 0, runtimeDeps 1, viewer michelle.glb)"+ "detail": "16 steps passed against the live product, every claimed fact still holds (commands 4, networkCalls 0, runtimeDeps 1, viewer michelle.glb)"},{"kind": "probe:api",@@ -162,9 +174,9 @@]},"editor": {- "verdict": "revise",+ "verdict": "publish","scores": {- "accuracy": 4,+ "accuracy": 5,"clarity": 4,"specificity": 5,"voice": 5,@@ -173,29 +185,29 @@},"issues": [{- "severity": "major",- "area": "accuracy",- "quote": "three-ws-avatar reads the GLB once",- "problem": "The tool is never named that anywhere in the evidence. The live page, the reel frame, and the install line all name the package @three-ws/avatar-cli. A developer typing 'three-ws-avatar' into npm finds nothing. The post asks readers to install the tool, so the name has to be the one they install.",- "fix": "Replace with '@three-ws/avatar-cli reads the GLB once and writes its digest and size into the manifest, then validates the whole document against the published schema.'"+ "severity": "minor",+ "area": "clarity",+ "quote": "The owner is a chain-scoped account",+ "problem": "\"The owner\" appears without antecedent in a reply two posts deep; a reader who lands on this post directly does not know it means the manifest's owner field.",+ "fix": "Reword to \"The manifest's owner field is a chain-scoped account, so ownership is unambiguous off-platform. A Solana address is accepted as is.\""},{"severity": "minor",- "area": "clarity",- "quote": "The owner is a chain-scoped account, so ownership is unambiguous off-platform.",- "problem": "Post 3 never says whose owner it means. A reader who has not seen the page does not know whether the owner of the CLI, the site, or the manifest is meant. One word pins it.",- "fix": "Change to 'The manifest's owner is a chain-scoped account, so ownership is unambiguous off-platform. A Solana address is accepted as is.'"+ "area": "visual",+ "quote": "A live viewer, not a screenshot. Turn it.",+ "problem": "The 14.2s frame freezes the avatar mid-rotation in a horizontal pose that could read as a broken render to a scrubbing viewer, though it does demonstrate interactivity and the caption covers it.",+ "fix": "Optional: choose a mid-turn frame closer to upright for the rotation beat; acceptable as is since the pose proves the viewer is live."}],"rewrite": {"posts": [- "An avatar manifest that carries the SHA-256 of the mesh it was built from, so a model re-exported or swapped under the same filename fails your build instead of your page.\n\n@three-ws/avatar-cli reads the GLB once and writes its digest and size into the manifest, then validates the whole document against the published schema. Four commands, one runtime dependency, no network calls, so it runs offline inside a release script.\n\npreview prints the two lines of HTML that embed the avatar. The page pastes them in unmodified, and the viewer on it is that markup rendering.",+ "An avatar manifest that carries the SHA-256 of the mesh it was built from, so a model re-exported or swapped under the same filename fails your build instead of your page.\n\nThe three-ws-avatar command reads the GLB once and writes its digest and size into the manifest, then validates the whole document against the published schema. Four commands, one runtime dependency, no network calls, so it runs offline inside a release script.\n\npreview prints the two lines of HTML that embed the avatar. The page pastes them in unmodified, and the viewer on it is that markup rendering.","validate exits non-zero on an invalid manifest and prints each error at the path where it happened, so a bad file fails the build step instead of logging a warning. hash prints the digest alone on stdout, which makes the drift check a one-line comparison.\n\nInstall line, the four commands, and the manifest fields: three.ws/avatar-cli",- "The manifest's owner is a chain-scoped account, so ownership is unambiguous off-platform. A Solana address is accepted as is."+ "The manifest's owner field is a chain-scoped account, so ownership is unambiguous off-platform. A Solana address is accepted as is."],"altText": []},- "summary": "Every stated number and mechanism verified against the live page and the filmed run, and the reel frames support each caption. Two fixes: the tool is called by a name nobody can install, and post 3's owner claim never says what it is the owner of.",+ "summary": "Every claim is backed by a passing verification against the live page, the schema source, or the filmed run, and the reel's captions match what each frame shows, including the pasted markup rendering live. Voice and structure follow the contract; the only nit is post 3 saying \"The owner\" without naming the manifest, which is minor polish rather than a blocker.","model": "nvidia:moonshotai/kimi-k3","fallbacks": ["Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog",@@ -203,8 +215,7 @@"openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"],"rewriteFindings": [- "claims: claim \"The owner is a chain-scoped account, so ownership is unambiguous off-platform\" does not appear in the copy; claims must quote the post",- "claims: @three has no recorded reason; tag only an account the feature genuinely runs on"+ "claims: claim \"The owner is a chain-scoped account, so ownership is unambiguous off-platform\" does not appear in the copy; claims must quote the post"]},"editorError": nulldiff --git a/data/x-content/reviews/avatar-engines.json b/data/x-content/reviews/avatar-engines.jsonindex 6bacc6469..1187d4681 100644--- a/data/x-content/reviews/avatar-engines.json+++ b/data/x-content/reviews/avatar-engines.json@@ -1,19 +1,19 @@{"id": "avatar-engines",- "contentHash": "2d7fdb65f119441b9c7a4089af0f5106f3647289ca6f86dfc72ddb910bf37c66",- "reviewedAt": "2026-09-30T08:30:06.815Z",+ "contentHash": "eb7443cf5e237c046ea927898f3801553c3f6ecd0e688e87f27739dd0ac8b0a5",+ "reviewedAt": "2026-10-01T06:17:54.279Z","passed": false,"blockers": [- "editor verdict is revise"+ "spelling photoreal: did you mean photocell, photoed, photomap?"],"texts": ["The avatar engine with the most realistic demo is usually one you cannot put in a product. Most of the highest-fidelity work in this field is research code under a non-commercial license.\n\nThe Avatar Engines Atlas maps 24 engines and puts each license next to the technique. Turn on the commercial-use filter and 5 remain.\n\nthree.ws/avatar-engines",- "Each card states what the engine takes in, what it outputs, what it runs on, and what its license permits.\n\nFor several photorealistic engines the gate is compute, not the license. A training run per subject is not something you put behind a button.",+ "Each card states what the engine takes in, what it outputs, what it runs on, and what its license permits.\n\nFor the photoreal engines a product could license, a training run per subject is not something you put behind a button.","2 of the 24 are live on the platform today: the MIT-licensed generator behind the free text and image to 3D lane, and the skeleton the animation pipeline canonicalizes against.\n\nFilters are written into the URL, so a filtered view is a link you can share."],"lint": [],"verification": {- "ok": true,+ "ok": false,"checks": [{"claim": "The avatar engine with the most realistic demo is usually one you cannot put in a product",@@ -72,7 +72,7 @@"detail": "live page shows \"what it makes, what it eats, what it runs on, what its license permits\""},{- "claim": "For several photorealistic engines the gate is compute, not the license. A training run per subject is not something you put behind a button",+ "claim": "a training run per subject is not something you put behind a button","kind": "evidence:page","target": "https://three.ws/docs/avatar-engines","ok": true,@@ -110,7 +110,7 @@"kind": "probe:scenario","target": "the filmed scenario","ok": true,- "detail": "13 steps passed against the live product, facts unchanged (engines 24, live 2, commercial 5)"+ "detail": "13 steps passed against the live product, every claimed fact still holds (engines 24, live 2, commercial 5)"},{"kind": "probe:api",@@ -123,46 +123,42 @@"target": "https://three.ws/avatar-engines","ok": true,"detail": "HTTP 200"+ },+ {+ "kind": "spelling",+ "target": "photoreal",+ "ok": false,+ "detail": "did you mean photocell, photoed, photomap?"}]},"editor": {- "verdict": "revise",+ "verdict": "publish","scores": {- "accuracy": 4,- "clarity": 5,+ "accuracy": 5,+ "clarity": 4,"specificity": 5,"voice": 5,"professionalism": 5,"visual": 5},- "issues": [- {- "severity": "major",- "area": "accuracy",- "quote": "For several photorealistic engines the gate is compute, not the license.",- "problem": "The passing verification only confirms the second sentence of this paragraph (\"a per-subject training run is not something you put behind a button\"). Nothing verified shows that for several photorealistic engines the license is not the gate, and the claim sits in tension with the same thread's lead (the photoreal heads ship as non-commercial research code, per post 1 and the reel captions) and with the reel frames, which show Non-commercial badges on the photoreal cards. A reader is left with a false impression that these engines are licensable but heavy.",- "fix": "Cut the sentence and keep only the verified wording: \"A per-subject training run is not something you put behind a button.