Can independent parties act on verifiable commitments instead of requests?
A principal signs what is allowed, a source signs what is true, and one executor makes the effect and answers with a signed outcome. Can anyone holding the records check the action, and what does that add over ordinary integration, a shared topic or a hub?
Where this stands
What the evidence shows 2
- From the English contract alone, a separate AI agent in the project built a checker that matched the Go reference on 435 of 435 version-0 cases and 75,229 re-signed mutants; the reference matched 629 of 629 cases the agent wrote. CMT-06 re-checked.
- On an Apple M4, a proof that a sealed record rests on a withdrawn premise, built to hide which one, took 31.88 to 32.82 s to make and 9.54 to 9.85 ms to check, nearly unchanged from 2^10 to 2^20 log entries. CMT-03 re-checked.
What remains open 3
- Only the checker role has conformance vectors. Executor vectors and a reader written outside the project are missing, and both sides of the agreement came from one project's AI-agent workflow, with isolation self-reported. CMT-06 re-checked.
- The withdrawal proof is not wired to the layer and its log entries are not bound to an issuer; its costs come from one run per configuration on one machine, and hiding the premise was the zkVM's design target, not tested. CMT-03 re-checked.
- The published studies do not yet answer what the layer adds over ordinary integration, a shared topic or a hub, or test most design-record claims, such as one effect per intent and withdrawal reaching dependents; by design it gives attribution, not enforcement. CMT-01 working note.
Not yet published
Eleven more studies are not yet published, because their records are not yet in the repository or an owner decision on their release is pending.
Every published study on this question
Attacks with harm, on the product in in-process stands: authority 1/33, cross-organisation 0/34, withdrawal 7/33 (all listed for the host), repeats 4/16 (one steward case), contradictions 0/28. Legal actions refused: 1 of 112.
Does combining help? Layer test
An internal, contract-only Python reader agreed with the Go reference on 435 of 435 version-0 cases, 11 of 11 network cases and 75,229 re-signed mutants; the reference agreed with 629 of 629 cases the reader's author wrote.
Does it work? Layer test
An accountable-action layer for agent tool calls: design of signed intents, outcomes and withdrawals
No measurement of its own: a reviewed specification with English contracts and conformance vectors, whose three review rounds found 17, 3 and about 40 blockers. The chain test and the contract-only reader check parts of it.
Does it work? Design record
On an Apple M4, one run per configuration: proof generation 31.88 to 32.82 s, receipts 225,384 to 227,432 bytes, verification 9.54 to 9.85 ms, nearly unchanged between 2^10 and 2^20 log entries within one segment.
Where does it stop? Prototype
Other questions
Who allowed this action, and within what limits?
3 published, 3 re-checked
10 published, 4 re-checked
One effect per intent, and an honest answer
11 published, 7 re-checked; 2 planned
When the basis of an action is withdrawn
2 published, 2 re-checked
What the guard costs, and how we measure
14 published, 3 re-checked