Use cases · Agent outcomes & evidence reports

Show what your agents delivered, not what they logged.

A customer paying for outcomes, or a board signing off on an agent rollout, wants to know whether the work was good. A log export shows that something happened. It doesn't show whether it met what you promised, and the reader has to take your summary on trust.

Request a demo → See what runs online

What gets recorded

Each agent action, sealed as an Agent Action Capsule where it happens, in a log that stays on your side.
The criteria for the month, sealed as a record before the window opens, so they can't be tuned after the results are in.
One judgment per case: met, not met or not evaluable, with the method named where a model judges.

What anyone can check

The report, its rows and the receipts behind them verify offline against public keys, without an Action State account.
A declared case count, checked at the end, so a missing case shows as a gap.

What runs today

What's built, and what isn't yet.

Runs today Sealing and offline verification (capsule-emit, the open verifier).
Runs today The evaluation compiler, a Claude Code skill: evidencebook-skills ↗.
Preview The monthly Evidence Report is produced with you today, as a retrospective audit on your own records.

How an Evidence Report reads · The sample report