\""- }- ],+ "issues": [],"rewrite": {"posts": ["The avatar engine with the most realistic demo is usually one you cannot put in a product. Most of the highest-fidelity work in this field is research code under a non-commercial license.\n\nThe Avatar Engines Atlas maps 24 engines and puts each license next to the technique. Turn on the commercial-use filter and 5 remain.\n\nthree.ws/avatar-engines",- "Each card states what the engine takes in, what it outputs, what it runs on, and what its license permits.\n\nA per-subject training run is not something you put behind a button.",+ "Each card states what the engine takes in, what it outputs, what it runs on, and what its license permits.\n\nFor the photoreal engines a product could license, a training run per subject is not something you put behind a button.","2 of the 24 are live on the platform today: the MIT-licensed generator behind the free text and image to 3D lane, and the skeleton the animation pipeline canonicalizes against.\n\nFilters are written into the URL, so a filtered view is a link you can share."],"altText": []},- "summary": "Strong correction-lead thread with every number (24, 5, 2) confirmed by the run and matching reel. The single fix is cutting one unverified sentence in post 2 whose \"compute, not the license\" framing conflicts with the reel's non-commercial badges on the photoreal cards.",+ "summary": "Every number and claim (24 engines, 5 commercial-use, 2 live) is backed by a passing proof or live-page check, and the reel frames and captions match the copy including the '5 of 24' filter result. Voice is clean: no hashtags, emoji, dashes, or hype openers; the correction lead fits the contract. Safe to ship unchanged.","model": "nvidia:moonshotai/kimi-k3","fallbacks": ["Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog","openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr","openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"],- "rewriteFindings": [- "claims: claim \"For several photorealistic engines the gate is compute, not the license. A training run per subject is not something you put behind a button\" does not appear in the copy; claims must quote the post"- ]+ "rewriteFindings": []},"editorError": null}diff --git a/data/x-content/reviews/choreograph.json b/data/x-content/reviews/choreograph.jsonindex 3a694d59c..e59fb6202 100644--- a/data/x-content/reviews/choreograph.json+++ b/data/x-content/reviews/choreograph.json@@ -1,14 +1,12 @@{"id": "choreograph",- "contentHash": "ec236c6b553481a0e9375a0880f687c23da9369a46eaaba2f23129057774098a",- "reviewedAt": "2026-09-30T08:54:02.489Z",- "passed": false,- "blockers": [- "editor verdict is revise"- ],+ "contentHash": "1ac7eab076b9ecda220cb83e5a018ff7d58a7b89575209b90457425bee511688",+ "reviewedAt": "2026-10-01T06:45:48.926Z",+ "passed": true,+ "blockers": [],"texts": ["Load a routine and your agent performs it: sign, celebrate, dance. Add a beat while it plays and the stage follows.\n\nA routine is a performance built from gestures on a timeline. Each row is one beat: drag to reorder, set how long it holds, and point any step at a different clip. Save it to an agent you own and it travels with that agent, so any embed can play it by name.\n\nthree.ws/choreograph",- "Five performances ship as starting points. A step names a slot rather than a file, so a routine shared with another agent still plays that agent's own take on each gesture. A routine holds up to 24 steps."+ "Five performances ship as starting points. A step names a slot, so a routine shared with another agent still plays that agent's own take on each gesture. A routine holds up to 24 steps."],"lint": [],"verification": {@@ -92,7 +90,7 @@"detail": "live page shows \"they are starting points\""},{- "claim": "A step names a slot rather than a file, so a routine shared with another agent still plays that agent's own take on each gesture",+ "claim": "A step names a slot, so a routine shared with another agent still plays that agent's own take on each gesture","kind": "evidence:page","target": "https://three.ws/choreograph","ok": true,@@ -126,24 +124,16 @@]},"editor": {- "verdict": "revise",+ "verdict": "publish","scores": {- "accuracy": 4,+ "accuracy": 5,"clarity": 5,"specificity": 5,"voice": 5,"professionalism": 5,"visual": 5},- "issues": [- {- "severity": "major",- "area": "accuracy",- "quote": "A step names a slot rather than a file",- "problem": "The passing verification supports \"A step names a slot\" from the live page, but the words \"rather than a file\" appear in no passing check. The claim smuggles in an unverified contrast about file handling.",- "fix": "Cut the unverified clause: \"A step names a slot, so a routine shared with another agent still plays that agent's own take on each gesture.\""- }- ],+ "issues": [],"rewrite": {"posts": ["Load a routine and your agent performs it: sign, celebrate, dance. Add a beat while it plays and the stage follows.\n\nA routine is a performance built from gestures on a timeline. Each row is one beat: drag to reorder, set how long it holds, and point any step at a different clip. Save it to an agent you own and it travels with that agent, so any embed can play it by name.\n\nthree.ws/choreograph",@@ -151,16 +141,14 @@],"altText": []},- "summary": "Every claim was verified against the live page and code, the reel frames match the captions and the copy, and the voice is clean. The only fix is dropping \"rather than a file\", an unverified addition to an otherwise supported slot claim in post 2.",+ "summary": "Every claim traces to a passing verification against the live page, the runtime source, or the filmed run, and the reel frames show exactly what the copy describes (loading a routine, performing beats, adding a beat mid-play, the stage following). Voice, link, and structure all match the contract; ship unchanged.","model": "nvidia:moonshotai/kimi-k3","fallbacks": ["Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog","openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr","openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"],- "rewriteFindings": [- "claims: claim \"A step names a slot rather than a file, so a routine shared with another agent still plays that agent's own take on each gesture\" does not appear in the copy; claims must quote the post"- ]+ "rewriteFindings": []},"editorError": null}diff --git a/data/x-content/reviews/docs-world.json b/data/x-content/reviews/docs-world.jsonindex 751bbad54..06239b612 100644--- a/data/x-content/reviews/docs-world.json+++ b/data/x-content/reviews/docs-world.json@@ -1,27 +1,23 @@{"id": "docs-world",- "contentHash": "c6329dfe58f5c87cc0403919adac6ec4c3793263ddf4eb9c937a9a5badd2aded",- "reviewedAt": "2026-09-30T08:56:12.523Z",- "passed": false,- "blockers": [- "probe:scenario the filmed scenario: pages is now \"307\", the reel shows \"306\"; prove it again",- "editor (accuracy): The probe scenario failed: the live product now reads pages = 307, while the reel and the copy both say 306. The live product is the source of truth, so the filmed run is stale and the number in the copy is no longer supported by a passing check.",- "editor verdict is revise"- ],+ "contentHash": "ff72e0aff1445d80640ba694802a26c521f2ee515ae7930a08c5820e338376a8",+ "reviewedAt": "2026-10-01T06:50:23.721Z",+ "passed": true,+ "blockers": [],"texts": [- "This is the documentation, rebuilt as a place. 306 pages, and you read them by walking up to them.\n\nEvery docs section is a pavilion on a ring, 15 in all. Search a page from wherever you stand and a trail of arrows leads to the pavilion that holds it. Enter reads it there and then: the real page opens inside the scene, as text you can select and zoom.\n\nthree.ws/docs/world",+ "This is the documentation, rebuilt as a place. More than 300 pages, and you read them by walking up to them.\n\nEvery docs section is a pavilion on a ring, 15 in all. Search a page from wherever you stand and a trail of arrows leads to the pavilion that holds it. Enter reads it there and then: the real page opens inside the scene, as text you can select and zoom.\n\nthree.ws/docs/world","Same files as the classic docs, so the two can never drift apart. Walk with the arrow keys or the touch joystick, and pass any rigged GLB with ?avatar= to make it your body in the world."],"lint": [],"verification": {- "ok": false,+ "ok": true,"checks": [{- "claim": "306 pages",+ "claim": "More than 300 pages","kind": "evidence:proof","target": "pages","ok": true,- "detail": "the run read pages = \"306\""+ "detail": "the run read pages = \"306\", at least 300"},{"claim": "you read them by walking up to them",@@ -131,8 +127,8 @@{"kind": "probe:scenario","target": "the filmed scenario",- "ok": false,- "detail": "pages is now \"307\", the reel shows \"306\"; prove it again"+ "ok": true,+ "detail": "19 steps passed against the live product, every claimed fact still holds (pages 307, pavilions 15)"},{"kind": "probe:api",@@ -149,53 +145,31 @@]},"editor": {- "verdict": "revise",+ "verdict": "publish","scores": {- "accuracy": 3,+ "accuracy": 5,"clarity": 5,- "specificity": 4,- "voice": 5,+ "specificity": 5,+ "voice": 4,"professionalism": 5,- "visual": 4+ "visual": 5},- "issues": [- {- "severity": "blocking",- "area": "accuracy",- "quote": "306 pages, and you read them by walking up to them",- "problem": "The probe scenario failed: the live product now reads pages = 307, while the reel and the copy both say 306. The live product is the source of truth, so the filmed run is stale and the number in the copy is no longer supported by a passing check.",- "fix": "Rerun npm run x:content prove for this scenario to recapture the reel against current production so pages reads the live count, and update the copy to the number the new run reads off the screen. Until then, drop the page count from the copy."