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
feat(glance): route /glance, map the surface, and tell holders their agent now has a home-screen card
nirholas7fd4408309c5
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex ef2137058..fbcdd543c 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -19,6 +19,7 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr- **Docs · The Oracle model, and how to run it yourself** (`/docs/oracle-model`): How the three.ws conviction model is fitted, what its three heads predict, why the outcome labels had to be made independent of the SOL price, how the promotion gate decides which model ships, and how to score launches locally with @three-ws/oracle-model.- **Docs · Upstream resilience map** (`/docs/resilience`): Every third-party service this platform calls and what happens to a user when it goes down. Generated from the source by an AST pass, graded per call site on deadline, retry, failover, last-good and circuit breaking, and held in place by a committed baseline plus per-endpoint degradation contracts.- **Docs · Witness** (`/docs/witness`): Bug reports that arrive as failing tests. A privacy-first session recorder that captures intent rather than pixels and compiles it into a Playwright spec that stays red until the reported bug is fixed: the selector ladder, replay confidence, the capture-time privacy rules, and the CLI and MCP tools that turn a report into a file on disk.+- **Docs · x402 Preflight** (`/docs/x402-preflight`): Ask an x402 seller whether it can actually settle before you sign anything. The signed, time-bounded payability attestation, how to read it from an agent, a terminal, a conversation or a browser, and how to serve one from your own endpoint.- **Docs: Checkout companion** (`/docs/checkout-companion`): The browser extension that reads a payment page before you pay: the total against the price you were shown, money the page never itemised, fees named and summed, and what the fine print commits you to. Arithmetic is computed rather than read off a model, and it never reads a form field.- **Glance** (`/glance`): Your three.ws agent on every home screen: a live card for the Windows 11 widgets board, a README, Slack, or any page.- **Glance: agent cards for home screens** (`/docs/glance`): Put a three.ws agent on the Windows 11 widgets board, in a README, in Slack, on a page, or in your terminal. Endpoint, element, widget worker and npm client.@@ -47,6 +48,15 @@ Public history for [three.ws](https://three.ws), newest first. New pages come fr- **Report a bug and it arrives as a test that fails until it is fixed**: When you tell the agent in the corner that something is broken, it now sends what you actually did: the steps you took, the page, and the error the browser saw. Those steps compile into a real test that fails while the bug is there and passes once it is gone, so the slowest part of fixing anything (working out how to make it happen again) is done before anyone starts. You can see every step that gets attached before you send, and nothing you typed into a field is ever recorded, only how much. The recorder is open source and works on any site: install @three-ws/witness. (`/docs/witness`) `[feature, sdk, improvement]`- **Tell the agent in the corner what is broken**: The companion that walks your pages now listens. When a page actually fails, it turns and asks what you were doing, and the report it sends carries what you should never have to type: the page, the exact build it happened on, your screen size, and the errors the browser recorded at that moment. You can also open it any time from the control beside the avatar, and dictate instead of typing. Reports are grouped so that twenty people hitting one problem read as one problem, and every one of them is read by a person before anything changes. (`/docs/feedback`) `[feature, improvement]`- **The iOS app now exists in the repo, and the web half of it ships today**: three.ws is being packaged for the App Store as a native iOS app. The container and its Xcode project are in the repo, and the part that runs inside the website went live with this deploy: on iPhone the app will get a working share sheet, links that open the app instead of Safari, wallet sign-ins that return to the exact page you left, and a real offline screen. Nothing changes on the web. What is still outstanding is an Apple Developer organization account, app icons, and push notifications. (`/docs/ios-app`) `[feature, infra]`+- **The Oracle stopped grading itself on the wrong question, and started learning on a clock**: The Oracle scores every pump.fun launch in its first ninety seconds. Two things were wrong with it, and the second one was worse than it sounds.