{
  "note": "Only these numbers may appear on the new home page. Values and wording follow BRIEF.md section 2 with CRITIC.md fixes. Hosted-model results (H7, H8, H10, H11, M3, M4, M7, M8) are held for owner clearance (Q5).",
  "labels": {
    "MEASURED": "counted or re-run by the named script on the stated date",
    "ESTABLISHED": "read from a dated record or code, and a check confirmed it",
    "INFERRED": "derived from other facts, not observed directly",
    "SYNTHETIC": "made by model personas or simulations; never user or demand evidence",
    "PLANNED": "not built or not run yet"
  },
  "claims": [
    {"id": "H2", "area": "harness", "text": "Python tests collected in the harness", "value": "3,511", "date": "2026-10-21", "source": "pytest --collect-only", "label": "MEASURED"},
    {"id": "H4", "area": "harness", "text": "Workflow skills with a deterministic eval; eval tests pass (eval not public yet)", "value": "10 skills; 54 tests", "date": "2026-10-21", "source": "skills repo, evals/", "label": "PLANNED"},
    {"id": "H5", "area": "harness", "text": "Route cells promoted by the promotion gate (3 route tables)", "value": "0", "date": "2026-10-21", "source": "route_table.*.json", "label": "MEASURED"},
    {"id": "H6", "area": "harness", "text": "Calls where the requested tier matches the resolved model", "value": "74/74", "date": "2026-10-21", "source": "conformance.jsonl", "label": "MEASURED"},
    {"id": "M1", "area": "models", "text": "A 1.75M-parameter JEPA world model, about 15k training steps (25,311 in the dataset), failed gate G1 three times; the stop rule fired", "value": "1.868 vs logistic 1.681 nats/step; -11.1 % [-11.7, -7.2]", "date": "2026-10-08", "source": "2026-10-08-p6-g1-attempt-3.md", "label": "ESTABLISHED"},
    {"id": "M2", "area": "models", "text": "Next-action prediction, best of 5 models tested on one private dataset: hand features (logistic)", "value": "logistic 1.681; Markov-2 1.812; prior 2.120 nats/step", "date": "2026-10-08", "source": "2026-10-08-p6-g1-attempt-3.md", "label": "ESTABLISHED"},
    {"id": "M5", "area": "models", "text": "HMM regime model vs Markov chain, held-out Brier (lower is better); the regime was a self-made proxy fix", "value": "HMM 0.110; chain 0.116; prior 0.162; gain +0.006 [+0.002, +0.011]", "date": "2026-10-08", "source": "2026-10-08-l1-regime-and-retry-ab.md", "label": "ESTABLISHED"},
    {"id": "M6", "area": "models", "text": "Action shapes predict failed SWE-rebench changes only weakly (public data, CC BY 4.0)", "value": "AUC 0.649 +/- 0.009 at n=12,000; rule 0.505", "date": "2026-10-18", "source": "ml/swe_shape_probe.json", "label": "ESTABLISHED"},
    {"id": "M12", "area": "models", "text": "QLoRA fine-tune loop: code and tests built; no run yet (0 examples approved)", "value": "0 runs", "date": "2026-10-21", "source": "rag_cycle.jsonl", "label": "PLANNED"},
    {"id": "CC3", "area": "change-chain", "text": "Commits linked to their agent session before the hook (one repo since 2026-10-01)", "value": "0 % of 142", "date": "2026-10-20", "source": "chain.py coverage", "label": "MEASURED"},
    {"id": "CC4", "area": "change-chain", "text": "Commits linked after the hook (one repo, about 8 hours)", "value": "session 95.0 % of 20; policy 100 %", "date": "2026-10-21", "source": "chain.py coverage --since 2026-10-21", "label": "MEASURED"},
    {"id": "CC5", "area": "change-chain", "text": "Same repo since 2026-10-01 (show next to CC4)", "value": "session 12.3 % of 163; policy 12.9 %", "date": "2026-10-21", "source": "chain.py coverage --since 2026-10-01", "label": "MEASURED"},
    {"id": "CC6", "area": "change-chain", "text": "change-chain v1 tests pass", "value": "108", "date": "2026-10-21", "source": "python3 -m unittest", "label": "MEASURED"},
    {"id": "CC9", "area": "change-chain", "text": "Coverage decision A2 (>= 90 % of new commits linked after 7 days)", "value": "2026-10-28", "date": "2026-10-28", "source": "nightly LOG.md", "label": "PLANNED"},