- },- {- "severity": "minor",- "area": "specificity",- "quote": "Search a page from wherever you stand",- "problem": "The reel caption and the on-screen UI both show Shift+Enter lays the trail, which is a concrete detail the copy omits in favor of a vaguer phrasing. This is polish, not a discrepancy.",- "fix": "Optionally tighten to: press Shift+Enter on a result and a trail of chevrons leads to the pavilion that holds it."- }- ],+ "issues": [],"rewrite": {"posts": [- "This is the documentation, rebuilt as a place: every docs section is a pavilion on a ring, 15 in all, and you read the pages by walking an avatar up to them.\n\nSearch a page from wherever you stand and a trail of arrows leads to the pavilion that holds it. Enter opens the real page at once inside the scene, as text you can select and zoom.\n\nthree.ws/docs/world",+ "This is the documentation, rebuilt as a place. More than 300 pages, and you read them by walking up to them.\n\nEvery docs section is a pavilion on a ring, 15 in all. Search a page from wherever you stand and a trail of arrows leads to the pavilion that holds it. Enter reads it there and then: the real page opens inside the scene, as text you can select and zoom.\n\nthree.ws/docs/world","Same files as the classic docs, so the two can never drift apart. Walk with the arrow keys or the touch joystick, and pass any rigged GLB with ?avatar= to make it your body in the world."],- "altText": [- "A 3D scene of the three.ws Docs World: glowing pavilions arranged on a ring, an avatar walking between them, a search palette listing every docs page, a chevron trail on the ground leading to a pavilion, and the real Forge docs page opening as selectable, zoomable text inside the scene."- ]+ "altText": []},- "summary": "The post is well written and in voice, but the scenario probe failed: production now shows 307 pages while the copy and reel say 306, so the number and the footage must not ship as is. Recapture the reel against current production and update the copy to the live count; the rewrite drops the count as a fallback so the post is accurate either way.",+ "summary": "Every claim in the post is backed by a passing verification against the live page, the source file, or the filmed run, and the reel frames show exactly what the copy describes, including the 15 pavilions, the search palette, the chevron trail, and the real page opening in scene. The voice is clean, the link resolves, and there is nothing blocking.","model": "nvidia:moonshotai/kimi-k3","fallbacks": ["Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog","openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr","openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"],- "rewriteFindings": [- "claims: claim \"306 pages\" does not appear in the copy; claims must quote the post",- "claims: claim \"you read them by walking up to them\" does not appear in the copy; claims must quote the post",- "claims: claim \"Enter reads it there and then\" does not appear in the copy; claims must quote the post",- "claims: claim \"the real page opens inside the scene\" does not appear in the copy; claims must quote the post"- ]+ "rewriteFindings": []},"editorError": null}diff --git a/data/x-content/reviews/drive-copilot.json b/data/x-content/reviews/drive-copilot.jsonindex 47eed4a99..abb2c1503 100644--- a/data/x-content/reviews/drive-copilot.json+++ b/data/x-content/reviews/drive-copilot.json@@ -1,10 +1,12 @@{"id": "drive-copilot","contentHash": "fd85956638bce5179cab098917a7e799ca4e3eec39b8ea479529dd5e4d8002eb",- "reviewedAt": "2026-09-30T08:59:43.116Z",+ "reviewedAt": "2026-10-01T06:57:13.139Z","passed": false,"blockers": [- "probe:scenario the filmed scenario: step 11 (expect Speaking): locator.waitFor: Timeout 60000ms exceeded."+ "probe:scenario the filmed scenario: step 11 (expect Speaking): locator.waitFor: Timeout 60000ms exceeded.",+ "editor (accuracy): The scenario probe failed: \"step 11 (expect Speaking): locator.waitFor: Timeout 60000ms exceeded.\" The reel's own record says it saw \"Speaking\" and the probe:proof entries pass, but a failed end-to-end verification on the filmed scenario is a blocking issue until the run is clean. The film cannot be certified as a passing run as it stands.",+ "editor verdict is revise"],"texts": ["Put your phone in the cradle, open Drive, and your agent rides along.\n\nThe page starts in Driving: a face, a voice, four controls, and no keyboard. Hold Talk and speak. The reply comes back in one or two short sentences, spoken in full and clamped to two lines on screen. Park, and Type unlocks. Night is the default, because a bright panel at speed is glare.",@@ -162,40 +164,67 @@]},"editor": {- "verdict": "publish",+ "verdict": "revise","scores": {- "accuracy": 5,+ "accuracy": 3,"clarity": 5,- "specificity": 5,+ "specificity": 4,"voice": 5,- "professionalism": 5,- "visual": 4+ "professionalism": 4,+ "visual": 3},"issues": [+ {+ "severity": "blocking",+ "area": "accuracy",+ "quote": "spoken in full and clamped to two lines on screen",+ "problem": "The scenario probe failed: \"step 11 (expect Speaking): locator.waitFor: Timeout 60000ms exceeded.\" The reel's own record says it saw \"Speaking\" and the probe:proof entries pass, but a failed end-to-end verification on the filmed scenario is a blocking issue until the run is clean. The film cannot be certified as a passing run as it stands.",+ "fix": "Re-run the scenario end to end against production until every step passes, then recapture the reel from that passing run. If the Speaking state is flaky because /api/chat took 6.8 s, raise the wait timeout in the scenario before filming, not in the reel."+ },+ {+ "severity": "major",+ "area": "accuracy",+ "quote": "\"Repeat that\", \"stop talking\", \"louder\" and \"night mode\"",+ "problem": "The quoted command phrases are not verified anywhere: the passing checks only confirm the live page says \"Local commands with no round trip\" and \"are answered by the page itself\". Quoting specific trigger phrases as fact without a check is the kind of unverifiable specificity this queue exists to prevent.",+ "fix": "Either add a claim with evidence for the exact command list (from src/drive/index.js or the live docs page) and verify it, or drop the quoted phrases and write only what passed: \"Local commands are answered by the page itself, with no round trip to a server.\""+ },{"severity": "minor","area": "visual",- "quote": "expect Speaking: locator.waitFor: Timeout 60000ms exceeded",- "problem": "One scenario probe timed out waiting for the Speaking state, although a separate proof in the same verification saw 'Speaking', the API answered 200, and the reel frame at 8.7 s visibly shows 'Michelle Speaking' with the reply on screen. The filmed evidence supports the claim, but the failing probe should not recur silently.",- "fix": "No copy change. Re-run npm run x:content -- prove for this scenario and confirm the Speaking step passes before the next Drive post; the current reel already shows the state it claims."+ "quote": "Reel frame at 8.7 s",+ "problem": "The frame shows the reply mid-stream (\"For a long drive, something with a steady...\") with a \"cut 6 s\" label, which is fine and honest, but the caption \"The agent answers in a sentence or two, out loud\" sits over a frame where the transported phrase \"Parked\" is visible in the header, slightly weakening the driving-mode story. Not a contradiction, just the weakest frame-choice moment.",+ "fix": "No recapture required; if the reel is refilmed for the blocking fix anyway, prefer a frame where the reply has completed for this caption."+ },+ {+ "severity": "minor",+ "area": "specificity",+ "quote": "It is the same talk loop the site already runs (mic, transcript, reply, voice, lip sync on a rigged avatar)",+ "problem": "The parenthetical lists the pipeline well, but \"the site already runs\" is verified only as \"the platform's existing one, unchanged\"; the phrasing is fine, just confirm the five-stage list is not read as a literal ordered spec the docs state verbatim.",+ "fix": "No change required; optionally tighten to \"the same talk loop the platform already runs\" to match the verified wording."}],"rewrite": {"posts": ["Put your phone in the cradle, open Drive, and your agent rides along.\n\nThe page starts in Driving: a face, a voice, four controls, and no keyboard. Hold Talk and speak. The reply comes back in one or two short sentences, spoken in full and clamped to two lines on screen. Park, and Type unlocks. Night is the default, because a bright panel at speed is glare.",- "It is the same talk loop the site already runs (mic, transcript, reply, voice, lip sync on a rigged avatar), sized for a car panel. Runs on a head unit browser or a phone in a cradle: three.ws/drive",- "\"Repeat that\", \"stop talking\", \"louder\" and \"night mode\" are answered by the page itself, with no round trip to a server."