{"data":[{"project":"default","id":"mcp.triage-ci60-ae4d07b","benchmark":"mcp.triage","split":"ci60","system":"hanzo-mcp/ae4d07b","version":"one Kai choice over six categories","baseline":true,"study":"kai-tools","embedder":"","reader":"","k":null,"temperature":null,"max_tokens":null,"prompt":"","prompt_digest":"","dataset_digest":"","store_digest":"","facts_digest":"","commit":"ae4d07b","questions":60,"answered":60,"when":"","ended":"2026-10-15T01:23:34Z","by":"","completion":"complete","key":"8521ac1e030417a21e467550e96a78f67584094809290290f81caa123abb7a61","measures":[{"category":"nameable","metric":"accuracy","value":0.5,"lo":null,"hi":null,"n":20,"of":40},{"category":"all","metric":"p50_ms","value":1627.87,"lo":null,"hi":null,"n":null,"of":null}],"notes":"60 real CI logs from 48 public repositories, labelled from the excerpt alone without the rules in view; nameable = the 40 whose class the old six categories can name; p50_ms is the handler's wall time against kai-1.2 on dgx's CPU while other training reads ran, not a serving latency","revision":"original","canonical":true,"visibility":"public","trainable":false,"publishable":true},{"project":"default","id":"mcp.triage-ci60-kai-failure","benchmark":"mcp.triage","split":"ci60","system":"hanzo-mcp/kai-failure","version":"Kai's failure capability alone","baseline":false,"study":"kai-tools","embedder":"","reader":"","k":null,"temperature":null,"max_tokens":null,"prompt":"","prompt_digest":"","dataset_digest":"","store_digest":"","facts_digest":"","commit":"kai-1.2 + failure capability","questions":60,"answered":60,"when":"","ended":"2026-10-15T01:57:17Z","by":"","completion":"complete","key":"6afff509f2376e7aebc040f14d55f5a876320400aed93aaf141e69db9bcbe0d4","measures":[{"category":"all","metric":"accuracy","value":0.4167,"lo":null,"hi":null,"n":25,"of":60}],"notes":"A choice capability (key failure, nine labels) over kai-1.2's pooled read of the log's last 4,000 characters, fit on 345 labelled CI logs from repositories disjoint from the 48 here; L2 100 and temperature 1.456 by five folds grouped by repository; the joining policy (Kai only where the rules answer unknown) chosen on those folds (out-of-fold .678 against the rules' .571) before the test was read once. Not in a release yet","revision":"original","canonical":true,"visibility":"public","trainable":false,"publishable":true},{"project":"default","id":"mcp.triage-ci60-measured","benchmark":"mcp.triage","split":"ci60","system":"hanzo-mcp/measured","version":"failure signatures, root cause first","baseline":false,"study":"kai-tools","embedder":"","reader":"","k":null,"temperature":null,"max_tokens":null,"prompt":"","prompt_digest":"","dataset_digest":"","store_digest":"","facts_digest":"","commit":"863cb8a7a47f8110dfd24c779b989442d0d01c0c","questions":60,"answered":60,"when":"","ended":"2026-10-15T01:23:34Z","by":"","completion":"complete","key":"c449883a3c6b44faa16312a2ebd1cf78ce2d53216c9b447d7d3b6dce49224278","measures":[{"category":"all","metric":"accuracy","value":0.7,"lo":null,"hi":null,"n":42,"of":60},{"category":"all","metric":"unknown","value":9,"lo":null,"hi":null,"n":null,"of":null},{"category":"nameable","metric":"accuracy","value":0.7,"lo":null,"hi":null,"n":28,"of":40},{"category":"all","metric":"p50_ms","value":0.37,"lo":null,"hi":null,"n":null,"of":null}],"notes":"60 real CI logs from 48 public repositories, labelled from the excerpt alone without the rules in view; nameable = the 40 whose class the old six categories can name","revision":"original","canonical":true,"visibility":"public","trainable":false,"publishable":true},{"project":"default","id":"mcp.triage-ci60-rules-failure","benchmark":"mcp.triage","split":"ci60","system":"hanzo-mcp/rules+kai-failure","version":"signatures, then Kai's failure capability where they answer unknown","baseline":false,"study":"kai-tools","embedder":"","reader":"","k":null,"temperature":null,"max_tokens":null,"prompt":"","prompt_digest":"","dataset_digest":"","store_digest":"","facts_digest":"","commit":"kai-1.2 + failure capability","questions":60,"answered":60,"when":"","ended":"2026-10-15T01:57:17Z","by":"","completion":"complete","key":"a398e5b08d6835c24e3075ff6bf6c0692c62317473f9ffe75b68ff65daa7e029","measures":[{"category":"all","metric":"accuracy","value":0.7333,"lo":null,"hi":null,"n":44,"of":60}],"notes":"A choice capability (key failure, nine labels) over kai-1.2's pooled read of the log's last 4,000 characters, fit on 345 labelled CI logs from repositories disjoint from the 48 here; L2 100 and temperature 1.456 by five folds grouped by repository; the joining policy (Kai only where the rules answer unknown) chosen on those folds (out-of-fold .678 against the rules' .571) before the test was read once. Not in a release yet","revision":"original","canonical":true,"visibility":"public","trainable":false,"publishable":true}],"total":4,"completion":{"complete":4,"partial":0,"inconsistent":0,"unknown":0},"gaps":[{"field":"by","runs":4},{"field":"dataset_digest","runs":4},{"field":"embedder","runs":4},{"field":"facts_digest","runs":4},{"field":"k","runs":4},{"field":"max_tokens","runs":4},{"field":"prompt","runs":4},{"field":"prompt_digest","runs":4},{"field":"reader","runs":4},{"field":"store_digest","runs":4},{"field":"temperature","runs":4},{"field":"when","runs":4}]}