The three verifiers

1. verify_mission_journal.py

Checks a durable mission journal: a hash-chained, signed log of a long-running agent mission, from the authority that started it to every effect that was observed. On the published sample it performs 31 kinds of check, 169 checks in all, including:

Group	Checks (count on the sample)
Genesis	first entry; mission signed by the authority; plan signed and bound; trust matches supplied keys
Chain	continuity (33); every entry signed by the pinned runtime key (33); contains the witnessed head
Admission	action in plan; authorization signed for the exact action, unexpired, never used before; delegation chain live; mission not expired or cancelled; request signed by the plan's agent; human approval present
Dispatch	admitted first; exactly once; idempotency key; authority live at dispatch; before expiry; not after cancellation; bound to the admitting contract version; recovery class; result follows intent
Effect	observation follows a dispatch; observation signed by the pinned observer
Decision and completion	every input decided in order (26); decision digests (13); no unresolved dispatch; completion still supported by the evidence

python3 -I verify_mission_journal.py mission.journal.jsonl --trust trust.public.json --head head.json

2. verify_execution_truth.py

Answers one claim: "exactly the authorized effects occurred inside the observed roots." It re-checks every signature and digest in the receipt and re-derives the grade from approver-signed and sensor-signed records. On the sample: AUTHORIZED_EFFECTS_OCCURRED, level INDEPENDENTLY_CORROBORATED, postconditions HOLDS, no gaps, nothing undeclared. Give it several receipts and it keeps one record of receipt and authorization ids across them, so a replayed receipt fails (RECEIPT_REPLAYED). If a sensor you anchored did not report, the result is INCOMPLETE.

python3 -I verify_execution_truth.py valid.receipt.json valid.anchors.json
python3 -I verify_execution_truth.py replayed.receipt0.json replayed.receipt1.json replayed.anchors.json   # FAILED

3. verify_launchpad.py

Checks a deployment export from the Launchpad: a gate ledger of effects, the steps that authorized them and a checkpoint signed by a witness federation. On the sample, six checks, all VERIFIED: trust material (4 witness keys, threshold 3); gate ledger signature (12 effects, sequence 1..12); checkpoint signature; deployment bound to the checkpoint; witness quorum (4 verified signatures, 3 needed); every effect has an authorized step.

python3 -I verify_launchpad.py deployment export.json --trust trust.public.json

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.