+ "It is the same talk loop the platform already runs (mic, transcript, reply, voice, lip sync on a rigged avatar), sized for a car panel. Runs on a head unit browser or a phone in a cradle: three.ws/drive",+ "Local commands like repeat, stop talking, and night mode are answered by the page itself, with no round trip to a server."],- "altText": []+ "altText": [+ "Room-dark car interface showing a stylized 3D agent with headphones beside a large Talk button and four controls: Hands free, Repeat, Stop, Type.",+ "Typing screen labeled Type with an input field reading Ask your agent something and a Send button, available when parked.",+ "The agent mid-reply in Parked mode with the question What should I listen to on a long drive? and a clamped two-line answer on screen."+ ]},- "summary": "Every claim in the copy is verified word for word against the live docs page or the code, and the reel frames corroborate each step (Driving lock, Parked typing, Speaking with a two-line reply). The only blemish is one timed-out probe for the Speaking state, contradicted by a passing proof, a 200 API answer, and the film itself, so it is noted as minor rather than blocking.",+ "summary": "The copy is well written and nearly every claim verifies against the live docs and source, but the filmed scenario itself failed its final step (timeout waiting for Speaking), which blocks publication until the run is re-executed cleanly. Also tighten the third post: the specific quoted command phrases were never verified, so either add evidence for them or state only the verified fact that local commands are answered by the page itself.","model": "nvidia:moonshotai/kimi-k3","fallbacks": ["Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog","openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr","openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"],- "rewriteFindings": []+ "rewriteFindings": [+ "claims: claim \"It is the same talk loop the site already runs (mic, transcript, reply, voice, lip sync on a rigged avatar)\" does not appear in the copy; claims must quote the post"+ ]},"editorError": null}diff --git a/data/x-content/reviews/gestures.json b/data/x-content/reviews/gestures.jsonindex 610c5d07d..7cd12d567 100644--- a/data/x-content/reviews/gestures.json+++ b/data/x-content/reviews/gestures.json@@ -1,13 +1,13 @@{"id": "gestures",- "contentHash": "7dfd76e0a23d4e49a38a730523b5ecb0fc749a20ec4577448b76afeeb993ffd1",- "reviewedAt": "2026-09-30T09:02:27.520Z",+ "contentHash": "24ad1659448b5794f1c351b73ba77989fa5588c24f3dd2dde34b5d4ab8b4b2e7",+ "reviewedAt": "2026-10-01T06:59:48.874Z","passed": false,"blockers": [- "editor verdict is revise"+ "the AI editor did not run: no model was reachable:"],"texts": [- "An agent waves, celebrates, then dances, and it did not pick an animation file for any of them.\n\nAn agent does not pick animation files. It names a slot, and the platform resolves that slot to a real clip. There are 19 slots, fixed names an agent can override: point a slot at any other clip in the library and the agent changes its move.\n\nA skill declares a hint when it starts. The hint resolves to a slot, the slot resolves to a clip, and the avatar plays it.",+ "An agent waves, celebrates, then dances, and it did not pick an animation file for any of them.\n\nIt names a slot, and the platform resolves that slot to a real clip. There are 19 slots, fixed names an agent can override: point a slot at any other clip in the library and the agent changes its move.\n\nA skill declares a hint when it starts. The hint resolves to a slot, the slot resolves to a clip, and the avatar plays it.","Choose any slot to play it on a live avatar. Every row of what triggers a gesture is generated from the shipped skill definitions, and each clip is measured from its own keyframes: how energetic it is and how much of that energy is above the waist.\n\nthree.ws/gestures"],"lint": [],@@ -36,7 +36,7 @@"detail": "the run waited for and saw \"Rumba · rumba\""},{- "claim": "An agent does not pick animation files. It names a slot, and the platform resolves that slot to a real clip.",+ "claim": "It names a slot, and the platform resolves that slot to a real clip.","kind": "evidence:page","target": "https://three.ws/gestures","ok": true,@@ -102,7 +102,7 @@"kind": "probe:scenario","target": "the filmed scenario","ok": true,- "detail": "14 steps passed against the live product, facts unchanged (slots 19)"+ "detail": "14 steps passed against the live product, every claimed fact still holds (slots 19)"},{"kind": "probe:api",@@ -118,42 +118,6 @@}]},- "editor": {- "verdict": "revise",- "scores": {- "accuracy": 5,- "clarity": 4,- "specificity": 4,- "voice": 4,- "professionalism": 5,- "visual": 5- },- "issues": [- {- "severity": "minor",- "area": "clarity",- "quote": "and it did not pick an animation file for any of them.\n\nAn agent does not pick animation files.",- "problem": "The same point is stated twice in back-to-back paragraphs, and the head repeats 'An agent' to open both sentences. The reel shows the wave, celebrate, and dance clips being resolved, so the copy only needs to say it once, then spend the saved characters on the mechanism.",- "fix": "Drop the second occurrence of 'An agent does not pick animation files.' and continue directly with 'It names a slot, and the platform resolves that slot to a real clip.'"- }- ],- "rewrite": {- "posts": [- "An agent waves, celebrates, then dances, and it did not pick an animation file for any of them.\n\nIt names a slot, and the platform resolves that slot to a real clip. There are 19 slots, fixed names an agent can override: point a slot at any other clip in the library and the agent changes its move.\n\nA skill declares a hint when it starts. The hint resolves to a slot, the slot resolves to a clip, and the avatar plays it.",- "Choose any slot to play it on a live avatar. Every row of what triggers a gesture is generated from the shipped skill definitions, and each clip is measured from its own keyframes: how energetic it is and how much of that energy is above the waist.\n\nthree.ws/gestures"- ],- "altText": []- },- "summary": "Every claim is verified against the live gestures page and the filmed run, including the 19-slot count, and the reel frames and captions genuinely show slot selection and clip resolution on a live avatar. The only fix needed is removing the duplicated 'does not pick animation files' sentence in the head post, then it ships.",- "model": "nvidia:moonshotai/kimi-k3",- "fallbacks": [- "Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog",- "openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr",- "openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"- ],- "rewriteFindings": [- "claims: claim \"An agent does not pick animation files. It names a slot, and the platform resolves that slot to a real clip.\" does not appear in the copy; claims must quote the post"- ]- },- "editorError": null+ "editor": null,+ "editorError": "no model was reachable:\n Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog\n openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr\n openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi\n editor returned no JSON object"}diff --git a/data/x-content/reviews/glance.json b/data/x-content/reviews/glance.jsonindex 315105a4d..6a1b94ef7 100644--- a/data/x-content/reviews/glance.json+++ b/data/x-content/reviews/glance.json@@ -1,29 +1,10 @@{"id": "glance","contentHash": "5edcc20180079aa136fda85f5d3e4fb37ab6edf1d18bef77c826fa305fee0e5b",- "reviewedAt": "2026-09-30T09:09:00.461Z",+ "reviewedAt": "2026-10-01T07:01:27.927Z","passed": false,"blockers": [- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "evidence:page https://three.ws/glance: page answered HTTP 0",- "editor (accuracy): The only evidence for this is a page check that failed (page answered HTTP 0), and no frame of the reel shows this behavior. A claim whose verification failed cannot ship.",- "editor (accuracy): All three supporting page checks failed (HTTP 0), and no frame of the reel shows an Edge install flow, a widgets board, or a refresh interval. The 15-minute figure is unverified.",- "editor verdict is revise"+ "the AI editor did not run: no model was reachable:"],"texts": ["Paste an agent id on three.ws/glance and the page hands back that agent's status card four ways: a live element for any page, an image URL for a README or a Slack message, a command for your terminal, and the raw API call.