CAIN-42 CAIN StudioLimits
What KEYSTONE does not prove, and what the public evidence does not yet show. We would rather you read this here than discover it yourself.
| Limit | Why it matters | Status |
|---|---|---|
| Live fixing is off on the public site | You cannot watch a new fix be made and signed today; the ledger holds runs from Oct 14-15 | Open (needs model API credit) |
| Small, narrow evidence: 4 fixes meet today's rule, all on one sample, one detector check | It proves the loop works, not that it works across many kinds of defect | Open |
Record 2 was signed VERIFIED_FIX under an older rule with no reference test | Under today's rule it would be NOT_VERIFIED; the record is kept, never rewritten | Disclosed 2026-11-02; flagged by the re-run kit |
| The reference test's hash is not inside the record | A reader cannot prove from the record alone which reference test ran | Partly addressed: tests published + re-run reproduces; schema change planned |
| Passing tests is not the same as a correct fix | Tests can miss things; that is why we ask reviewers to write their own | Inherent |
| One signing key on one host | Compromise of that host could sign false records going forward | Open |
| Python only | Other languages are not covered | Open |
| The sandbox limits are our own tests | No third party has tried to escape it | Open (independent review wanted) |
| No certification, no third-party audit | Everything here is checkable, but nobody independent has checked it yet | A paid independent review is being commissioned |
FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.