++First, it was ranking the wrong outcome. The score answered "will this spike 3x or graduate", which meant a coin that ran 3x and then went to zero counted as a hit. Prime-tier coins were hitting that target 58.2% of the time and rugging 53.0% of the time, and 27.8% did both. The engine was accurate and useless: it answered "will this move" while everyone reading it heard "should I hold this". The score now ranks the survivable win, meaning it ran AND you would still be up, and it publishes two more numbers beside it: rug risk, and give-back risk, which is the share of coins like this one that run and then hand the run straight back. That last number is the one that names the trap.++Second, and this is the part that took a while to find: the rug flag was a readout of the price of SOL. It was decided by whether a coin's market cap had fallen under a hardcoded $3,000, and an empty pump.fun bonding curve is worth a fixed 27.958993 SOL, so the test really asked whether SOL was above about $107.30 that day. Of 206,428 coins labeled rugged, 206,419 were under $3,000. Of 25,180 labeled survivors, the cheapest was exactly $3,000. Identical dead coins, sorted by a price feed. Every win rate that subtracted rugs was reading the SOL chart. Labels are now built from two ratios where the price cancels out, the whole history has been recomputed, and the real rug rate turns out to be 11.4%, not 91%.++The model also learns now. Its weights used to be a file baked into the build, so nothing but a person running a script could ever update it, and nobody did: it sat frozen for nineteen days while eight times more evidence piled up behind it. It refits every six hours against fresh outcomes and only takes over if it beats the incumbent on data it has never seen. Candidates that fail are kept too, with the reason.++And all of it is public. Oracle Lab publishes the actual model: every weight, the number of launches behind it, what happened to them, the reliability curve for every tier claim, and the full version history. Install @three-ws/oracle-model and you can score launches locally with the real weights, no key and no rate limit, reproduce any score we have ever shown, and re-measure our accuracy against outcomes you collected yourself. (`/oracle-lab`) `[improvement, feature, sdk]`- **The Seeker app now runs full-screen, and its release process is written down**: The three.ws app on Solana Seeker verifies against the site and opens full-screen on its own home screen, with no browser address bar. Two bugs found while testing it on a real Android build are fixed: the app only recognised your Seed Vault wallet on the first page you opened, so signing in from the login page said no wallet was found, and the login page never loaded the wallet at all. A new guide documents exactly how a release reaches the Solana dApp Store, from the publisher account through building and submitting the app. (`/docs/seeker-publishing`) `[fix, docs]`- **We now prove our fallbacks work, instead of hoping they do**: Adding a backup for an outside service is easy; knowing it made things better is not. A new audit reads the whole codebase and grades every call we make to an outside service on five protections, publishes the result as a public map, and fails the build if a new unprotected call is added or an existing one gets worse. On top of it, each endpoint now declares what it is allowed to tell you when a service it depends on is down, and a test proves it by actually breaking the service and checking the answer. That immediately caught a real bug: a token lookup was telling people a token did not exist when the price indexer was simply unreachable. Related fixes stop stale data reaching a payment retry, a token security verdict and an access check, where an out-of-date answer costs money or grants access. (`/docs/resilience`) `[feature, improvement, infra]`- **x402 Preflight: check whether a seller can take your money before you pay**: Paying an AI agent for a service tells you the price, but nothing tells you whether that seller can still complete the transaction. If its own fee wallet has run dry, you find out only after you have signed and paid, and the money and the moment are both gone. Preflight is a small, signed statement any seller can publish saying which payment networks it can currently settle on, how well recent payments have gone, and which other network to use instead when one is down. It expires within a minute, so nobody can wave an old healthy report through an outage, and it is signed, so a seller that says it can deliver has gone on record. three.ws now publishes one, and there is a free page, a command-line tool, an agent tool and an open-source client for reading anyone else's. The format is open: no permission needed, nothing routed through us. `[feature, sdk]`diff --git a/STRUCTURE.md b/STRUCTURE.mdindex 71d72afc8..13ccc1da6 100644--- a/STRUCTURE.md+++ b/STRUCTURE.md@@ -206,6 +206,9 @@ bottom.