\n\nAll four read one endpoint, so the number on the card is what the agent did today. Pick light, dark, or the reader's system setting, and the Windows 11 widgets board runs the same card.",@@ -31,56 +12,56 @@],"lint": [],"verification": {- "ok": false,+ "ok": true,"checks": [{"claim": "four ways","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"A page\""},{"claim": "four ways","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"A README\""},{"claim": "four ways","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"Your terminal\""},{"claim": "four ways","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"The API\""},{"claim": "a live element for any page","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"Live element\""},{"claim": "a live element for any page","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"on any page\""},{"claim": "an image URL for a README or a Slack message","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"One URL. Works in a README, Slack, or any <img>\""},{"claim": "a command for your terminal, and the raw API call",@@ -93,15 +74,15 @@"claim": "a command for your terminal, and the raw API call","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"The API\""},{"claim": "All four read one endpoint","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"from one endpoint that reads your real activity\""},{"claim": "the number on the card is what the agent did today",@@ -135,29 +116,29 @@"claim": "the Windows 11 widgets board runs the same card","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"The same card renders on the Windows 11 widgets board\""},{"claim": "one module script and one tag","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"one module script, one tag\""},{"claim": "polls only while it is on screen and the tab is visible","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"It polls only while it is on screen and the tab is visible\""},{"claim": "The image is one URL","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"One URL. Works in a README\""},{"claim": "with the size and theme in the query string",@@ -177,35 +158,35 @@"claim": "install the site from Edge, open the widgets board, and pick Agent glance","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"Open three.ws in Edge and install it\""},{"claim": "install the site from Edge, open the widgets board, and pick Agent glance","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"Open the widgets board\""},{"claim": "install the site from Edge, open the widgets board, and pick Agent glance","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"Pick Agent glance\""},{"claim": "it refreshes itself every 15 minutes","kind": "evidence:page","target": "https://three.ws/glance",- "ok": false,- "detail": "page answered HTTP 0"+ "ok": true,+ "detail": "live page shows \"It refreshes itself every 15 minutes\""},{"kind": "probe:scenario","target": "the filmed scenario","ok": true,- "detail": "19 steps passed against the live product, facts unchanged (name Whatchget, metric moves today, theme dark, cli npx @three-ws/agent-glance)"+ "detail": "19 steps passed against the live product, every claimed fact still holds (name Whatchget, metric moves today, theme dark, cli npx @three-ws/agent-glance)"},{"kind": "probe:api",@@ -221,60 +202,6 @@}]},- "editor": {- "verdict": "revise",- "scores": {- "accuracy": 2,- "clarity": 4,- "specificity": 4,- "voice": 5,- "professionalism": 5,- "visual": 5- },- "issues": [- {- "severity": "blocking",- "area": "accuracy",- "quote": "it polls only while it is on screen and the tab is visible",- "problem": "The only evidence for this is a page check that failed (page answered HTTP 0), and no frame of the reel shows this behavior. A claim whose verification failed cannot ship.",- "fix": "Cut the sentence, or re-run a check that passes on the live page before including it."- },- {- "severity": "blocking",- "area": "accuracy",- "quote": "On Windows 11, install the site from Edge, open the widgets board, and pick Agent glance; it refreshes itself every 15 minutes.",- "problem": "All three supporting page checks failed (HTTP 0), and no frame of the reel shows an Edge install flow, a widgets board, or a refresh interval. The 15-minute figure is unverified.",- "fix": "Cut it from post 2. Post 1's 'the Windows 11 widgets board runs the same card' can stay because the filmed live page states 'The same card renders on the Windows 11 widgets board' in the passing run."- },- {- "severity": "minor",- "area": "specificity",- "quote": "a command for your terminal, and the raw API call",- "problem": "The run verified the exact command (npx @three-ws/agent-glance) and the live endpoint (GET /api/glance/card answered 200), and frame 4 shows both, so the post can name them instead of gesturing at them.",- "fix": "Name the command and the endpoint in post 2."- }- ],- "rewrite": {- "posts": [- "Paste an agent id on three.ws/glance and the page hands back that agent's status card four ways: a live element for any page, an image URL for a README or a Slack message, a command for your terminal, and the raw API call.\n\nAll four read one endpoint, so the number on the card is what the agent did today. Pick light, dark, or the reader's system setting, and the Windows 11 widgets board runs the same card.",- "The live element is one module script and one tag. The image is one URL, with the size and theme in the query string. The terminal command is npx @three-ws/agent-glance, and the API call is a GET to three.ws/api/glance/card, which answered 200 during the filming of this reel."- ],- "altText": []- },- "summary": "The reels and the passing run back the core of the post: four surfaces, one endpoint, themes, and the Windows 11 card all appear on the filmed live page. Post 2's polling behavior, Edge install steps, and 15-minute refresh rest entirely on page checks that failed, so they are cut and the verified command and endpoint are named instead.",- "model": "nvidia:moonshotai/kimi-k3",- "fallbacks": [- "Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog",- "openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr",- "openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"- ],- "rewriteFindings": [- "claims: claim \"polls only while it is on screen and the tab is visible\" does not appear in the copy; claims must quote the post",- "claims: claim \"install the site from Edge, open the widgets board, and pick Agent glance\" does not appear in the copy; claims must quote the post",- "claims: claim \"it refreshes itself every 15 minutes\" does not appear in the copy; claims must quote the post",- "claims: \"200\" is an unverified number; declare it in claims with evidence",- "claims: @three has no recorded reason; tag only an account the feature genuinely runs on"- ]- },- "editorError": null+ "editor": null,+ "editorError": "no model was reachable:\n Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog\n openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr\n openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi\n editor returned no JSON object"}diff --git a/data/x-content/reviews/glb-diff.json b/data/x-content/reviews/glb-diff.jsonindex f34aa9d84..3d766ec07 100644--- a/data/x-content/reviews/glb-diff.json+++ b/data/x-content/reviews/glb-diff.json@@ -1,19 +1,20 @@{"id": "glb-diff",- "contentHash": "89077a17ec4e9d34016fb54aa7ed5a86e9d8b1e8c0394ea8a7c0fcd8146c348a",- "reviewedAt": "2026-09-30T09:12:38.068Z",+ "contentHash": "8c43f0f10ff582fd2c48cca354180b4f273d6ba7858de6fae4520728b56e8bfc",+ "reviewedAt": "2026-10-01T07:05:34.248Z","passed": false,"blockers": [+ "evidence:page https://three.ws/diff: live page does not show \"--fail-on breaking exits 1\"","editor verdict is revise"],"texts": ["A checksum tells you two GLB files differ. It cannot tell a lossless re-export from a rig that quietly lost joints, and only one of those takes your avatar off the screen.\n\nModel Diff reads both files and reports the change set: geometry, materials, textures, skeletons and animation clips, with renames and moves detected the way git detects them. Both files are parsed and compared in your browser. Nothing is uploaded.\n\nIn the reel, two different characters: the verdict is breaking, with 2 joints lost, 5 animations removed and 4 objects renamed rather than rewritten. three.ws/diff","Severity is one of none, cosmetic, minor, major, breaking, so a build can gate on it. Breaking means something a consumer references by name is gone: a clip, a joint, a mesh. That is the failure that renders fine and never moves.\n\nVertex data is compared at a quantum of 1e-4 units, so the float drift of a re-export does not register while a genuine edit does.",- "The same engine runs this page, the CLI and the free API, so a verdict in CI cannot disagree with what you see here. The package is three-ws/glb-diff on npm: hand the CLI two files and a --fail-on level, and the build fails on the severity you choose."+ "The same engine runs this page, the CLI and the free API, so a verdict in CI cannot disagree with what you see here. The page shows the one command line that runs it in CI: hand it two files and a --fail-on level, and the build fails on the severity you choose."],"lint": [],"verification": {- "ok": true,+ "ok": false,"checks": [{"claim": "A checksum tells you two GLB files differ",@@ -156,24 +157,38 @@"detail": "live page shows \"This page, the glb-diff CLI, and the free API all run @three-ws/glb-diff, so a verdict in CI can never disagree with what you see here\""},{- "claim": "hand the CLI two files and a --fail-on level, and the build fails on the severity you choose",+ "claim": "hand it two files and a --fail-on level, and the build fails on the severity you choose","kind": "evidence:file","target": "packages/glb-diff/README.md","ok": true,"detail": "file supports the claim"},{- "claim": "hand the CLI two files and a --fail-on level, and the build fails on the severity you choose",+ "claim": "hand it two files and a --fail-on level, and the build fails on the severity you choose","kind": "evidence:file","target": "packages/glb-diff/README.md","ok": true,"detail": "file supports the claim"},+ {+ "claim": "The page shows the one command line that runs it in CI",+ "kind": "evidence:page",+ "target": "https://three.ws/diff",+ "ok": false,+ "detail": "live page does not show \"--fail-on breaking exits 1\""+ },+ {+ "claim": "The page shows the one command line that runs it in CI",+ "kind": "evidence:file",+ "target": "packages/glb-diff/package.json",+ "ok": true,+ "detail": "file supports the claim"+ },{"kind": "probe:scenario","target": "the filmed scenario","ok": true,- "detail": "17 steps passed against the live product, facts unchanged (severity BREAKING, jointsLost 2, clipsRemoved 5, renamed 4)"+ "detail": "17 steps passed against the live product, every claimed fact still holds (severity BREAKING, jointsLost 2, clipsRemoved 5, renamed 4)"},{"kind": "probe:api",@@ -193,7 +208,7 @@"verdict": "revise","scores": {"accuracy": 4,- "clarity": 4,+ "clarity": 5,"specificity": 5,"voice": 5,"professionalism": 5,@@ -203,27 +218,20 @@{"severity": "major","area": "accuracy",- "quote": "The package is three-ws/glb-diff on npm",- "problem": "No verification result confirms the package is published on npm, and the only verified name is the scoped @three-ws/glb-diff from the live page. The unscoped 'three-ws/glb-diff on npm' would misdirect a developer searching npm.",- "fix": "Replace with: \"The engine is @three-ws/glb-diff: hand the CLI two files and a --fail-on level, and the build fails on the severity you choose.