Feedback Loop

The Feedback Loop asks whether each release moved what it said it would, from the readings that actually came in. For every released release it shows the key results it expected to move, the readings in each window, the checks at +7 days and at each horizon, the verdict, and what was done about it.

Checks

Each expected signal is checked a week after release and again at its horizon. The page shows each check as due, done, or due on a date, and how many readings its window holds. Owners and studio admins run the checks as they fall due, or take an early look before then.

Verdicts

On track — the readings moved the expected way and are at, or well on the way to, the target.

Behind — moved the right way too slowly, or not at all.

Regressed — moved the wrong way from the baseline.

Inconclusive — too few readings, contradictory readings, or no baseline to compare with.

The Telemetry Analyst gives the verdict from the readings in the window and the server’s own arithmetic, with the observations it rests on, a confidence, and a recommendation. With no connector or no readings for a key result, the verdict is inconclusive and no agent is asked. Each check is kept; running it again adds a check rather than replacing one.

Course corrections

A behind or regressed verdict raises a course correction on the Recommendations page, proposing an intent that names the release and the key result. A person accepts or rejects it. Nothing creates an intent on its own — not even in a Fully Autonomous studio — and an intent created this way records the release it corrects, which the Feedback Loop then shows.

All AI Product Studio documentation

Build v0.3.2 · deployed 2026-10-19T23:52:31Z