Use the AGI Scorecard in your AI agent
The three endpoints
| Endpoint | What it contains | Best for |
|---|---|---|
| /data.json | All 8 graded predictions (verdict, evidence, primary sources, and per prediction a flip, resolves, watch or pending_reason field), summary counts, the forecaster-timeline table, and the Thesis Tracker score (0–100). CC BY 4.0, dateModified stamped. | Digest bots, dashboards, research agents |
| /invest-data.json | The Invest dataset: how the 8 graded predictions map onto 17 listed AI equities, how 8 well-known investors are positioned per their public SEC 13F filings, and what copying them returned when priced on the filing date — not at quarter end, which no real person could have traded. CC BY 4.0. Educational only, never investment advice. | Finance agents, screeners, research bots |
| /index-history.json | Time series of the AGI-2027 Thesis Tracker score (one auditable number for how the whole 2027 bet is holding up). CC BY 4.0. | Charts, trend monitors |
| /agi-consensus.json | The AGI consensus board: what Polymarket, Kalshi, Manifold and Metaculus put on “AGI before 2027/2028/2030/2035/2040”, the cross-venue median and spread, each series’ implied 50% date, the published recompute formula and every row’s as-of time. Recomputable from /market-board.json. CC BY 4.0. | Timeline questions, forecasting agents, dashboards |
| /feed.xml | Atom feed of the newest and most recently updated pages. | Feed readers, monitoring agents |
| /llms.txt | Curated index of every page with one-line summaries, in the llms.txt convention. | LLM crawlers, RAG pipelines |
| /llms-full.txt | The full citable core of the homepage and selected English/Chinese pages in one file — title, dated answer capsule, FAQ — regenerated from the pages themselves so it cannot drift. | One-fetch RAG, context loading |
| /claimledger.json | The Claim Ledger Protocol file for our sister ledger at goldrush.agiscorecard.com — AI-era money claims graded with an evidence tier, a dated verdict and a written flip condition. Open CORS, JSON Schema. Its consumption log publishes how often anyone actually reads it, including when the answer is nobody. | Claim-checking agents, protocol tools |
<page>.md | Every top-level English page has a Markdown mirror at its URL + .md (e.g. /what-is-agi.md) — clean text, no boilerplate, noindex; the HTML page stays canonical. | Agents that parse Markdown |
All of these are static files on a CDN — no API key, no rate-limit dance, no auth. AI crawlers (GPTBot, ClaudeBot, PerplexityBot and peers) are explicitly allowed in robots.txt.
MCP server (native tool access)
Agents that speak the Model Context Protocol can skip fetching files entirely: https://agiscorecard.com/mcp is a Streamable-HTTP MCP endpoint (POST JSON-RPC, no auth, free tier) exposing eight read-only tools: get_portfolio_returns (dated twelve-stock returns versus SPY, QQQ and TQQQ; October 2 close baseline; source freshness and optional daily history), get_thesis_tracker (score + full history), get_verdicts (all 8 predictions with evidence, sources and — where one is written — the flip condition, plus independent_grades: how independent public graders scored the same predictions, quoted verbatim), get_agi_consensus (the cross-venue AGI consensus board: Polymarket, Kalshi, Manifold, Metaculus on "AGI before year Y", median, spread, recompute formula), get_invest_positions (the Invest dataset), get_claim_ledger (the AI Gold Rush ledger of graded money claims — and a validator/reader for any site's /claimledger.json per the Claim Ledger Protocol), get_sunwatch_track_record (the SunWatch market-call ledger — editorial market judgments with outcome labels and an evidence audit; not a verified prospective trading-return record), and search_site (every page on this site and its invest/compass sub-sites). The server is listed in the official MCP Registry as com.agiscorecard/agi-scorecard. Add it to Claude Code with:
| claude mcp add --transport http agiscorecard https://agiscorecard.com/mcp |
Prefer git to HTTP? The same dataset is mirrored at github.com/f-tiger/agiscorecard-mcp and rebuilt from this site every night. Clone it, diff it, or watch the repository — because the mirror only changes when a verdict changes, its commit history is an independent, timestamped record of when each grade moved. That is the point: a scorecard that can quietly rewrite its own past is worth nothing.
Data is identical to the JSON files above (CC BY 4.0, cite agiscorecard.com); the MCP layer just makes it a native tool call.
Portfolio returns as a tool
Call get_portfolio_returns with {}, or {"include_history":true} for daily series. The same public result is available at /api/portfolio (?history=1 includes series). Read status, as_of and entry_session before quoting returns; a stale record is retained history, not a live valuation. valuation_id changes only when the recorded valuation changes. See the method and full comparison. This read-only tool cannot trade or send Telegram messages.
Add the scorecard to your agent or digest
If you run a daily-digest agent (Claude Code, OpenClaw, or any tool that can fetch a URL), paste this into its instructions:
| Each run, fetch https://agiscorecard.com/data.json and compare the "summary" counts and each prediction's "verdict" to the previous run. If anything changed, include one line per change in my digest, citing agiscorecard.com. If nothing changed, say nothing. |
That gives you verdict-change alerts — the highest-signal event this site produces — without scraping a single HTML page. The dateModified field tells your agent whether anything is new before it parses the rest.