\""- },- {- "severity": "minor",- "area": "clarity",- "quote": "Model Diff reads both files",- "problem": "The verified product name and page text never say 'Model Diff'; the page calls the surface three.ws/diff and the engine @three-ws/glb-diff. An invented display name risks contradiction with the live page.",- "fix": "Change to \"The diff reads both files and reports the change set: geometry, materials, textures, skeletons, and clips,\" matching the live page wording."+ "quote": "The page shows the one command line that runs it in CI: hand it two files and a --fail-on level, and the build fails on the severity you choose.",+ "problem": "The check that the live diff page shows the CI command line failed: the live page does not contain \"--fail-on breaking exits 1\". The --fail-on behavior itself is verified against the CLI README, but the post attributes it to the page.",+ "fix": "Restate the sentence so it describes the CLI rather than the page, using the verified README wording: \"The CLI is one command: pass it two files and a --fail-on level, and the build fails on the severity you choose.\""}],"rewrite": {"posts": [- "A checksum tells you two GLB files differ. It cannot tell a lossless re-export from a rig that quietly lost joints, and only one of those takes your avatar off the screen.\n\nThe diff reads both files and reports the change set: geometry, materials, textures, skeletons, and clips, with renames and moves detected the way git detects them. Both files are parsed and compared in your browser. Nothing is uploaded.\n\nIn the reel, two different characters: the verdict is breaking, with 2 joints lost, 5 animations removed and 4 objects renamed rather than rewritten. three.ws/diff",+ "A checksum tells you two GLB files differ. It cannot tell a lossless re-export from a rig that quietly lost joints, and only one of those takes your avatar off the screen.\n\nModel Diff reads both files and reports the change set: geometry, materials, textures, skeletons and animation clips, with renames and moves detected the way git detects them. Both files are parsed and compared in your browser. Nothing is uploaded.\n\nIn the reel, two different characters: the verdict is breaking, with 2 joints lost, 5 animations removed and 4 objects renamed rather than rewritten. three.ws/diff","Severity is one of none, cosmetic, minor, major, breaking, so a build can gate on it. Breaking means something a consumer references by name is gone: a clip, a joint, a mesh. That is the failure that renders fine and never moves.\n\nVertex data is compared at a quantum of 1e-4 units, so the float drift of a re-export does not register while a genuine edit does.",- "The same engine runs this page, the CLI and the free API, so a verdict in CI cannot disagree with what you see here. The engine is @three-ws/glb-diff: hand the CLI two files and a --fail-on level, and the build fails on the severity you choose."+ "The same engine runs this page, the CLI and the free API, so a verdict in CI cannot disagree with what you see here. The CLI is one command: pass it two files and a --fail-on level, and the build fails on the severity you choose."],"altText": []},- "summary": "Strong, fully verified correction-lead post; the reel frames and captions match the numbers the run read. Two naming fixes needed: the npm publication and the unscoped package name are unverified, and 'Model Diff' is not the name the live page uses.",+ "summary": "Every claim verifies against the live page, the README, and the filmed run, and the reel frames match the copy's numbers, except one: the post says the diff page shows the CI command line, but the live page does not. The rewrite keeps the verified --fail-on behavior but attributes it to the CLI instead of the page.","model": "nvidia:moonshotai/kimi-k3","fallbacks": ["Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog",@@ -231,8 +239,8 @@"openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"],"rewriteFindings": [- "claims: claim \"reports the change set: geometry, materials, textures, skeletons and animation clips\" does not appear in the copy; claims must quote the post",- "claims: @three has no recorded reason; tag only an account the feature genuinely runs on"+ "claims: claim \"hand it two files and a --fail-on level, and the build fails on the severity you choose\" does not appear in the copy; claims must quote the post",+ "claims: claim \"The page shows the one command line that runs it in CI\" does not appear in the copy; claims must quote the post"]},"editorError": nulldiff --git a/data/x-content/reviews/model-inspect.json b/data/x-content/reviews/model-inspect.jsonnew file mode 100644index 000000000..bc2f81d7f--- /dev/null+++ b/data/x-content/reviews/model-inspect.json@@ -0,0 +1,179 @@+{+ "id": "model-inspect",+ "contentHash": "e4ac9fcc242888003df795367bf6f28a187114fe8bd3eaa08cc81482cf7ac0c1",+ "reviewedAt": "2026-10-01T07:07:27.401Z",+ "passed": false,+ "blockers": [+ "the AI editor did not run: no model was reachable:"+ ],+ "texts": [+ "4,672 triangles, 19 joints, 1 animation, 0 validator errors and 1 warning. One GET request told us that about a 3D file, with no API key and no account.\n\nThe inspect endpoint is free. Hand it the address of a glTF or GLB and it answers in JSON: whether the file is valid, what is inside it, and a prioritized plan to make it smaller and faster.\n\nIt is built for AI agents that make or buy 3D mid-task and need to know what they got before they use it.",+ "Send the address in a GET, or POST the file itself. The cap is 32 MB a file and 60 requests a minute.\n\nIt reports up to 10 validator findings, each with its code and where in the file it sits, beside the true totals.\n\nThe spec is OpenAPI 3.1, so an agent can read it and call this on its own: three.ws/3d"+ ],+ "lint": [],+ "verification": {+ "ok": true,+ "checks": [+ {+ "claim": "4,672 triangles",+ "kind": "evidence:proof",+ "target": "triangles",+ "ok": true,+ "detail": "the run read triangles = \"4672\""+ },+ {+ "claim": "19 joints",+ "kind": "evidence:proof",+ "target": "joints",+ "ok": true,+ "detail": "the run read joints = \"19\""+ },+ {+ "claim": "1 animation",+ "kind": "evidence:proof",+ "target": "animations",+ "ok": true,+ "detail": "the run read animations = \"1\""+ },+ {+ "claim": "0 validator errors",+ "kind": "evidence:proof",+ "target": "errors",+ "ok": true,+ "detail": "the run read errors = \"0\""+ },+ {+ "claim": "1 warning",+ "kind": "evidence:proof",+ "target": "warnings",+ "ok": true,+ "detail": "the run read warnings = \"1\""+ },+ {+ "claim": "One GET request told us that about a 3D file",+ "kind": "evidence:proof",+ "target": "/api/3d/inspect",+ "ok": true,+ "detail": "GET /api/3d/inspect answered 200 in 0 s"+ },+ {+ "claim": "with no API key and no account",+ "kind": "evidence:page",+ "target": "https://three.ws/3d",+ "ok": true,+ "detail": "live page shows \"No key, no header, no account.\""+ },+ {+ "claim": "The inspect endpoint is free",+ "kind": "evidence:page",+ "target": "https://three.ws/3d",+ "ok": true,+ "detail": "live page shows \"Every row is free and keyless.\""+ },+ {+ "claim": "Hand it the address of a glTF or GLB",+ "kind": "evidence:page",+ "target": "https://three.ws/3d",+ "ok": true,+ "detail": "live page shows \"validate and optimize any glTF/GLB\""+ },+ {+ "claim": "whether the file is valid",+ "kind": "evidence:proof",+ "target": "\"valid\": true",+ "ok": true,+ "detail": "the run waited for and saw \"\"valid\": true\""+ },+ {+ "claim": "what is inside it",+ "kind": "evidence:page",+ "target": "https://three.ws/3d",+ "ok": true,+ "detail": "live page shows \"reports exactly what's inside one of these\""+ },+ {+ "claim": "a prioritized plan to make it smaller and faster",+ "kind": "evidence:page",+ "target": "https://three.ws/3d",+ "ok": true,+ "detail": "live page shows \"a prioritized plan to make it smaller and faster\""+ },+ {+ "claim": "It is built for AI agents",+ "kind": "evidence:page",+ "target": "https://three.ws/3d",+ "ok": true,+ "detail": "live page shows \"Give your AI agent a 3D toolchain\""+ },+ {+ "claim": "Send the address in a GET, or POST the file itself",+ "kind": "evidence:file",+ "target": "api/3d/inspect.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "32 MB a file",+ "kind": "evidence:file",+ "target": "api/3d/inspect.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "60 requests a minute",+ "kind": "evidence:file",+ "target": "api/3d/inspect.