Product Evidence

Framing without evidence generates strategy from vibes. This skill runs first: it collects what is actually known and sorts it by how much you can lean on it. The frame that follows is a compression of this ledger, not an invention.

The ledger — five buckets, strict definitions

Every entry carries a source: someone can resolve: a dashboard URL, a ticket ID, a quote with a date, a file in the repo. An entry with no source is an assumption whatever it claims to be. Promoting a signal to fact because it supports the plan is the whole failure mode this skill exists to prevent.

Where evidence lives, in order of cost

1. The existing product: what's shipped, what's used, what's abandoned mid-flow. Cheapest and most ignored.
2. Analytics: funnels, drop-offs, feature usage, search queries inside the product.
3. Support tickets and issue history: users describing their problem in their own words, at their own expense.
4. Sales and churn conversations: why people bought, why they left.
5. Interviews: expensive, richest. Ask about past behavior ("walk me through the last time you..."), never about intent ("would you use...?"). Answers about the future are polite fiction.
6. Competitor and category patterns: what the audience's other tools taught them to expect. Convention evidence, not desire evidence.

Read sources 1-4 before scheduling a single interview. Most framing questions are already answered by what the product and its users have written down.

Output

A one-page ledger, buckets in order, strongest first, each line claim — source. Close it with the three most expensive unknowns: the ones that, answered wrong, kill the project. Those go to the validation skill or into the frame's Unknowns line.

What this skill never does: recommend a solution. Evidence describes the territory; the frame picks the destination. When the ledger is thin, say so plainly: "we are betting" is a legitimate, honest frame input. A padded ledger is worse than a thin one.