The Proof tab

Every promise your build made, its stamp, and the actual evidence that earned it — readable, not just trustable.

The Proof tab is your build’s receipts. Two panes: on the left, every promise the build made — worst news first; on the right, the recorded request transcript for whichever claim you select. A stamp asks you to trust the platform; the transcript beside it lets you check the platform.

The stamps

Every promise carries exactly one stamp, and the different stamps are never blended into a single “mostly fine” number:

Stamp	What it means	Whose move it is
PROVED	A machine check drove the app and demonstrated it.	Nobody’s — it held.
CONFIRMED · YOU	You judged it yourself, on a recorded date.	Nobody’s — you settled it.
DID NOT WORK	A check ran and the promise failed.	The agent’s — it needs fixing.
COULDN’T CONFIRM	Something was tried and came back unclear.	Open — it isn’t settled either way.
UNPROVEN	A prover exists but hasn’t been able to run yet.	The platform’s debt — there is nothing for you to do. When it lands, this resolves to PROVED or DID NOT WORK.
NOT CHECKED	Nothing was attempted, so nothing is known. Not a failure.	Stated honestly so you can decide whether it matters.
NEEDS YOU	The check is waiting on your judgment.	Yours — see Your verdicts.

The headline and the bar

At the top, one sentence states the closure count — how many promises are discharged out of how many made — and a segmented bar shows the shape of the evidence. It is deliberately not a percentage: a rate can hide one broken promise inside “94% verified,” while a bar drawn from countable rows can’t say anything the list below it doesn’t also say.

The evidence pane

Selecting a claim shows who proved it, how, and when — and the recorded request/response exchange that earned the stamp, when one exists. When there is no transcript, the pane says exactly why, and the reasons are kept distinct:

• Nothing ran for this promise — no check tried it on this attempt.
• This check ran, but keeps no transcript — the verdict is real; the evidence isn’t replayable on this screen.
• We can’t tie a transcript to this promise — rather than show you a transcript that may belong to a different claim, it shows none.

Below both panes, an “Also ran — not tied to a promise” section lists checks that executed but don’t belong to any single promise (delivery checks, platform probes). They’re shown so nothing that ran is hidden — and kept out of the evidence pane so nothing gets attributed by guesswork.

When the record itself is incomplete

The Proof tab tells you about its own limits, too:

• If the agent is still finishing something a story needs, a banner says the list is not complete yet.
• If a run couldn’t finish at all, the page says so and adds: “Nothing here says anything about your app yet — not that it works, and not that it is broken.”
• Spec rules that were declared but never exercised are listed under what we could not check — and why, each with its reason.

The same record feeds the publish screen — the numbers you see there and the rows you see here come from one place, so two screens can never disagree about your build.