Workshop: The feedback loop

This workshop writes the operational half of your PRD — and then makes you run one week of it against simulated data, because loops designed but never rehearsed have a way of being decorative.

1. 1Write the episode schema for your feature: the five core fields adapted to your context, the segments (pick 4-6 and defend each with the launch question it answers), and the miss-path events with their triggers. One page, PRD-ready.
2. 2Design the implicit signals into the UX: for your feature's surfaces, name where acceptance, edit, rejection, and abandonment become distinguishable events — and sketch the one UX change that makes the best signal exist (Dana's case: 'accept resolution' and 'dismiss' as separate affordances, never a generic close).
3. 3Write the privacy paragraph and the retention clock — honestly, for your actual company's posture. If you can't write it, you've found a real launch blocker early; log it as such.
4. 4Spec the weekly ritual: the dashboard's four views, who attends, the worst-20 selection rule, the harvest format, and the layer-sorted fix backlog's owner. Plus the drift review's quarterly checklist (3 questions minimum).
5. 5Run the triage drill: below-simulated week-one data — acceptance 61% overall but 34% for orders >$200; edit-distance high on the apology sentence specifically; miss-path 'route to human' at 22% vs. 8% forecast; one episode where the feature messaged about an order the customer had already called about (the non-goal that had a bug). Sort into layers, pick your top three actions with reasons, write the two-line update you'd send your exec sponsor. There are defensible different answers; there is also a clearly right first action — the non-goal bug, because it's the trust budget, not the metrics, that's bleeding.