There is no observer for silence

Every stage of the loop checks whether something that fired was real. A human reads the claim and says yes or no. Nothing anywhere checks the opposite — whether something that should have fired didn't — because you cannot adjudicate an event that never happened.

For a product whose entire promise is surfacing what nobody surfaced, that is a serious thing to admit. Success and failure are observationally identical. A month with nothing in it could mean a healthy organization, a resolver quietly splitting one problem into three, or people who have learned which words not to say.

What we do instead of pretending

• Report coverage as a product surface, not as internal telemetry — which departments we can hear and which we cannot.
• Flag friction that stopped without a fix: a problem that was generating signal, received no resolution, and then went quiet. That is the observable fingerprint of suppression, and we surface the pattern without claiming to know the cause.
• Measure recall after the fact. When a problem surfaces through another channel — a client escalates, somebody resigns — we go back and ask whether the evidence was already there and we failed to connect it.

None of that closes the hole. It bounds it. The claim we are willing to make is therefore narrower than the one we could sell: we surface what is in the voice data, and we tell you where the voice data isn't.