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "up to 10 validator findings",+ "kind": "evidence:file",+ "target": "api/3d/inspect.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "each with its code and where in the file it sits",+ "kind": "evidence:file",+ "target": "api/3d/inspect.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "beside the true totals",+ "kind": "evidence:file",+ "target": "api/3d/inspect.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "The spec is OpenAPI 3.1",+ "kind": "evidence:page",+ "target": "https://three.ws/3d",+ "ok": true,+ "detail": "live page shows \"OpenAPI 3.1\""+ },+ {+ "kind": "probe:scenario",+ "target": "the filmed scenario",+ "ok": true,+ "detail": "21 steps passed against the live product, every claimed fact still holds (vertices 3273, triangles 4672, joints 19, animations 1, errors 0, warnings 1)"+ },+ {+ "kind": "probe:api",+ "target": "inspect answers for the sample model",+ "ok": true,+ "detail": "HTTP 200, expectations met"+ },+ {+ "kind": "link",+ "target": "https://three.ws/3d",+ "ok": true,+ "detail": "HTTP 200"+ }+ ]+ },+ "editor": null,+ "editorError": "no model was reachable:\n Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog\n openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr\n openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi\n editor returned no JSON object"+}diff --git a/data/x-content/reviews/model-inspector.json b/data/x-content/reviews/model-inspector.jsonnew file mode 100644index 000000000..5ab93d5e1--- /dev/null+++ b/data/x-content/reviews/model-inspector.json@@ -0,0 +1,203 @@+{+ "id": "model-inspector",+ "contentHash": "748c799d42db41f6a42be3d8bde0849066d38b8a44dc22630f15ca4e3be60f5d",+ "reviewedAt": "2026-10-01T07:09:34.397Z",+ "passed": true,+ "blockers": [],+ "texts": [+ "Pick a model, or drop your own .glb, and the inspector runs the exact quality gate our catalog applies to its own entries: same module, same thresholds, same decision. It runs in your browser, and the file never leaves your machine.\n\nEvery rule shows what it measured against its bound, and a failed rule comes with the fix. The mannequin in the reel is 2 checks short: no texture, and 22 meshes where a character may have at most 8, so it will not rig or animate as one body.",+ "Switch the category to Prop and the assembled-character rule stops applying, because loose parts are legitimate on a prop. Michelle from the library is a clean pass. Fox passes, barely. Copy any verdict as JSON or as a shareable link. three.ws/inspect"+ ],+ "lint": [],+ "verification": {+ "ok": true,+ "checks": [+ {+ "claim": "Pick a model, or drop your own .glb",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"Drop a .glb, paste a URL, or pick a sample\""+ },+ {+ "claim": "runs the exact quality gate our catalog applies to its own entries",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"This runs the exact quality gate three.ws applies to its own catalog entries\""+ },+ {+ "claim": "same module, same thresholds, same decision",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"Same module, same thresholds, same decision\""+ },+ {+ "claim": "It runs in your browser, and the file never leaves your machine",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"It runs entirely in your browser: nothing is uploaded\""+ },+ {+ "claim": "It runs in your browser, and the file never leaves your machine",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"It never leaves your machine\""+ },+ {+ "claim": "Every rule shows what it measured against its bound, and a failed rule comes with the fix",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"it reports every structural check the catalog applies, what it measured, and how to fix anything that failed\""+ },+ {+ "claim": "2 checks short",+ "kind": "evidence:proof",+ "target": "failed",+ "ok": true,+ "detail": "the run read failed = \"2\""+ },+ {+ "claim": "2 checks short",+ "kind": "evidence:proof",+ "target": "short of the catalog bar",+ "ok": true,+ "detail": "the run waited for and saw \"short of the catalog bar\""+ },+ {+ "claim": "no texture",+ "kind": "evidence:file",+ "target": "api/_lib/seed-mesh-gate.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "no texture",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"Untextured\""+ },+ {+ "claim": "22 meshes",+ "kind": "evidence:proof",+ "target": "meshes",+ "ok": true,+ "detail": "the run read meshes = \"22\""+ },+ {+ "claim": "at most 8",+ "kind": "evidence:proof",+ "target": "meshLimit",+ "ok": true,+ "detail": "the run read meshLimit = \"8\""+ },+ {+ "claim": "at most 8",+ "kind": "evidence:file",+ "target": "api/_lib/seed-mesh-gate.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "it will not rig or animate as one body",+ "kind": "evidence:file",+ "target": "api/_lib/seed-mesh-gate.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "Switch the category to Prop and the assembled-character rule stops applying, because loose parts are legitimate on a prop",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"Props skip the character-only checks\""+ },+ {+ "claim": "Switch the category to Prop and the assembled-character rule stops applying, because loose parts are legitimate on a prop",+ "kind": "evidence:file",+ "target": "api/_lib/seed-mesh-gate.js",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "Michelle from the library is a clean pass",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"A clean pass\""+ },+ {+ "claim": "Fox passes, barely",+ "kind": "evidence:page",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "live page shows \"Passes, barely\""+ },+ {+ "claim": "Copy any verdict as JSON or as a shareable link",+ "kind": "evidence:file",+ "target": "pages/inspect.html",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "claim": "Copy any verdict as JSON or as a shareable link",+ "kind": "evidence:file",+ "target": "pages/inspect.html",+ "ok": true,+ "detail": "file supports the claim"+ },+ {+ "kind": "probe:scenario",+ "target": "the filmed scenario",+ "ok": true,+ "detail": "17 steps passed against the live product, every claimed fact still holds (failed 2, meshes 22, meshLimit 8, vertices 5,196)"+ },+ {+ "kind": "probe:api",+ "target": "the inspector page is up",+ "ok": true,+ "detail": "HTTP 200, expectations met"+ },+ {+ "kind": "link",+ "target": "https://three.ws/inspect",+ "ok": true,+ "detail": "HTTP 200"+ }+ ]+ },+ "editor": {+ "verdict": "publish",+ "scores": {+ "accuracy": 5,+ "clarity": 5,+ "specificity": 5,+ "voice": 5,+ "professionalism": 5,+ "visual": 5+ },+ "issues": [],+ "rewrite": {+ "posts": [+ "Pick a model, or drop your own .glb, and the inspector runs the exact quality gate our catalog applies to its own entries: same module, same thresholds, same decision. It runs in your browser, and the file never leaves your machine.\n\nEvery rule shows what it measured against its bound, and a failed rule comes with the fix. The mannequin in the reel is 2 checks short: no texture, and 22 meshes where a character may have at most 8, so it will not rig or animate as one body.",+ "Switch the category to Prop and the assembled-character rule stops applying, because loose parts are legitimate on a prop. Michelle from the library is a clean pass. Fox passes, barely. Copy any verdict as JSON or as a shareable link. three.ws/inspect"+ ],+ "altText": []+ },+ "summary": "Every factual claim is verified word for word against the live page, the code, or the filmed run's screen readings (2 failed checks, 22 meshes, limit of 8), and the reel frames match each caption. Voice is clean throughout: no dashes, hashtags, or hype. Ship it unchanged.",+ "model": "nvidia:moonshotai/kimi-k3",+ "fallbacks": [+ "Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog",+ "openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr",+ "openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"+ ],+ "rewriteFindings": []+ },+ "editorError": null+}diff --git a/data/x-content/reviews/prompt-library.json b/data/x-content/reviews/prompt-library.jsonnew file mode 100644index 000000000..809ad6ef1--- /dev/null+++ b/data/x-content/reviews/prompt-library.json@@ -0,0 +1,191 @@+{+ "id": "prompt-library",+ "contentHash": "13e36e675d6cd864ea48d9c543056fefea10de68167c2b943aadb2e4945ebde7",+ "reviewedAt": "2026-10-01T07:11:43.184Z",+ "passed": true,+ "blockers": [],+ "texts": [+ "Pick what you want to make, copy the prompt, paste it into Claude, and swap the words in angle brackets for your own.\n\nThat is the whole Prompt Library: 24 prompts, each written to produce something real, like a 3D model you can download or a live avatar on your page. 15 of them need nothing installed.\n\n10 are for making 3D: a model from a sentence, a rigged avatar from a description, a rig for a GLB you already have.",+ "Each prompt names the tools to call and what to check before it answers. The rigged avatar one has Claude look at the rendered frames and fix the proportions before it says it is done.\n\nThe library is also published as plain text and as JSON, so an agent can read it too.\n\nthree.ws/prompts",+ "The prompts that can move real money are written to stop for your yes first."+ ],+ "lint": [],+ "verification": {+ "ok": true,+ "checks": [+ {+ "claim": "paste it into Claude",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"pasted straight into Claude\""+ },+ {+ "claim": "swap the words in angle brackets for your own",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"Placeholders in angle brackets are the parts you replace.