    {"id": "S1", "area": "h1", "text": "H1 preregistered: rules hashed and tagged before launch", "value": "2026-10-17", "date": "2026-10-17", "source": "docs/prereg-h1.md; /prereg-h1.txt", "label": "ESTABLISHED"},
    {"id": "S2", "area": "h1", "text": "Window 2026-10-20 to 2026-11-09 (21 days); the owner stopped H1 on day 2 to reposition the site; no verdict", "value": "day 2 of 21", "date": "2026-10-21", "source": "D22; nightly LOG.md", "label": "ESTABLISHED"},
    {"id": "S3", "area": "h1", "text": "Counted page loads at the stop, against a 500-load floor", "value": "242 of 500 (15:39Z)", "date": "2026-10-21", "source": "tools/daily_report.py", "label": "MEASURED"},
    {"id": "S4", "area": "h1", "text": "Qualified requests", "value": "0", "date": "2026-10-21", "source": "private request log not re-read", "label": "INFERRED"},
    {"id": "S6", "area": "h1", "text": "Page versions live in the window (exposure pooled)", "value": "v0, R1, R2, R3", "date": "2026-10-21", "source": "docs/launch-record-h1.md", "label": "ESTABLISHED"},
    {"id": "S8", "area": "method", "text": "Honesty scorer (number allow-list, deny words, labels, prereg byte-equal)", "value": "R3 91.1/100 (R1 91.3)", "date": "2026-10-21", "source": "tools/rsi_score.py; METRICS.tsv", "label": "MEASURED"},
    {"id": "RSI5", "area": "rsi", "text": "Gate waivers in the decision log, each a written entry: two WAIVER entries (D10, D21) and one reviewer substitute (D11)", "value": "3 entries", "date": "2026-10-21", "source": "docs/rsi/DECISIONS.md", "label": "ESTABLISHED"},
    {"id": "RSI6", "area": "rsi", "text": "Rule for unattended loops: the loop may decide, but it may never loosen a gate; a status changes only after the check re-runs; the loop never sets a waiver. A rule in a skill file, with no automatic check yet", "value": "rule text", "date": "2026-10-21", "source": "workflow skills repo, unattended-loop/SKILL.md", "label": "ESTABLISHED"},
    {"id": "RSI7", "area": "rsi", "text": "Never-loosen check that compares policy and gate values before and after each step: not built", "value": "0 steps checked", "date": "2026-10-21", "source": "not built", "label": "PLANNED"},
    {"id": "RSI8", "area": "rsi", "text": "Readout-driven loop that changes the harness or a model: not built; no external readout received; intake not open", "value": "0 steps run", "date": "2026-10-21", "source": "not built", "label": "PLANNED"},
    {"id": "SEC1", "area": "security", "text": "Permission decisions on record in one operator's sessions: 4 explicit rejections by a person and 3 auto-mode classifier denials, all 3 on removal of scratch folders. Counted by a throwaway script that was not reviewed and not re-run since", "value": "7 in 5,040 tool calls", "date": "2026-10-12", "source": "own session transcripts, 44 files; throwaway count script; archive/h1-r3/site/data/facts.json", "label": "MEASURED"},
    {"id": "SEC2", "area": "security", "text": "Bash calls and active hours used for the ruleset replay (same sessions, Bash calls only)", "value": "3,555 calls; 116 hours", "date": "2026-10-12", "source": "own session transcripts; replay script; archive/h1-r3/site/data/facts.json", "label": "MEASURED"},
    {"id": "SEC3", "area": "security", "text": "Replay of a throwaway ruleset (6 deny, 8 ask, 33 allow prefixes) over those Bash calls. A model of the policy, not observed decisions: matching approximated from vendor docs, then checked 2026-10-22 against the vendor matcher (CLI 2.1.288, 300 sampled calls, no call run): split within 1.7 to 8.3 points, 81 % per-call agreement", "value": "47 prefixes (6 deny, 8 ask, 33 allow); allow 731; ask 2,815; deny 9; about 24.3 added prompts per active hour", "date": "2026-10-12", "source": "replay script; vendor permission docs; vendor matcher check of 2026-10-22; archive/h1-r3/site/data/facts.json", "label": "INFERRED"},
