Status
What the record holds, how much of it has been re-derived, what counts under each evidence policy, who has been contacted, and what stands between this preview and a citable tag. Built 2026-09-15; event-log digest 524dfab128a7; RULES.md v0.1 (2026-09-15). Every statistic on this page covers the 24 ranked organizations; the 1 watchlist entry and the 2 retained out-of-scope entries contribute to none.
Gates for a citable tag
| Gate | Where it stands | |
|---|---|---|
| Every binding signal has a quoted span | FAIL | 7 binding signal(s) without a span: equistamp.07, farai.07, farai.11, msft.07, palisade.09, redwood.08, saferai.04; all bind to non-extreme values (1-3), so no displayed score violates the clamp rule |
| Every extreme has a second coder | PASS | 17 of 17 extremes second-coded |
| Every ranked organization and every materially-named person has been contacted | FAIL | 24 of 25 contacted (24 ranked organizations, 1 materially-named person); the log lists 28 recipients, including 3 outside the gate: epoch, hal, hfoai |
| Default policy is Retrieved and confirmed and the Primary-only view is live | PASS | default standard; policies leads, standard, against_interest, spans, primary |
| RULES.md is published and every conflict cites a rule | PASS | RULES.md present; every conflict resolved by rule |
| Homepage carries the byline and the competence chip | PASS | checked in site/template.html |
Recomputed at every build by python -m bench gates. bench release --stage published refuses while any gate fails.
Coverage
| Ranked organizations | 24 |
| Watchlist (not ranked) | 1 |
| Out of scope, retained | 2 |
| Signals (ranked) | 331 |
| Signals with an exact quote (ranked) | 301 |
| Sources | 251 (236 confirmed, 10 unaudited, 2 imported, 3 unverifiable) |
| Ledger transfers | 103 (70 confirmed, 13 unaudited, 1 imported, 3 differs, 14 unverifiable, 2 superseded) |
| Ledger roles | 91 |
| Checked and not found (bounded negatives) | 27 |
| Entities | 115 |
| Regimes | 16 |
Exposure (ranked only)
| Ranked evaluators with a direct lab tie (hop 0), any kind | 10 of 24 |
| of which: lab cash for evaluation work (evaluation credits are recorded, never counted) | 6 |
| of which: a lab owns a stake or is acquiring the evaluator | 2 |
| of which: no-fee partnership or membership only | 2 |
| Ranked evaluators with an inflow from a lab or a lab-tied party (hop 0 or 1) | 15 of 24 |
| Funding sources with a second hop traced | 48 of 56 |
Evidence standard
Imported rows are leads copied from another project's ledger and satisfy no gate. A 4 on funding requires a confirmed bounded negative in a filing or index. A 0 needs a quoted span; a 4 needs a span and a second source. Scores are readings of the public record at the build date, not endorsements.
What counts under each evidence policy
| Evidence policy | Rule | Ranked signals that count | Assessments unevidenced |
|---|---|---|---|
| Leads included | Every signal that is not quarantined counts, including imported and unverifiable leads, which can support only a 2. For research, not for citation. | 329 of 331 | 1 of 192 |
| Retrieved & confirmed (default) | A signal counts only if at least one cited source was re-fetched and confirmed. The default: no number moves on a source you cannot open and check. | 313 of 331 | 8 of 192 |
| Against interest | Retrieved & confirmed, and an organization's own statements count only when they are against its interest or backed by a non-self source. | 235 of 331 | 49 of 192 |
| Verified spans | Retrieved & confirmed, and the signal carries a quoted span you can find on the page. | 291 of 331 | 13 of 192 |
| Primary only | Only confirmed filings, funder indexes, or third-party ledgers with a quoted span: what can be verified from outside the field. | 37 of 331 | 167 of 192 |
Population
45 organizations have been checked against the population criteria (RULES.md section 11): 23 out of scope, 14 in-scope candidates, 8 unclear. The full table is on the exclusions page.
Right of reply
28 recipients (27 organizations, 1 people); 24 contacted, 1 replied. The citable-tag gate tracks the ranked organizations and the materially-named people (the bench's claim text names them in a score-binding signal — RULES 12); any remaining rows are outside the gate. Packets are generated by bench outreach --all; sending is a human action, logged here with the date.
| Recipient | Kind | Contacted | Replied | Status |
|---|---|---|---|---|
| Andon Labs | organization | 2026-09-15 | sent, awaiting reply | |
| Apollo Research | organization | 2026-09-15 | sent, awaiting reply | |
| AVERI | organization | 2026-09-15 | sent, awaiting reply | |
| Center for AI Safety | organization | 2026-09-15 | sent, awaiting reply | |
| Dreadnode | organization | 2026-09-15 | sent, awaiting reply | |
| Epoch AI | organization | packet ready, not sent | ||
| EquiStamp | organization | packet ready, not sent | ||
| EU AI Office | organization | 2026-09-15 | sent, awaiting reply | |
| FAR.AI | organization | 2026-09-15 | sent, awaiting reply | |
| Gray Swan | organization | 2026-09-15 | sent, awaiting reply | |
| Holistic Agent Leaderboard (Princeton) | organization | packet ready, not sent | ||
| Hugging Face Open Alignment Initiative | organization | packet ready, not sent | ||
| Humane Intelligence | organization | 2026-09-15 | sent, awaiting reply | |
| Irregular (formerly Pattern Labs) | organization | 2026-09-15 | sent, awaiting reply | |
| METR | organization | 2026-09-15 | sent, awaiting reply | |
| Microsoft AI Red Team | organization | 2026-09-15 | sent, awaiting reply | |
| MLCommons (AILuminate) | organization | 2026-09-15 | sent, awaiting reply | |
| Nemesys Insights | organization | 2026-09-15 | sent, awaiting reply | |
| Palisade Research | organization | 2026-09-15 | sent, awaiting reply | |
| RAND | organization | 2026-09-15 | sent, awaiting reply | |
| Redwood Research | organization | 2026-09-15 | sent, awaiting reply | |
| SaferAI | organization | 2026-09-15 | sent, awaiting reply | |
| Scale AI (SEAL / Scale Labs) | organization | 2026-09-15 | sent, awaiting reply | |
| SecureBio | organization | 2026-09-15 | sent, awaiting reply | |
| Transluce | organization | 2026-09-15 | 2026-09-16 | replied — correction received |
| UK AI Security Institute | organization | 2026-09-15 | sent, awaiting reply | |
| US CAISI (NIST) | organization | 2026-09-15 | sent, awaiting reply | |
| Jaan Tallinn | person | 2026-09-15 | sent, awaiting reply |
Second coder
17 assessment(s) second-coded; 17 agree with the stored value. A human second coder on every extreme (every stored 0 or 4) is a gate for a citable tag; the full population follows by v0.2. Disagreements are logged in data/coding/second-coder.csv and in the assessment's resolution.
Open questions
Kept in paper/OPEN-QUESTIONS.md with status, routes tried, and the document that would close each. Audits in paper/audits/. Replies from named organizations and people are filed as signals; after first publication, a score that moves by more than one anchor triggers a record packet (bench outreach).
Data
bench.json (evaluators, assessments with per-policy derivations, signals, sources), exposure.json, timeline.json, ledger CSVs, event log, RULES.md. Code Apache-2.0; data CC BY 4.0. Disclosure: DISCLOSURE.md.