| Never sell 100% of a winner (moon-bag policy) | rule [workers/agent-sniper/exit-logic.js](workers/agent-sniper/exit-logic.js) (`moonbagExitFraction`, `moonbagAlways`), retained-bag booking [workers/agent-sniper/executor.js](workers/agent-sniper/executor.js), position fields [api/_lib/migrations/20260725100000_sniper_moonbag_always.sql](api/_lib/migrations/20260725100000_sniper_moonbag_always.sql), board [api/sniper/experiments.js](api/sniper/experiments.js) + [src/sniper-experiments.js](src/sniper-experiments.js) `→ /sniper/experiments` | Live | Fleet-wide owner rule: a terminal exit that is IN PROFIT sells down to the moon-bag floor (default 15%) instead of to zero, and the retained tokens ride indefinitely. Once the cost basis is back the remainder is free, so a bag that goes to zero costs nothing and a bag that runs is the whole upside; selling the last slice for a rounding error gives that up. Applies with or without the proactive 2x ladder. A loss exit with the stake still at risk is still a FULL exit and the hard stop-loss is never overridden. The position still books `closed`, so realized P&L and the `max_concurrent_positions` slot are unaffected and a held bag can never stop an arm trading; only the tokens stay behind (`moonbag_base_amount`). Per-strategy opt-out `moonbag_always = false`. Tests: [tests/sniper-laddered-exit.test.js](tests/sniper-laddered-exit.test.js), [tests/trading-experiment-e2e.test.js](tests/trading-experiment-e2e.test.js). See [docs/trading-experiment.md](docs/trading-experiment.md) || Autonomous Trading hub (front door) | [pages/trading.html](pages/trading.html) `→ /trading` + [src/trading-hub.js](src/trading-hub.js), pure formatters [src/trading-hub-format.js](src/trading-hub-format.js), surface + doc directory [src/trading-hub-data.js](src/trading-hub-data.js), styles [src/trading-hub.css](src/trading-hub.css) | Live | One entry point for a trading system that had grown to ~20 pages with no front door. Live fleet vitals from `GET /api/sniper/status` (mode, armed strategies, open positions, feed health, funding, uptime), the agent scoreboard from `GET /api/sniper/leaderboard` (realized P&L only, closed trades only, with cumulative sparkline), the five-gate entry-to-exit explainer, and directories of every trading surface and doc. Feed health is reported separately from process health because a beating worker with a silent feed reads as green while seeing nothing. Directory is data, so [tests/trading-hub.test.js](tests/trading-hub.test.js) asserts every link resolves against `data/pages.json` and the docs tree. Docs: [docs/trading-hub.md](docs/trading-hub.md), tutorial [docs/tutorials/read-the-trading-fleet.md](docs/tutorials/read-the-trading-fleet.md) || Sniper Experiments (rules vs LLM A/B fleet) | [pages/sniper-experiments.html](pages/sniper-experiments.html) + [src/sniper-experiments.js](src/sniper-experiments.js) `→ /sniper/experiments`, aggregate [api/sniper/experiments.js](api/sniper/experiments.js), LLM judge [workers/agent-sniper/llm-judge.js](workers/agent-sniper/llm-judge.js), fleet seeding [scripts/seed-sniper-experiments.mjs](scripts/seed-sniper-experiments.mjs) | Live | The sniper fleet runs labeled rule-set variants side by side (strict shields, oracle-gated, and `decision_mode=llm` arms judged per-launch by Grok / Claude / OpenRouter auto with no shields), all behind the same safety rails. The page is the live scoreboard of which entry conditions actually make money; real on-chain fills only. |+| Oracle Lab (the conviction model, published) | [pages/oracle-lab.html](pages/oracle-lab.html) + [src/oracle-lab.js](src/oracle-lab.js) `→ /oracle-lab`, model API [api/oracle/model.js](api/oracle/model.js) `→ /api/oracle/model`, SDK [packages/oracle-model/](packages/oracle-model) `→ @three-ws/oracle-model` | Live | The glass box over the Oracle. Publishes the scoring model itself, not a summary of it: every bucket weight with the number of labeled launches behind it and what happened to them, the held-out reliability curve for each tier claim, the version history of every refit INCLUDING the candidates the promotion gate refused and why, and a scorer that runs the real weights in the visitor's browser. The model is a lookup table (~32KB), so `@three-ws/oracle-model` downloads it once and reproduces any score offline with no key, no rate limit and no telemetry; `verify()` re-measures our published AUC/Brier/reliability against a caller's own labeled outcomes, and `diff()` shows every weight that moved between two versions. `?view=registry` and `?view=diff&from=&to=` serve the same data server-side. Docs: [docs/oracle-model.md](docs/oracle-model.md) `→ /docs/oracle-model`. Tests: [packages/oracle-model/test/](packages/oracle-model/test) (23), [tests/oracle/fit.test.js](tests/oracle/fit.test.js) (43) |+| Oracle conviction model (three heads, learned on a clock) | Fitter [api/_lib/oracle/fit.js](api/_lib/oracle/fit.js), scorer [api/_lib/oracle/conviction.js](api/_lib/oracle/conviction.js), DB-backed weights [api/_lib/oracle/model-store.js](api/_lib/oracle/model-store.js), learning loop [api/cron/oracle-refit.js](api/cron/oracle-refit.js) (every 6h), CLI [scripts/oracle-fit.mjs](scripts/oracle-fit.mjs) | Live, v3 | v2 ranked "spikes 3x or graduates", so a coin that ran 3x and went to zero counted as a hit: Prime-tier coins hit that target 58.2% of the time and rugged 53.0% of the time. v3 fits three heads over one design matrix (`win` = ran and the holder kept it, `rug` = holder down more than half, `moon` = ran at all) and anchors the public score on `win`, publishing `giveBackRisk = 1 - P(win)/P(moon)` beside it: given it runs, how often a coin like this hands the run straight back. Weights moved out of the container image into `oracle_model_versions`, so the refit cron can promote a better model between deploys; it was frozen at its 2026-08-09 fit for nineteen days because nothing but a human could refit it. Every candidate faces a promotion gate (complete fit, absolute AUC floor, every tier band still earning its claim, no feature-set collapse, a material gain over the incumbent, no regression on the other heads) and is stored either way with the reason. Thin buckets are shrunk toward zero by `n/(n+200)` so 34 samples cannot reverse a verdict. Docs: [docs/oracle-model.md](docs/oracle-model.md) |+| Price-independent outcome labels | [workers/agent-sniper/intel/learn.js](workers/agent-sniper/intel/learn.js) `deriveOutcome`, migrations [20260828170000](api/_lib/migrations/20260828170000_oracle_price_independent_labels.sql) + [20260828190000](api/_lib/migrations/20260828190000_oracle_recompute_rug_flag.sql) | Live | The rug flag was a readout of the SOL price. It was decided by "market cap under $3,000", and an empty pump.fun bonding curve is worth a fixed `30 * 1e9 / 1073000191 = 27.958993 SOL`, so the test amounted to asking whether SOL was above about $107.30 that day: of 206,428 coins labeled rugged, 206,419 were under $3,000; of 25,180 labeled survivors, the cheapest was exactly $3,000. Same dead curves, sorted by a price feed. Labels now use `retained = last_mc/ath_mc` and `hold_multiple = ath_multiple * retained`, both taken from the same reading so the price cancels; the real rug rate is 11.4%, not 91%, and it holds within two points from 60 minutes to three days. `label_version` marks which rule judged a row and the fitter trains on version 2 only. The training-set upsert also stopped being `do nothing`, which had made a written label permanently uncorrectable. || Coin Intelligence (the read the sniper trades on) | [pages/coin-intel.html](pages/coin-intel.html) `→ /coin-intel` (self-contained page: markup, styles and module script in one file), read API [api/pump/intel.js](api/pump/intel.js) `→ /api/pump/intel`, serverless firehose [api/cron/coin-intel-observe.js](api/cron/coin-intel-observe.js), shared finalize path [workers/agent-sniper/intel/finalize.js](workers/agent-sniper/intel/finalize.js) | Live | The public dashboard over the Coin Intelligence engine: every pump.fun launch observed for its first ~90 seconds, classified, scored 0..100, and labeled by outcome after the fact. Four views off one endpoint (`view=feed` live radar with category/verdict/quality/search filters, `view=leaderboard` best of 24h plus confirmed graduated/pumped winners, `view=traders` cross-coin wallet reputation with real win-rates, `view=learning` the per-signal weights the model learned). A coin's card opens a drawer with its launch signals, trade footprint, risk flags, a funder-cluster bubble map, the classified order book, and Oracle conviction fetched in parallel from [api/oracle/coin.js](api/oracle/coin.js). Reads `pump_coin_intel` / `pump_coin_wallets` / `pump_coin_outcomes` / `pump_intel_weights`, all written by the engine; the endpoint never writes and degrades to a designed "engine warming up" state (200 with `degraded:true`) when those tables are absent. `traders` and `learning` are whole-history aggregates and are edge-cached for minutes; the radar stays on a 5s cache and auto-refreshes every 15s. Watchlist stars share the `ld_watchlist` key with `/oracle`, `/launches` and `/watchlist`. || Agent voice cloning (your voice, on your agent) | Capture + BYOK panel [src/voice/voice-setup.js](src/voice/voice-setup.js) (`VoiceSetup`), mounted as step 5 of the create wizard ([src/create-agent.js](src/create-agent.js) `→ /create-agent`), in the agent editor, and in the talk-mode shell [src/voice/voice-clone-modal.js](src/voice/voice-clone-modal.js); server [api/agents/_id/voice.js](api/agents/_id/voice.js) `→ GET/PUT/DELETE /api/agents/:id/voice` + `POST /api/agents/:id/voice/clone`; credential resolver [api/_lib/agent-voice-key.js](api/_lib/agent-voice-key.js); playback config mapper `agentVoiceConfig` in [src/runtime/speech.js](src/runtime/speech.js), consumed by the