    {"id": "SEC4", "area": "security", "text": "Rule that decided each call: the vendor event does not name the matching permission rule. Since version 2.1.301 an auto-mode classifier block names its built-in rule", "value": "not recorded", "date": "2026-10-17", "source": "vendor monitoring documentation and changelog", "label": "ESTABLISHED"},
    {"id": "ORG1", "area": "organism", "text": "Agent jobs of one operator's swarm in one snapshot (51 done, 6 partial or queued, 1 failed, 1 killed): 39 on this research, 20 on a sibling project. All ORG rows count both.", "value": "59 jobs", "date": "2026-10-20", "source": "docs/rsi/a3/organism-data.md; tools/organism", "label": "MEASURED"},
    {"id": "ORG2", "area": "organism", "text": "Handoffs: a file that one agent wrote and another agent read later. Lower bound: writes through bash are found by pattern.", "value": "422 handoffs", "date": "2026-10-20", "source": "docs/rsi/a3/organism-data.md; tools/organism", "label": "MEASURED"},
    {"id": "ORG3", "area": "organism", "text": "Files written by an agent that no other agent read. Most are scratch or final reports; a lower bound.", "value": "508 of 666 (76 %)", "date": "2026-10-20", "source": "docs/rsi/a3/organism-data.md; tools/organism", "label": "MEASURED"},
    {"id": "ORG4", "area": "organism", "text": "Stale reads: reads of a file that another agent changed later", "value": "133 of 2,083 (6.4 %)", "date": "2026-10-20", "source": "docs/rsi/a3/organism-data.md; tools/organism", "label": "MEASURED"},
    {"id": "ORG5", "area": "organism", "text": "Code writes followed by a test or a build run in the same job", "value": "71 %", "date": "2026-10-20", "source": "docs/rsi/a3/organism-data.md; tools/organism", "label": "MEASURED"},
    {"id": "SCAN1", "area": "scan", "text": "datapce-scan replay of two 6-rule deny sets over one operator's past Bash calls, subagents included, to 2026-10-21. Naive set: its rm -rf / prefix rule fired 257 times; a root-only rule fired 0 times. Fixed set: all denies are git reset --hard. Temp-dir targets mark likely false denies (a lower bound)", "value": "48,982 Bash calls; naive set 269 denies, 233 with a temp-dir target; fixed set 20 denies, 10 with a temp-dir target", "date": "2026-10-22", "source": "datapce-scan replay --rules datapce-6-v0,datapce-6 --until 2026-10-21 --include-subagents --json (re-run 2026-10-22); own logs, read only", "label": "MEASURED"},
    {"id": "SCAN2", "area": "scan", "text": "datapce-scan published: MIT licence, CI on Linux and macOS, Python 3.9 and 3.13", "value": "97 tests; main de6700f", "date": "2026-10-22", "source": "codeberg.org/datapce/datapce-scan; python3 -m unittest", "label": "MEASURED"},
    {"id": "CC10", "area": "change-chain", "text": "Commits of the site repo since 2026-10-21, by link: session established or inferred; policy established; deployment", "value": "57 commits; session 50 (87.7 %) established, 7 inferred; policy 57 (100 %); deployment 5 (8.8 %)", "date": "2026-10-22", "source": "chain.py coverage --since 2026-10-21", "label": "MEASURED"},
    {"id": "LH2", "area": "h2", "text": "H2, the preregistered launch test of datapce-scan: one post, a 14-day window. PASS needs 2 or more rule-set requests; KILL is fewer than 15 stars and 0 rule-set requests; else INCONCLUSIVE. The window is open and no result is read", "value": "no result", "date": "2026-10-23", "source": "docs/prereg-h2.md, SHA-256 4b0c0276e1db6a7cc5698bd1fecff7f8b830918080acae4f08e1d7c2e61dbdbc; D40", "label": "PLANNED"},
    {"id": "AGG1", "area": "readout", "text": "Minimum cell of the public aggregate: a cell shows a value only when 5 or more contributors give to it; set before any outside readout exists", "value": "5 contributors", "date": "2026-10-22", "source": "readout-v0.schema.json contributor.aggregate_ok; BI loop move 7", "label": "PLANNED"}
  ]
}