{"site":"match.inocta.io — inocta-bench","site_url":"https://match.inocta.io/","version":"1.0","manifest_url":"https://match.inocta.io/api/mcp/manifest","execute_url":"https://match.inocta.io/api/mcp/execute","request":{"method":"POST","url":"https://match.inocta.io/api/mcp/execute","content_type":"application/json","body":{"tool":"<tool name>","params":"<object matching the tool's inputSchema>"},"note":"Send arguments under 'params'. 'input' is accepted as a synonym. Read tools take no auth."},"description":"Read-only tools over a completed, published AI benchmark: three AI reasoning architectures, the betting market, an ELO rating and one human predicted all 104 FIFA World Cup 2026 matches before kickoff, scored on calibration. The Council is the best-calibrated architecture. Pre-registered at DOI 10.17605/OSF.IO/R5SBJ; papers at /paper.","documentation_url":"https://match.inocta.io/llms.txt","tools":[{"name":"get_benchmark_info","description":"Returns what this site is and the final results of the completed, published study: the inocta-bench methodology summary (three AI architectures vs the betting market, an ELO rating and one human, scored on calibration), the finding (the Council is the best-calibrated architecture tested), the OSF pre-registration links and DOI, the /paper links (scientific and business reads, EN and FR), final match counts, total AI spend, and the page map in both languages. Use this first to orient.","group":"site","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{}}},{"name":"get_scoreboard","description":"Returns the final leaderboard for all eight analysts over all 104 matches: Brier score, expected calibration error (ECE), picks made and refused, hit rate, total cost in USD and total model calls. Lower Brier and ECE are better. The Council is the best-calibrated architecture.","group":"results","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{}}},{"name":"list_matches","description":"Lists World Cup 2026 matches tracked by the benchmark, ordered by kickoff. All 104 matches are finished. Returns match id, teams, kickoff time (UTC ISO), stage, venue, status and final score. isDevWindow=true marks run-in matches played before the rules locked (they count as training and are excluded from final results). Use the match id with get_match for the full prediction detail.","group":"matches","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{"status":{"type":"string","description":"Which matches to list: 'finished' (default, all 104) or 'upcoming' (none remain).","enum":["upcoming","finished"]},"limit":{"type":"integer","description":"Maximum matches to return, 1-50. Default 10.","minimum":1,"maximum":50}}}},{"name":"get_match","description":"Returns one match with the analysts' locked pre-kickoff predictions: winner pick, confidence, score line, risk factors, reasoning, abstentions, cost and latency per lane, plus the closing betting odds snapshot and the final score. Predictions were locked about 75 minutes before kickoff (the market closing-line horizon) and are on the record, complete for all 104 matches.","group":"matches","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{"matchId":{"type":"integer","description":"Numeric match id, e.g. 1489369. Get ids from list_matches."}},"required":["matchId"]}},{"name":"get_analysts","description":"Returns the eight analysts: the three AI architectures (Solo, Pipeline, Council), their stripped-down control twins, the ELO statistician, the betting market baseline and the human — with persona, one-liner, type and model rosters. Available in English and French.","group":"site","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{"locale":{"type":"string","description":"Language of the descriptions: 'en' (default) or 'fr'.","enum":["en","fr"]}}}},{"name":"get_calibration","description":"Returns the final calibration metrics (M8): per-analyst reliability-curve points (stated confidence vs actual hit rate, in 10 bins), expected calibration error (ECE) with bootstrap 95% interval, and the final-ECE 'architecture race' series over all 104 scored matches. Lower ECE means more honest confidence. The pre-registered gate (meaningful from ~24 matches, leader asserted from 30) is long since cleared at 104; the response still carries asOfMatchCount and a gate flag. The Council posts the lowest ECE; the leaders' intervals overlap, a disclosed limit rather than a null.","group":"results","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{}}},{"name":"get_standings","description":"Returns the final World Cup 2026 group standings, the real football tables, not benchmark data. The 12 groups (A-L), each with the four teams ranked by points then goal difference: played, won, drawn, lost, goals for/against, goal difference, points, recent form, and a qualification tag ('Round of 32' for advancing teams). Also returns the cross-group third-place ranking ('Group Stage' table) that decided which eight of the twelve third-placed teams advanced. This is fan information, separate from the calibration research.","group":"football","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{}}},{"name":"get_bracket","description":"Returns the World Cup 2026 knockout bracket — the real tournament tree (Round of 32, Round of 16, Quarter-finals, Semi-finals, Final), plus the third-place playoff. Each round lists its matchups with teams, kickoff, status and final score where played. Before the group stage resolves the rounds are empty (hasKnockoutFixtures=false); they fill in as fixtures are drawn. This is fan information, separate from the calibration research.","group":"football","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{}}},{"name":"get_citation","description":"Returns a ready-to-cite reference for the published study: the APA and BibTeX strings, the DOI and DOI URL, the Zenodo preprint, the four paper PDFs (scientific and business, EN and FR), and the one-sentence finding to attribute. Use this to cite inocta-bench correctly: Kahlain, M. (2026), Calibration over Accuracy, DOI 10.17605/OSF.IO/R5SBJ.","group":"site","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{}}},{"name":"get_paper","description":"Returns the published study as a scholarly object: title (EN and FR), abstract, author, year, DOI, keywords, the headline result (Council best-calibrated, ECE 0.100 over 104 matches, with the full ranking), and the four read variants (scientific and business, each EN and FR) as fetchable PDF URLs. For a completed study archive, this is the paper node.","group":"site","annotations":{"readOnlyHint":true,"idempotentHint":true},"inputSchema":{"type":"object","properties":{}}}],"browser_tools":{"note":"Additionally registered on navigator.modelContext (Chrome WebMCP) on every page. The write tool runs via its on-page form, not the HTTP execute endpoint.","tools":[{"name":"subscribe_final_report","description":"Subscribes an email address to inocta's research updates. The 2026 World Cup study is published (read it at /paper); this feeds the ongoing research list for new benchmarks and findings. Occasional, no spam. WRITE ACTION, has a side effect (adds the address to the research list).","group":"site","annotations":{"readOnlyHint":false,"idempotentHint":true},"inputSchema":{"type":"object","properties":{"email":{"type":"string","description":"Email address to receive the one final report."}},"required":["email"]}}]}}