A Signed AI Agent Receipt Can Still Be Wrong
Your AI agent returns a signed receipt: 0 defects found. The signature is valid. The receipt has not been altered. The agent was authorized to run the check. The result can still be wrong. Perhaps the scanner hit a rate limit and silently converted eleven failures into eleven empty results. Perhaps a watchdog inspected 8 machines and issued a conclusion about 68. Perhaps a database health check…
An AI agent can generate a signed receipt indicating no defects were found, but that does not guarantee the accuracy of the measurement. The signature only proves the integrity of the receipt itself, not the validity of the results. There are various ways the scanner, watchdog, or database check could produce misleading outcomes, such as hitting rate limits, sampling too few targets, or overlooking critical columns.
The distinction between provenance integrity (who created the statement and if it was altered) and measurement integrity (what was actually observed and if it supports the conclusion) is crucial. A signed receipt alone cannot prove the business question has been adequately measured. The failure modes are not limited to bad code; many systems only have two result states - yes or no - and silence can be mistaken for a value.
An "UNKNOWN" receipt is different from no receipt at all; the latter could mean the check never started, crashed, or the target was unreachable. A signed "UNKNOWN" explicitly states that evidence did not justify a VERIFIED or REFUTED conclusion, and such a statement can carry additional information like a reason code, timestamp, verifier identity, and causal link to the work being checked.
This makes it auditable and prevents interpreting absence as success. The protocol should require every completed verification attempt to produce a signed verdict, with separate fields for measurement status and verdict. A check can be fully measured yet still return UNKNOWN if the evidence is insufficient for the claim. Another check may be incomplete due to limited observation targets.
A third may never have run at all. Separating these operational causes from the verdict helps downstream systems diagnose issues accurately and not mistakenly accept flawed results.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