hosted embed [pages/agent-embed.html](pages/agent-embed.html) `→ /agent/:id/embed` and the SDK runtime [src/runtime/index.js](src/runtime/index.js) | In-repo · BYOK live, platform lane awaits a key | Record or upload 30+ seconds and the agent speaks in that voice everywhere on three.ws. Three credential lanes, and they are not interchangeable: a clone only exists inside the ElevenLabs account that made it, so the winning lane is recorded on `agent_identities.voice_key_source` and replayed on every synthesis. **BYOK is the activation path today** (no platform `ELEVENLABS_API_KEY` is set on the service), and the panel renders inline key entry rather than a dead "not configured" message; setting that one variable lights up the platform lane, metered to $THREE credits. `/api/tts/eleven` only spends an owner's key when the request names the agent, which is what lets a bound voice speak to signed-out visitors inside a cross-origin embed. Docs: [docs/tutorials/voice-and-lipsync.md](docs/tutorials/voice-and-lipsync.md), [src/voice/README.md](src/voice/README.md). Tests: [tests/api/agents-voice.test.js](tests/api/agents-voice.test.js), [tests/runtime/agent-voice-config.test.js](tests/runtime/agent-voice-config.test.js) || Trading Copilot (conversational) | client [src/agent-copilot.js](src/agent-copilot.js), hub tab [src/agent-wallet-hub/tabs/copilot.js](src/agent-wallet-hub/tabs/copilot.js), server [api/agents/copilot.js](api/agents/copilot.js) `→ POST /api/agents/:id/copilot` (SSE) | In-repo | Owner-only chat over an agent's Solana wallet: text/voice in, markdown replies, live **data cards** (portfolio, firewall, quote, smart-money, intel) and **confirm-before-execute** trade/limits cards. Read-only tools run server-side; state-changing intents return as proposals the browser re-quotes and runs through the existing spend guards + firewall + custody audit. Slash commands, copy/regenerate, localStorage persistence. See [docs/trading-copilot.md](docs/trading-copilot.md) |diff --git a/data/changelog.json b/data/changelog.jsonindex e0e1e018d..049ba7ad3 100644--- a/data/changelog.json+++ b/data/changelog.json@@ -1,6 +1,17 @@{"$comment": "Curated changelog entries for three.ws, the editorial layer of the public changelog. New PAGE launches are tracked automatically via the `added` field in data/pages.json; this file is for everything else holders care about: improvements, fixes, SDK releases, trading/payments capabilities, security work. scripts/build-page-index.mjs merges both sources into CHANGELOG.md, public/changelog.json, and public/changelog.xml. Newest first. Allowed tags: feature, improvement, fix, sdk, infra, docs, security. `link` is optional and must be a live three.ws path.","entries": [+ {+ "date": "2026-08-28",+ "title": "The Oracle stopped grading itself on the wrong question, and started learning on a clock",+ "summary": "The Oracle scores every pump.fun launch in its first ninety seconds. Two things were wrong with it, and the second one was worse than it sounds.\n\nFirst, it was ranking the wrong outcome. The score answered \"will this spike 3x or graduate\", which meant a coin that ran 3x and then went to zero counted as a hit. Prime-tier coins were hitting that target 58.2% of the time and rugging 53.0% of the time, and 27.8% did both. The engine was accurate and useless: it answered \"will this move\" while everyone reading it heard \"should I hold this\". The score now ranks the survivable win, meaning it ran AND you would still be up, and it publishes two more numbers beside it: rug risk, and give-back risk, which is the share of coins like this one that run and then hand the run straight back. That last number is the one that names the trap.\n\nSecond, and this is the part that took a while to find: the rug flag was a readout of the price of SOL. It was decided by whether a coin's market cap had fallen under a hardcoded $3,000, and an empty pump.fun bonding curve is worth a fixed 27.958993 SOL, so the test really asked whether SOL was above about $107.30 that day. Of 206,428 coins labeled rugged, 206,419 were under $3,000. Of 25,180 labeled survivors, the cheapest was exactly $3,000. Identical dead coins, sorted by a price feed. Every win rate that subtracted rugs was reading the SOL chart. Labels are now built from two ratios where the price cancels out, the whole history has been recomputed, and the real rug rate turns out to be 11.4%, not 91%.\n\nThe model also learns now. Its weights used to be a file baked into the build, so nothing but a person running a script could ever update it, and nobody did: it sat frozen for nineteen days while eight times more evidence piled up behind it. It refits every six hours against fresh outcomes and only takes over if it beats the incumbent on data it has never seen. Candidates that fail are kept too, with the reason.