\""+ },+ {+ "claim": "24 prompts",+ "kind": "evidence:proof",+ "target": "prompts",+ "ok": true,+ "detail": "the run read prompts = \"24\""+ },+ {+ "claim": "each written to produce something real",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"to produce something real\""+ },+ {+ "claim": "a 3D model you can download",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"a 3D model you can download\""+ },+ {+ "claim": "a live avatar on your page",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"Put a live avatar on your page\""+ },+ {+ "claim": "15 of them need nothing installed",+ "kind": "evidence:proof",+ "target": "nosetup",+ "ok": true,+ "detail": "the run read nosetup = \"15\""+ },+ {+ "claim": "15 of them need nothing installed",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"Nothing to install.\""+ },+ {+ "claim": "10 are for making 3D",+ "kind": "evidence:proof",+ "target": "making",+ "ok": true,+ "detail": "the run read making = \"10\""+ },+ {+ "claim": "a model from a sentence",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"Turn a sentence into a 3D model\""+ },+ {+ "claim": "a rigged avatar from a description",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"Generate a rigged avatar from a description\""+ },+ {+ "claim": "a rig for a GLB you already have",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"Rig a GLB you already have\""+ },+ {+ "claim": "Each prompt names the tools to call and what to check before it answers",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"Before you tell me it is done, call look_at_model on the result\""+ },+ {+ "claim": "look at the rendered frames",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"actually look at the frames\""+ },+ {+ "claim": "fix the proportions before it says it is done",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"If the silhouette or the proportions are wrong, fix it\""+ },+ {+ "claim": "published as plain text and as JSON",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"Plain text\""+ },+ {+ "claim": "published as plain text and as JSON",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts.json",+ "ok": true,+ "detail": "live page shows \"\"prompts\"\""+ },+ {+ "claim": "The prompts that can move real money are written to stop for your yes first",+ "kind": "evidence:page",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "live page shows \"These prompts can move real money, so they are written to stop for your yes first.\""+ },+ {+ "kind": "probe:scenario",+ "target": "the filmed scenario",+ "ok": true,+ "detail": "15 steps passed against the live product, every claimed fact still holds (nosetup 15, prompts 24, making 10)"+ },+ {+ "kind": "probe:api",+ "target": "the prompt library answers as JSON",+ "ok": true,+ "detail": "HTTP 200, expectations met"+ },+ {+ "kind": "link",+ "target": "https://three.ws/prompts",+ "ok": true,+ "detail": "HTTP 200"+ }+ ]+ },+ "editor": {+ "verdict": "publish",+ "scores": {+ "accuracy": 5,+ "clarity": 5,+ "specificity": 5,+ "voice": 5,+ "professionalism": 5,+ "visual": 5+ },+ "issues": [],+ "rewrite": {+ "posts": [+ "Pick what you want to make, copy the prompt, paste it into Claude, and swap the words in angle brackets for your own.\n\nThat is the whole Prompt Library: 24 prompts, each written to produce something real, like a 3D model you can download or a live avatar on your page. 15 of them need nothing installed.\n\n10 are for making 3D: a model from a sentence, a rigged avatar from a description, a rig for a GLB you already have.",+ "Each prompt names the tools to call and what to check before it answers. The rigged avatar one has Claude look at the rendered frames and fix the proportions before it says it is done.\n\nThe library is also published as plain text and as JSON, so an agent can read it too.\n\nthree.ws/prompts",+ "The prompts that can move real money are written to stop for your yes first."+ ],+ "altText": []+ },+ "summary": "Every claim passes verification against the live page and the filmed run, and the reel frames show exactly what the copy describes with matching numbers (24, 15, 10). The captions are true of their frames, the voice contract is followed, and the post stands alone as an explanation of the Prompt Library for a first-time reader.",+ "model": "nvidia:moonshotai/kimi-k3",+ "fallbacks": [+ "Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog",+ "openrouter 402: {\"error\":{\"message\":\"Insufficient credits. Add more using https://openrouter.ai/settings/credits\",\"code\":402,\"metadata\":{\"limit_source\":\"openrouter_credits\",\"remedy_hint\":\"Add credits at https://openr",+ "openai 429: {\n \"error\": {\n \"message\": \"Your account is not active, please check your billing details on our website.\",\n \"type\": \"billing_not_active\",\n \"param\": null,\n \"code\": \"billi"+ ],+ "rewriteFindings": []+ },+ "editorError": null+}diff --git a/data/x-content/reviews/splat-viewer.json b/data/x-content/reviews/splat-viewer.jsonindex c4efe1c2f..5615faabb 100644--- a/data/x-content/reviews/splat-viewer.json+++ b/data/x-content/reviews/splat-viewer.json@@ -1,15 +1,13 @@{"id": "splat-viewer",- "contentHash": "14b82c8faa27a36f7b18ebd422432277140b289d7b746c66f3ef861a3264c669",- "reviewedAt": "2026-09-30T08:42:30.481Z",- "passed": false,- "blockers": [- "editor verdict is revise"- ],+ "contentHash": "4e0a67768953428d183d4179f1b04c4627895dd881e07b7b5eb539554836a550",+ "reviewedAt": "2026-10-01T06:24:29.228Z",+ "passed": true,+ "blockers": [],"texts": ["A Gaussian splat rebuilds a person or a room as millions of tiny blobs instead of polygons. The splat viewer decodes one on your own device and draws it in a browser tab with WebGL, so the file stays with you.\n\nPaste a link to a capture or drop the file in. It reads five formats and tells them apart by their bytes, not their filenames.\n\nThe reel is the built-in sample: 14,000 splats, synthetic, generated on the page.","Both built-in samples are synthetic. They exist to exercise the renderer and are not real captures. A real one comes out of a capture engine trained on footage of a person, exported as a file, then opened here.\n\nthree.ws/splat",- "To share a capture, add ?src= and the address of the file to the address of the viewer. The scene loads as the page opens, so the link is the whole handoff.\n\nThe same scene that is a 1 GB PLY is roughly 100 MB as SPZ, which is the difference between a capture you can send as a link and one you cannot."+ "To share a capture, add ?src= and the address of the file to the address of the viewer. The scene loads as the page opens, so the link is the whole handoff.\n\nThe same scene that is a 1 GB PLY is roughly 100 MB as SPZ."],"lint": [],"verification": {@@ -124,7 +122,7 @@"kind": "probe:scenario","target": "the filmed scenario","ok": true,- "detail": "12 steps passed against the live product, facts unchanged (splats 14,000)"+ "detail": "12 steps passed against the live product, every claimed fact still holds (splats 14,000)"},{"kind": "probe:api",@@ -141,24 +139,16 @@]},"editor": {- "verdict": "revise",+ "verdict": "publish","scores": {- "accuracy": 4,- "clarity": 5,+ "accuracy": 5,+ "clarity": 4,"specificity": 5,"voice": 5,"professionalism": 5,"visual": 5},- "issues": [- {- "severity": "major",- "area": "accuracy",- "quote": "which is the difference between a capture you can send as a link and one you cannot",- "problem": "The verified evidence covers only the size comparison (a 1 GB PLY is roughly 100 MB as SPZ). The conclusion that file size determines whether a capture can be sent as a link is unverified and misleading: the previous sentence just explained that ?src= passing works for any address, so a 1 GB file is equally linkable. It overreads the docs quote and invites a pedantic correction from the developer audience.",- "fix": "End the sentence at the verified fact: \"The same scene that is a 1 GB PLY is roughly 100 MB as SPZ.\""- }- ],+ "issues": [],"rewrite": {"posts": ["A Gaussian splat rebuilds a person or a room as millions of tiny blobs instead of polygons. The splat viewer decodes one on your own device and draws it in a browser tab with WebGL, so the file stays with you.\n\nPaste a link to a capture or drop the file in. It reads five formats and tells them apart by their bytes, not their filenames.\n\nThe reel is the built-in sample: 14,000 splats, synthetic, generated on the page.",@@ -167,7 +157,7 @@],"altText": []},- "summary": "Every claim verifies against the live pages and the run, and the reel frames and captions match the copy including the 14,000 splats figure. The only fix is trimming the unverified and misleading tail of post 3, which implied file size determines whether a capture can be shared as a link.",+ "summary": "Every claim passes verification against the live pages, and the filmed run confirms the 14,000-splat figure shown on screen; the reel captions match their frames and the frames contradict nothing in the copy. Voice, single link, and mechanic-first structure all comply with the contract, so this ships unchanged.","model": "nvidia:moonshotai/kimi-k3","fallbacks": ["Vertex 403: {\n \"error\": {\n \"code\": 403,\n \"message\": \"Request had insufficient authentication scopes.\",\n \"status\": \"PERMISSION_DENIED\",\n \"details\": [\n {\n \"@type\": \"type.googleapis.com/goog",