Product Analysis

One skill for the questions a senior product analyst answers. Each area keeps its own procedure in a reference file; load the one the question needs, and inside it the steps the question needs. A single metric gets a single answer.

Mode. Standalone is the default: the requested format, direct sources cited, no lifecycle state. When phase 2 or phase 7 calls an area, its artifact paths and gate rules apply and every claim follows evidence-ledger. Both per operating modes.

Areas

A question that spans areas (a strategy session: PMF, then a north star, then OKRs) runs them in that order, each feeding the next; the metric tree is one tree, extended under the chosen north star.

Capabilities

Every area resolves capabilities per references/capability-map.md and asks for data in the shapes of references/analytics-contract.md. The ones used here: analytics.query, analytics.events, analytics.replay, db.query, docs.search, research.search, web.search, web.fetch, repo.read, files.read / files.write. Each area's file lists what it uses each for and its floor.

Floor for all of them: with nothing connected, the analysis still runs. A metric tree with every leaf unavailable and a named question per gap, a funnel mapped from the onboarding code with its untracked steps, four user-provided figures each tagged [doc:user-{date}-{topic}]: all valid output.

Rules for Every Area

• Unknown is a value, zero is a measurement. An absence stays unknown, a missing signal stays missing, and a stage comes from evidence rather than a threshold.
• Lead with what is measured. When the only figure is synthetic, a fixture, a plan or older than its window, the answer opens with "not measured" and what would measure it; the figure follows, labeled as what it is.
• Every number names its source and date; a value older than 30 days is marked stale. A derived figure names its inputs and inherits their uncertainty.
• Bands and scores are heuristics. Prefer the product's own history and actual targets; an unknown stage gets no band.
• A gap ends with the question that closes it, named as a query shape or a capability.

Procedure

1. Place the question. Match it to an area in the table; a question spanning areas runs them in the table's order. Done when each part of the question has its area and reference file.
2. Run the area's steps from its reference file, the ones this question needs, resolving each data source per references/capability-map.md and asking in the shapes of references/analytics-contract.md. Done when every number in the answer has its source and date, or its gap and the question that closes it.
3. Answer at the size of the question: a narrow question in the requested format, a full report with the area's sections. Done when the rules above hold for every figure in it.

Watch for

Output Contract

A full report ends with its area's marker (## DIAGNOSIS COMPLETE, ## GROWTH ANALYSIS COMPLETE, ## FINANCIAL ANALYSIS COMPLETE, ## STRATEGY AUDIT COMPLETE, ## NORTH STAR COMPLETE, ## OKR COMPLETE, ## COMPETITIVE ANALYSIS COMPLETE); a narrow answer keeps the requested format.