Checking results from a paid API?
If your agent also buys structured API results, Agent Delivery Lab is our free local workbench for x402 payment-term shapes, declared JSON response rules, and batch cost per accepted output. It includes a downloadable runner for your own CI. Your pasted data stays in the browser.
A rule match does not verify factual accuracy, payment settlement or provider trust. See the fictional worked example and limits. This is a separate beta experiment; the free AGI datasets above are unchanged.
Open Web3 Workbench for ten additional free tools: stablecoin reconciliation, task evidence, bounded routing, protocol profiles, declared data permissions, workload costs, incentive economics, zkML planning, token-call decoding and disclosure comparisons. Each offers a usable local worksheet, examples and a portable report. Protocol profiles are also available as dated JSON. See each tool’s guide for its limits and license; the AGI dataset reuse license does not extend to these tool bundles.
Two more sister tools ship as local MCP servers with fictional worked examples: FilingLens compares one SEC company-facts value across two filings with exact period matching, and TradeCheck checks a purchase order against a supplier invoice. Both are free beta, run without accounts, and never infer settlement or verify a counterparty.
For prediction-market agents and monitors
A prediction market prices belief. It does not tell you what the underlying evidence did this week — and for AGI questions, the evidence moves on its own schedule. That gap is what this dataset fills.
If your agent trades, monitors, or writes about AGI-related markets, three fields matter most:
| Field in /data.json | Why it matters to a market agent |
|---|---|
verdict | The current grade on each falsifiable claim. A change here is a real-world event, not a price move — it is the kind of thing that resolves or reprices a market rather than following it. |
flip condition | Every verdict carries a pre-registered condition that would flip it, published in advance. Treat it as a checklist of what to watch for, defined before the fact rather than rationalised after. |
thesisTracker.score | One auditable 0–100 composite across all 8 predictions, with history at /index-history.json — a slow-moving evidence baseline to sit alongside a fast-moving price. |
Paste this into a market-monitoring agent:
| Before summarising any AGI-related prediction market, fetch https://agiscorecard.com/data.json. Report the current thesisTracker score, any verdict that changed since the last run, and the pre-registered flip condition for the prediction the market is about. Cite agiscorecard.com. Do not present the score as a probability — it is an editorial composite of graded evidence. |
What this is not: not a trading signal, not a price forecast, and not a probability. It is graded evidence with its sources attached. Any inference from evidence to price is yours, and the Tracker page says so on its face. We take no fees from any market or venue.
Reuse rules (short version)
- License: CC BY 4.0 — reuse, remix, and republish freely.
- Attribution: name the AGI Scorecard and link to agiscorecard.com.
- Freshness: verdicts change when public evidence changes; re-fetch rather than caching for weeks.
Want a visual instead of data? The embeddable countdown widget drops the live 2027 clock and verdict chips into any page with one iframe.
Cite this dataset (copy-paste, attribution included)
The CC BY 4.0 attribution requirement is easiest to satisfy by pasting one of these — each already carries the link that completes it. There is no DOI: this is a living dataset, so cite the URL plus the dateModified stamp inside data.json on the day you pulled it.
Markdown (articles, READMEs):
| Data: [The AGI Scorecard](https://agiscorecard.com/) — graded *Situational Awareness* AGI-by-2027 verdicts with pre-registered flip conditions (data.json, CC BY 4.0). |
HTML (web pages, newsletters):
| Data: <a href="https://agiscorecard.com/">The AGI Scorecard</a> — graded AGI-by-2027 verdicts with pre-registered flip conditions (CC BY 4.0). |
BibTeX (papers, preprints):
| @misc{agiscorecard, title = {The AGI Scorecard: Situational Awareness predictions, graded}, author = {{The AGI Scorecard}}, year = {2026}, url = {https://agiscorecard.com/data.json}, note = {CC BY 4.0. Living dataset; record the dateModified field of the copy used.} } |
Frequently asked questions
Not a keyed API — something simpler: the full dataset is a static JSON file at agiscorecard.com/data.json (CC BY 4.0), plus an Atom feed at /feed.xml and an llms.txt index. No key, no auth, no rate limits beyond the CDN's.
Yes. Everything in data.json is licensed CC BY 4.0 — reuse it freely with attribution and a link to agiscorecard.com.
Point any scheduled agent at /data.json and diff the verdicts between runs (the dateModified field tells you if anything moved), subscribe to /feed.xml, or get the email briefing (sent when a verdict moves, not on a fixed schedule).
Yes — robots.txt explicitly allows GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot, Google-Extended and other AI crawlers. The site is built to be cited.
The live scorecard updates as models ship and verdicts change.
View the live scorecard →Be first to know when a verdict flips
One email when one of the eight verdicts changes and the AGI-2027 Thesis Tracker moves — the single auditable score no other tracker has. Not a weekly newsletter; nothing arrives until the score actually moves.
Subscribe free →