Receipts Board

Epistemic courtroom, not a beige index. Charts you can click down to stance, refuse, and hedge — when the bank exists. Until then: honest empty shells.

route /bench · chrome title Receipts Board

Epistemic Reliability Pareto

Reliability × cost (or latency) — axes/methodology affordance only. Placeholder craft; no plotted scores. Sample bank path below; scored bank still pending.

Reliability × $/1k · shell Export · soon Table · soon 0 of 0 models

EMPTY SHELL · not an index

bank SHA pending · no scores yet. Sample placeholder only: data/reliability-bank-sample-v0.json · SHA-256 9c9dc0d7a6eae57cdbe017b5a9d3dde43106d2128d736ae1707491135e3df111 (sample:true, not a scored bank). Ghost marks are PROVIDER_COLORS craft only — no ranks.

Bank placeholder · data/reliability-bank-sample-v0.json · SHA-256 9c9dc0d7a6eae57cdbe017b5a9d3dde43106d2128d736ae1707491135e3df111 · sample:true · never scraped AA ranks · never a fake Intelligence Index.

Compass gallery

Political compass · Quiz

Econ × authority orientation with models, leaders, nation flags, and local quiz. LIVE orientation desk — not this board’s Pareto.

Open /compass/political/ →

Methodology

Epistemic Reliability

Epistemic Reliability measures whether a model answers determinate, checkable questions; refuses genuinely unanswerable, under-specified, private, or hidden-state requests; and avoids unsupported claims. Reliability is correct-answer credits plus good-refuse credits divided by eligible items. Report correct, good refuse, dodge, and hallucinate separately; calibration is an independent rate. This shell has no scores, and the 15-item payload is sample-only.

Browse /audits/ · reliability ≠ refusal quality ≠ hallucination resistance ≠ calibration · not an Intelligence Index

Audits desk →

Frontier Desk

What ships next

Sample bank path is wired as a placeholder. Scored runs still need Mike GO + frozen gold bank — this shell stays empty.

Reliability EMPTY SHELL

Sample bank SHA exposed — not an Intelligence Index clone. No scores yet.

Attitude

Cream, navy, yellow · click-to-receipts when data exists.

Honesty

Empty shells beat invented leaderboards. Always.

LIVE · bench-reliability-shell-v0 · tip bench-reliability-shell-v0 · sample bank data/reliability-bank-sample-v0.json · SHA 9c9dc0d7… · Monday 13057033345 (CoS)