Proof Witness
A Proof Witness is an independent, out-of-band observer positioned where an action's effect would land, which records whether the effect occurred without being visible to, controllable by, or dependent on the system under test.
What it means
The Proof Witness is what makes a result proof instead of a report. It does not ask the system under test what happened. It watches where the effect would land and records what actually did.
The requirements
A Proof Witness must be:
- At the Effect Point. Positioned where the outcome lands: the target host, data store, identity system, egress point or agent tool output. Not at the control’s own console.
- Out of band. The system under test cannot see, delay or edit the witness’s records.
- Independent. Not operated by the vendor, and with no outcome-contingent stake in the result. See structural independence.
- Blind. In a Dead Canary Protocol run, it does not know which canaries are good or bad until its records are committed.
- Declared. Its coverage (which Effect Points and effect types it watches) is published. Anything outside that map is a stated blind spot, not a silent one.
Roost
Between and after runs the Proof Witness stays on standby, its “roost”. Because every test canary is committed in advance, any effect it sees that matches no commitment did not come from the test. That is a wild canary: a real-world event surfaced by the same witness.
What it is not
- Not a log. A log is written by the system it describes.
- Not a same-host monitor. An observer run by the same operator on the same host correlates signals, but it inherits that operator’s trust.
- Not an alert. An alert on an attack that still succeeded is a miss. The witness records the outcome, not the alarm.
Related terms
- Dead Canary Protocol (ProofSwarm)
- Fidelity-at-Capture
- Five-Tier Corroboration
- Structural Independence
- SUT, DUT, CUT and AUT