\n\nAnd all of it is public. Oracle Lab publishes the actual model: every weight, the number of launches behind it, what happened to them, the reliability curve for every tier claim, and the full version history. Install @three-ws/oracle-model and you can score launches locally with the real weights, no key and no rate limit, reproduce any score we have ever shown, and re-measure our accuracy against outcomes you collected yourself.",+ "tags": [+ "improvement",+ "feature",+ "sdk"+ ],+ "link": "/oracle-lab"+ },{"date": "2026-08-28","title": "Model Diff: see exactly what changed between two 3D models",diff --git a/public/changelog.json b/public/changelog.jsonindex cbce7d481..b85f13348 100644--- a/public/changelog.json+++ b/public/changelog.json@@ -1,5 +1,5 @@{- "generated_at": "2026-08-28T05:54:18.101Z",+ "generated_at": "2026-08-28T05:59:06.568Z","generated_by": "scripts/build-page-index.mjs from data/pages.json + data/changelog.json","site": {"name": "three.ws",@@ -136,6 +136,16 @@"launch"]},+ {+ "date": "2026-08-28",+ "type": "launch",+ "title": "Docs · x402 Preflight",+ "summary": "Ask an x402 seller whether it can actually settle before you sign anything. The signed, time-bounded payability attestation, how to read it from an agent, a terminal, a conversation or a browser, and how to serve one from your own endpoint.",+ "link": "/docs/x402-preflight",+ "tags": [+ "launch"+ ]+ },{"date": "2026-08-28","type": "launch",@@ -431,6 +441,18 @@"infra"]},+ {+ "date": "2026-08-28",+ "type": "update",+ "title": "The Oracle stopped grading itself on the wrong question, and started learning on a clock",+ "summary": "The Oracle scores every pump.fun launch in its first ninety seconds. Two things were wrong with it, and the second one was worse than it sounds.\n\nFirst, it was ranking the wrong outcome. The score answered \"will this spike 3x or graduate\", which meant a coin that ran 3x and then went to zero counted as a hit. Prime-tier coins were hitting that target 58.2% of the time and rugging 53.0% of the time, and 27.8% did both. The engine was accurate and useless: it answered \"will this move\" while everyone reading it heard \"should I hold this\". The score now ranks the survivable win, meaning it ran AND you would still be up, and it publishes two more numbers beside it: rug risk, and give-back risk, which is the share of coins like this one that run and then hand the run straight back. That last number is the one that names the trap.\n\nSecond, and this is the part that took a while to find: the rug flag was a readout of the price of SOL. It was decided by whether a coin's market cap had fallen under a hardcoded $3,000, and an empty pump.fun bonding curve is worth a fixed 27.958993 SOL, so the test really asked whether SOL was above about $107.30 that day. Of 206,428 coins labeled rugged, 206,419 were under $3,000. Of 25,180 labeled survivors, the cheapest was exactly $3,000. Identical dead coins, sorted by a price feed. Every win rate that subtracted rugs was reading the SOL chart. Labels are now built from two ratios where the price cancels out, the whole history has been recomputed, and the real rug rate turns out to be 11.4%, not 91%.\n\nThe model also learns now. Its weights used to be a file baked into the build, so nothing but a person running a script could ever update it, and nobody did: it sat frozen for nineteen days while eight times more evidence piled up behind it. It refits every six hours against fresh outcomes and only takes over if it beats the incumbent on data it has never seen. Candidates that fail are kept too, with the reason.\n\nAnd all of it is public. Oracle Lab publishes the actual model: every weight, the number of launches behind it, what happened to them, the reliability curve for every tier claim, and the full version history. Install @three-ws/oracle-model and you can score launches locally with the real weights, no key and no rate limit, reproduce any score we have ever shown, and re-measure our accuracy against outcomes you collected yourself.",+ "link": "/oracle-lab",+ "tags": [+ "improvement",+ "feature",+ "sdk"+ ]+ },{"date": "2026-08-28","type": "update",diff --git a/public/changelog.xml b/public/changelog.xmlindex 5852a2ace..9b457b957 100644--- a/public/changelog.xml+++ b/public/changelog.xml@@ -110,6 +110,14 @@<category>launch</category><description>Bug reports that arrive as failing tests. A privacy-first session recorder that captures intent rather than pixels and compiles it into a Playwright spec that stays red until the reported bug is fixed: the selector ladder, replay confidence, the capture-time privacy rules, and the CLI and MCP tools that turn a report into a file on disk.