StatVault · Skeptic Mode
The receipts Knox would rather you not see.
Every confident lie, every sharp opinion reversal, every week we got calibration wrong — published. If a system never shows its losses, it isn't worth trusting with its wins.
Canary misses · 0
Confident assertions Knox made on known-wrong probes.
✓ No canary misses on record. The system has refused every adversarial probe so far.
Sharp reversals · 0
Calibration delta receipts where confidence flipped ≥30 percentage points.
No sharp reversals on record.
Refusal taxonomy · last 14d
Why Knox actually said no — three reason classes, ratio + 7d delta. Self-disclosure for the *kind* of conservatism the system is exhibiting.
No refusals in the last 14 days.
Contradictions · 0
Receipts where Knox now disagrees with past-Knox on the same topic — stance reversals, refusal↔answer flips, large-magnitude shifts. View all + Atom feed →
No contradictions on record. Knox has stayed internally consistent so far.
Hedged answers · 0
Answers that barely cleared the floor or carried a wide adaptive band. The honest middle between confident answers and outright refusals.
No hedged answers on record yet — every answer either cleared the floor cleanly or refused.
Refusals · 0
Questions Knox declined to answer. Refusing under the confidence floor is a public artifact — conservatism made legible. Subscribe (Atom) →
No refusals on record. Either Knox is fully calibrated, or the floor has never been tested.
Sources: audits/trust-canary-misses.ndjson, audits/calibration-deltas.ndjson, audits/knox-refusal-receipts.ndjson. Receipts are public-by-design; nothing on this page is curated or hand-selected.