CAIN-42 CAIN StudioAnatomy of a record
Schema cain42.keystone.fix-proof.v1. Every field below is taken from record sequence 1
in the live ledger.
| Field | Example (record 1) | Meaning |
|---|---|---|
schema | cain42.keystone.fix-proof.v1 | Record format version |
sequence | 1 | Position in the chain |
run_id | ks-20261014T180213Z-591582b2 | Unique run id, with time |
at | 2026-10-14T18:02:13Z | When the run finished |
sample_id, path | forged-token, tokens.py | Which sample (null for pasted code) and file name |
finding | unconditional_verdict, line 15, MEDIUM | What FALSEWORK flagged |
falsework_before / _after | PARTIAL → REAL | Detector grade before and after the fix |
original_code, original_sha256 | a28e0099…7a17 | The file as found, and its hash |
fixed_code, fixed_sha256 | 5ae419b4…3d7c | The file after the fix, and its hash |
diff, lines_changed | unified diff, 7 | What changed |
claude_test_code | pytest file | The model's own test |
explanation | "The original verify_token returned valid=True for any non-empty signature…" | The model's explanation |
claude | claude-opus-5-5, 871 in / 992 out tokens, 10.2 s, req_011Cg2Zvbx… | Which model, cost and provider request id |
tests.claude.before/after | not all passed / all passed | The model's test on the original and on the fix, per test |
tests.reference.before/after | not all passed / all passed | The independent reference test on both (null when none existed) |
verdict, reasons | VERIFIED_FIX, [] | Outcome and, if not verified, why |
prev_record_hash | 004e92d9…d6bf | Hash of record 0: the chain link |
record_hash | 52396661…e46c | SHA-256 of this record's canonical JSON |
signature_b64, public_key_b64 | Ed25519 | Signature over record_hash, and the signer's key |
What the record does not contain
The reference test's code and hash. The record holds the reference test's per-test results, but not the test itself, so a reader cannot prove from the record alone which reference test was run. Since 2026-11-02 the reference tests for the four demo samples are published separately (reference_tests.json) and the re-run kit shows they reproduce the signed results exactly. Binding the test's hash into the record is planned for a future schema version; see limits.
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.