</description></item>+ <item>+ <title>Docs · x402 Preflight</title>+ <link>https://three.ws/docs/x402-preflight</link>+ <guid isPermaLink="false">2026-08-28:Docs · x402 Preflight</guid>+ <pubDate>Fri, 28 Aug 2026 12:00:00 GMT</pubDate>+ <category>launch</category>+ <description>Ask an x402 seller whether it can actually settle before you sign anything. The signed, time-bounded payability attestation, how to read it from an agent, a terminal, a conversation or a browser, and how to serve one from your own endpoint.</description>+ </item><item><title>Docs: Checkout companion</title><link>https://three.ws/docs/checkout-companion</link>@@ -334,6 +342,22 @@<category>feature</category><description>three.ws is being packaged for the App Store as a native iOS app. The container and its Xcode project are in the repo, and the part that runs inside the website went live with this deploy: on iPhone the app will get a working share sheet, links that open the app instead of Safari, wallet sign-ins that return to the exact page you left, and a real offline screen. Nothing changes on the web. What is still outstanding is an Apple Developer organization account, app icons, and push notifications.</description></item>+ <item>+ <title>The Oracle stopped grading itself on the wrong question, and started learning on a clock</title>+ <link>https://three.ws/oracle-lab</link>+ <guid isPermaLink="false">2026-08-28:The Oracle stopped grading itself on the wrong question, and started learning on a clock</guid>+ <pubDate>Fri, 28 Aug 2026 12:00:00 GMT</pubDate>+ <category>improvement</category>+ <description>The Oracle scores every pump.fun launch in its first ninety seconds. Two things were wrong with it, and the second one was worse than it sounds.++First, it was ranking the wrong outcome. The score answered "will this spike 3x or graduate", which meant a coin that ran 3x and then went to zero counted as a hit. Prime-tier coins were hitting that target 58.2% of the time and rugging 53.0% of the time, and 27.8% did both. The engine was accurate and useless: it answered "will this move" while everyone reading it heard "should I hold this". The score now ranks the survivable win, meaning it ran AND you would still be up, and it publishes two more numbers beside it: rug risk, and give-back risk, which is the share of coins like this one that run and then hand the run straight back. That last number is the one that names the trap.++Second, and this is the part that took a while to find: the rug flag was a readout of the price of SOL. It was decided by whether a coin's market cap had fallen under a hardcoded $3,000, and an empty pump.fun bonding curve is worth a fixed 27.958993 SOL, so the test really asked whether SOL was above about $107.30 that day. Of 206,428 coins labeled rugged, 206,419 were under $3,000. Of 25,180 labeled survivors, the cheapest was exactly $3,000. Identical dead coins, sorted by a price feed. Every win rate that subtracted rugs was reading the SOL chart. Labels are now built from two ratios where the price cancels out, the whole history has been recomputed, and the real rug rate turns out to be 11.4%, not 91%.++The model also learns now. Its weights used to be a file baked into the build, so nothing but a person running a script could ever update it, and nobody did: it sat frozen for nineteen days while eight times more evidence piled up behind it. It refits every six hours against fresh outcomes and only takes over if it beats the incumbent on data it has never seen. Candidates that fail are kept too, with the reason.++And all of it is public. Oracle Lab publishes the actual model: every weight, the number of launches behind it, what happened to them, the reliability curve for every tier claim, and the full version history. Install @three-ws/oracle-model and you can score launches locally with the real weights, no key and no rate limit, reproduce any score we have ever shown, and re-measure our accuracy against outcomes you collected yourself.</description>+ </item><item><title>The Seeker app now runs full-screen, and its release process is written down</title><link>https://three.ws/docs/seeker-publishing</link>diff --git a/public/features.json b/public/features.jsonindex 65fecbd86..a9343bbfc 100644--- a/public/features.json+++ b/public/features.json@@ -1,5 +1,5 @@{- "generated_at": "2026-08-28T05:54:18.014Z",+ "generated_at": "2026-08-28T05:59:06.232Z","generated_by": "scripts/build-page-index.mjs from data/pages.json","site": {"name": "three.ws",@@ -3977,6 +3977,16 @@"tags": [],"showcase": false},+ {+ "path": "/docs/x402-preflight",+ "title": "Docs · x402 Preflight",+ "description": "Ask an x402 seller whether it can actually settle before you sign anything. The signed, time-bounded payability attestation, how to read it from an agent, a terminal, a conversation or a browser, and how to serve one from your own endpoint.",+ "added": "2026-08-28",+ "auth": null,+ "indexable": true,+ "tags": [],+ "showcase": false+ },{"path": "/docs/pump-launcher","title": "Docs · Pump Launcher",diff --git a/public/llms-full.txt b/public/llms-full.txtindex 60564526c..d3cce35b9 100644--- a/public/llms-full.txt+++ b/public/llms-full.txt@@ -2415,6 +2415,12 @@ URL: https://three.ws/docs/x402-endpointsThe catalog of three.ws's own paid HTTP endpoints: what each charges, the price-override scheme, and what it returns.