Capstone: One Full Cycle on Your Real Product
Run a complete discovery loop on your real product and score it against the four-dimension rubric.
Time: 1–2 weeks, async. Coding required: Depends on your product and the solution you test — but the capstone passes or fails on judgment, not code. Prerequisite: Modules 0–6, completed on the ShiftLoop sample. You will produce: Evidence of one complete discovery loop run on your real product, reviewed against the rubric below. Passing is the gate to “certified on the system.”
What this is
Everything until now was a rehearsal on shared, fictional material. The capstone is the real performance: you run a complete discovery loop on your actual product, with real customers, real signals, and real consequences. This is the moment the curriculum stops being a course and becomes how you work.
It’s also the natural before/after artifact — for an internal “certified” bar, and (for consulting engagements) for the client’s own success story.
The loop you run
Run one full turn of the cycle, end to end. Lean on the skills and agents you learned — this is exactly what they’re for.
- Collect real signals. Gather a real batch this week — interviews, support tickets, reviews, NPS, sales notes. Not last quarter’s research; this week’s. (No minimum volume required, but a single source or single method is a Low confidence ceiling no matter how strong the signal feels — flag it in your ledger entry. Aim for at least 2 sources or 2 methods; 3–5 interviews plus a batch of tickets is a solid capstone corpus.)
- Synthesize (Module 3). Cluster by need, tag confidence, surface contradictions, write a real Signal Ledger entry.
- Update your OST (Module 2). Map the clusters in (add / update / challenge), keep the tree structurally honest, and confirm a focus branch.
- Decompose one assumption and test it cheaply (Module 4). Take your focus opportunity, name the riskiest assumption, design the cheapest test, write the kill condition first, and — if feasible in the window — run it. (The rubric scores test design, not whether the test ran in the capstone window. A well-designed test with a pre-written kill condition scores the same whether or not you had time to execute it.)
- Run it through a gate (Module 4). Get an honest investment-gate verdict and the cheapest next action.
- (Optional) Hand off to the team (Module 5). If a solution survived, run it to user stories / a design brief and note where your judgment was required.
- Report at your weekly synthesis (Module 6). Bring the loop to a real cadence session — yours or your team’s.
You don’t have to reach “Building.” You do have to run a complete, honest loop and show your reasoning at each step.
What to submit
- Your Signal Ledger entry for the week (real sources, real verbatims).
- The OST before and after, so the update is visible.
- A one-page experiment brief: riskiest assumption, chosen test type + why, success + kill conditions.
- The investment-gate verdict with its next action.
- A short reflection (½ page): where the evidence was thin, what you’re still unsure about, and what you’d do next.
The rubric
Reviewed across four dimensions. Each is scored 0 / 1 / 2. Passing = 6+ of 8, with no dimension at 0. A zero anywhere is an automatic re-do on that dimension — these are the load-bearing habits.
1. Evidence quality
| Score | Looks like |
|---|---|
| 0 | Clusters are paraphrase only; opportunities rest on a single source or the PM’s opinion; no verbatims. |
| 1 | Most clusters have verbatims and a confidence tag, but at least one acted-on opportunity is thinly evidenced or single-source. |
| 2 | Every cluster carries verbatims, source counts, and a calibrated confidence tag; nothing prioritized rests on a single source; biases named. |
2. Tree integrity
| Score | Looks like |
|---|---|
| 0 | Solutions sit at the opportunity layer; more than one outcome at the root; orphan branches; no clear focus. |
| 1 | Structurally mostly sound, but a minor issue remains (a near-duplicate, a weakly-connected branch, a fuzzy focus). |
| 2 | One measurable outcome at the root; opportunities framed as customer needs; signals attributed; one clear, justified focus branch. |
3. Test minimalism
| Score | Looks like |
|---|---|
| 0 | Jumped to “build it and measure,” or proposed an expensive test for a cheap question; no kill condition. |
| 1 | Reasonable test, but not the cheapest that answers the assumption, or the kill condition is vague/post-hoc. |
| 2 | Riskiest assumption correctly ranked; cheapest test that answers it chosen with rationale; success + kill conditions written before running. |
Self-scoring the 1/2 boundary: Walk the four test types cheapest-first (fake door → concierge → prototype → A/B). Ask: “Is there a cheaper type that still answers my specific riskiest assumption?” If no cheaper test answers the same question — you’re at 2. If a cheaper test would have answered it — you’re at 1. Note the “why not cheaper” rationale in your brief; that reasoning is what earns the 2.
4. Honest uncertainty
| Score | Looks like |
|---|---|
| 0 | Overstated confidence; contradictions smoothed over; no acknowledgment of what’s unknown. |
| 1 | Some uncertainty acknowledged, but a contradiction or a weak spot is glossed. |
| 2 | Contradictions surfaced as open questions; confidence honestly calibrated; the reflection names what’s still unknown and the cheapest way to learn it. |
Why these four: they’re the habits that make every future cycle trustworthy. Speed and polish aren’t scored — a slow, honest loop passes; a fast, confident, unevidenced one does not.
After you pass
- Internally: you’re certified on the system. The expectation now is that you keep the Module 6 cadences — the capstone proves you can run the loop; the cadences prove you do.
- As a facilitator: passing the capstone qualifies you to co-facilitate the next cohort. (If you can’t teach a step, the docs aren’t done — and you’ll know exactly where.)
- Keep your friction log. Anything that tripped you on your real product is a doc fix for the next learner — feed it back (see
_testing-the-curriculum.md).
Solo / no-cohort path
If you’re working without a cohort facilitator, self-certification counts — use the rubric exactly as written, be honest with yourself on each score, and keep your scored rubric alongside your submission artifacts. The test is whether you can produce the artifacts and critique them honestly against the criteria. A solo PM who scores themselves a 1 where they deserved a 1 learns more than one who rounds up.
Not sure what a passing submission looks like in practice? See the worked exemplar — a full fictional capstone submission scored against the rubric.
← Back to the learning path