Capstone: The AI PRD

The capstone is the document this course has been assembling since Module 1: a complete, review-ready AI PRD for your feature — the artifact that answers, in one place, every question a serious team asks before building a probabilistic product. Structure (each section is a workshop artifact, matured):

1. 1Problem & opportunity (Module 1): the evidence-backed problem, the three door questions answered, the error-cost map. Plus the sharpest objection you received, and your current answer to it.
2. 2Scope contract (Module 2): v1 in its tolerant position with the vision as roadmap, non-goals with teeth, the miss-behavior table with forecasts, prototype findings with the honest caveat list, unit economics with ranges.
3. 3Quality contract & eval plan (Module 3): dimensions, rubrics, per-dimension bars with signatures line, the golden set's composition and growth rule, the judge spec with calibration receipts, the human-panel plan for the unmeasurable dimension.
4. 4Instrumentation & operating loop (Module 4): episode schema with segments, implicit-signal UX design, the privacy paragraph, the weekly ritual and layer-sorted backlog, the improvement budget line, drift/decay watchpoints.
5. 5Launch plan (Module 5): stages with pre-agreed gates, support enablement, comms that promise the floor, rollback rehearsed, the three success numbers with windows, and the expansion ladder's first rung with its earning criteria.
6. 6Then defend it: a 15-minute review (live with a colleague/mentor, or recorded against the review-question bank) — expect the four questions every AI PRD review contains: 'why will users trust it?', 'what happens when it's wrong?', 'how do you know it's good?', and 'what kills it?'. Your PRD answers all four in writing; the review tests whether you do without looking.

Certification rubric

• Problem discipline (20%) — evidence-first framing, the three questions answered honestly, an error map that shaped real decisions downstream.
• Scope & feasibility judgment (25%) — the tolerant-position choice defended, non-goals with reasons, prototype evidence used at its honest weight, economics penciled.
• Quality ownership (30%) — the contract a stranger could enforce: unambiguous rubrics, bars argued from user-felt cost, a golden set shaped like reality, judge receipts demanded. The heaviest weight, because it's the section that makes AI PM a discipline rather than vibes with standups.
• Operational & launch maturity (25%) — instrumentation that could actually run the loop, a launch plan with brakes, success defined before exposure, the kill answer given without flinching, and — scored explicitly, not implied — the steady-state ownership line: the named ongoing cost of operating the feature (the ~15-20%-of-an-engineer upkeep from Module 4) with an owner attached. A PRD that ends at launch and never budgets the loop loses points here; the operating cost is the product.

Passing earns the Certified AI Product Manager credential (ID format EDOVA-PM-2026-XXXX, independently verifiable at edova.ai/verify) — attesting that you can take a probabilistic capability from evidence to scope to measurable quality to staged launch, and operate what you ship.