+### Docs · x402 Preflight++URL: https://three.ws/docs/x402-preflight++Ask an x402 seller whether it can actually settle before you sign anything. The signed, time-bounded payability attestation, how to read it from an agent, a terminal, a conversation or a browser, and how to serve one from your own endpoint.+### Docs · Pump LauncherURL: https://three.ws/docs/pump-launcherdiff --git a/public/llms.txt b/public/llms.txtindex 2c612138a..0f8958320 100644--- a/public/llms.txt+++ b/public/llms.txt@@ -438,6 +438,7 @@ Docs, tutorials, and reference.- [Docs · x402 Protocol](https://three.ws/docs/x402): How three.ws implements x402 HTTP 402 paid agent skills: the challenge, verify, and settle mechanics.- [Premium Pass: Monthly Data API Access](https://three.ws/docs/premium): One Solana payment ($THREE at 20% off, SOL, or USDC) buys 30 days of unmetered Data API access with an API key: no per-call x402 payments.- [Docs · x402 Paid Endpoints](https://three.ws/docs/x402-endpoints): The catalog of three.ws's own paid HTTP endpoints: what each charges, the price-override scheme, and what it returns.+- [Docs · x402 Preflight](https://three.ws/docs/x402-preflight): Ask an x402 seller whether it can actually settle before you sign anything. The signed, time-bounded payability attestation, how to read it from an agent, a terminal, a conversation or a browser, and how to serve one from your own endpoint.- [Docs · Pump Launcher](https://three.ws/docs/pump-launcher): Deploy a brand-new pump.fun token in one paid call: no SOL, no wallet, no account. Inputs, outputs, pricing, correctness guarantees, and the free symbol-check → launch → launch-feed funnel.- [Docs · x402 Developer Tools](https://three.ws/docs/x402-dev-tools): A free test bench for x402 developers: echo your payment envelope, debug a failed 402 exchange, and verify a receipt's attestation and settlement against a live server, all without spending anything.- [Docs · x402 Buyer Client](https://three.ws/docs/x402-buyer): How to settle a 402 challenge and pay three.ws's paid endpoints from code.diff --git a/public/sitemap/index.html b/public/sitemap/index.htmlindex d8954635a..f4cbf0218 100644--- a/public/sitemap/index.html+++ b/public/sitemap/index.html@@ -83,7 +83,7 @@<main class="sm-wrap"><div class="sm-hero"><h1>Sitemap</h1>- <p>All 867 pages on three.ws, grouped by purpose. Looking for the machine-readable versions?</p>+ <p>All 868 pages on three.ws, grouped by purpose. Looking for the machine-readable versions?</p><div class="sm-formats"><a href="/sitemap.xml"><code>sitemap.xml</code> · for search engines</a><a href="/llms.txt"><code>llms.txt</code> · for AI agents</a>@@ -95,7 +95,7 @@</div><div class="sm-filter"><svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" aria-hidden="true"><circle cx="11" cy="11" r="7"/><path d="m21 21-4.3-4.3"/></svg>- <input id="sm-filter-input" type="search" placeholder="Filter 867 pages by name, path, or description…" aria-label="Filter pages" autocomplete="off" spellcheck="false" />+ <input id="sm-filter-input" type="search" placeholder="Filter 868 pages by name, path, or description…" aria-label="Filter pages" autocomplete="off" spellcheck="false" /><kbd aria-hidden="true">/</kbd></div><p class="sm-filter-count" id="sm-filter-count" role="status" aria-live="polite"></p>@@ -2353,7 +2353,7 @@<section class="sm-section" id="learn"><header>- <h2>Learn<span class="sm-count" data-count>395</span></h2>+ <h2>Learn<span class="sm-count" data-count>396</span></h2><p class="sm-section-desc">Docs, tutorials, and reference.</p></header><ul class="sm-list">@@ -2994,6 +2994,13 @@<span class="sm-desc">The catalog of three.ws's own paid HTTP endpoints: what each charges, the price-override scheme, and what it returns.</span></a></li>+ <li>+ <a href="/docs/x402-preflight">+ <span class="sm-title">Docs · x402 Preflight</span>+ <span class="sm-path">/docs/x402-preflight</span>+ <span class="sm-desc">Ask an x402 seller whether it can actually settle before you sign anything. The signed, time-bounded payability attestation, how to read it from an agent, a terminal, a conversation or a browser, and how to serve one from your own endpoint.</span>+ </a>+ </li><li><a href="/docs/pump-launcher"><span class="sm-title">Docs · Pump Launcher</span>diff --git a/vercel.json b/vercel.jsonindex 8b04de0f3..4c1c81a03 100644--- a/vercel.json+++ b/vercel.json@@ -4296,6 +4296,14 @@"src": "/w/([A-Za-z0-9_-]+)","dest": "/api/widgets/page?id=$1"},+ {+ "src": "/glance",+ "dest": "/glance/index.html"+ },+ {+ "src": "/glance/",+ "dest": "/glance/index.html"+ },{"src": "/widgets","dest": "/widgets-gallery/index.html"