pm-hypotheses
Structure validation hypotheses, design low-cost experiments, track results, and issue a Go/No-Go decision
/pm-hypotheses▊When to use
Phase 3a hard gate. Runs after design-thinking. The only skill in the framework that produces a formal Go/No-Go verdict. No Phase 3b work begins until this verdict is GO.
Agent mode is not supported. The skill's value is the live iterative dialogue around assumptions and evidence. Invoking with --agent will warn once and proceed interactively.
What it produces
Two artifacts under pureinn-workspace/[project-slug]/artifacts/phase-3-define/:
- Hypothesis Register (
hypothesis-register.md) - ICP definition, assumption map, structured hypotheses, experiment plans with pre-defined success criteria - Go/No-Go Decision (
go-no-go.md) - evidence summary per hypothesis, verdict (GO / PIVOT / STOP)
How to invoke
/pm-hypotheses # interactive only - --agent not supported
At the start of every session, the skill asks which mode to run:
Plan mode (first run, or after a PIVOT):
- Structures hypotheses from
design-thinkingoutputs + Phase 2 evidence - Ranks by risk, assigns experiment type, defines success criteria before any experiment runs
- Output: complete Hypothesis Register with experiment plans
Results mode (after experiments complete):
- Records experiment outcomes against pre-defined success criteria
- Issues the Go/No-Go verdict
- Output: Go/No-Go decision with evidence summary
Both modes use the same Hypothesis Register - it is updated progressively, not recreated.
Go/No-Go decision logic
| Verdict | Condition |
|---|---|
| GO | Problem + Customer + Market hypotheses all confirmed (at least one payment/commitment signal) |
| PIVOT | Mixed signals - evidence points at what to change; new experiment defined with new success criteria |
| STOP | Problem not confirmed after 10+ ICP interviews, or no willingness-to-pay signal after quantitative test |
PIVOT escalation rule: After PIVOT 2, if all four hypothesis types (Problem / Customer / Solution / Market) have each been revised at least once, a structured review is triggered before more experiments run. The pattern indicates the original premise was fundamentally wrong, not misaligned on one dimension.
Hypothesis types
| Type | What it tests |
|---|---|
| Problem | Does the pain exist and is it severe enough to act on? |
| Customer | Are the people we're targeting the right ICP, and can we reach them? |
| Solution | Does our proposed solution solve the problem in a way users value? |
| Market | Is the market large enough and willing to pay? |
Experiment types (summary)
| Experiment | Best for | Signal strength |
|---|---|---|
| Customer Interview (Mom Test) | Problem + Customer | High (qualitative) |
| Landing Page | Problem + Market + Solution | Medium |
| Smoke Test / Fake Door | Solution | Medium |
| Targeted Ads | Market + Customer | Medium (reach + CAC) |
| Pre-order | Market + Solution | Highest (financial commitment) |
| Concierge MVP | Solution + Market | Highest (delivery + payment) |
| Waitlist | Market | Low-Medium |
| Wizard of Oz MVP | Solution | Medium (behaviour without build) |
| No-code / clickable Prototype | Solution + UX | Medium (behaviour, not commitment) - built via /pm-prototype |
| Survey | Customer + Market | Low (breadth confirmation only) |
Prototype experiments route to /pm-prototype. When the experiment for a hypothesis is a no-code or clickable prototype, this skill records the hypothesis and its success criteria and hands over; pm-prototype decides whether a prototype is the cheapest test at all, builds it, and records the result against that same criterion.
A prototype result comes back as one of four verdict states. Only the first two update a hypothesis:
| Verdict | Effect on the Hypothesis Register |
|---|---|
Supported within scope | Confirmed - for the stated population and dimension only, scope written into Result |
Refuted within scope | Rejected, scope written into Result |
Prototype or study failure | No update - the instrument failed, belief is unchanged. Fix it and re-run |
Inconclusive | No update - schedule another round or a different method |
A prototype result is behaviour, not commitment: it never satisfies a Market hypothesis on its own. A workbench prototype (built wide, cut down by roadmap phase) produces decisions and stakeholder alignment, not a verdict - it updates no hypothesis.
Commitment hierarchy (strongest to weakest):
- Credit card / pre-payment
- Letter of intent (B2B) or signed pilot agreement
- Email sign-up from targeted ICP (not friends)
- Click-through from paid traffic
- Verbal agreement (does not count as validation)
The "done elsewhere" path
If Phase 3a validation was done outside the framework (external workshop, investor due diligence, Miro session, etc.):
/pureinn "Phase 3a done elsewhere"
Provide the verdict (GO / PIVOT / STOP) and key evidence. Pureinn logs it and routes forward without re-running experiments.
Dependencies
Required before running:
design-thinking- Validation Hypotheses draft, Value Proposition Canvas, HMW questions are the primary inputs
Recommended:
pm-personas- ICP definition and pain intensityjtbd-building- Forces Diagram informs which assumptions are riskiestpm-problem-validation- Phase 2 evidence base
Produces for:
pm-prototype- builds the prototype for a hypothesis tested by a no-code / clickable prototype, returns a four-state verdictpm-lean-canvas- validated problem, customer, and solution feed Lean Canvas inputspm-prd- validation evidence and Go/No-Go verdict are required PRD sectionspm-product-roadmap(v1) - hypotheses and risks section
Related skills: design-thinking, pm-personas, jtbd-building, pm-problem-validation, pm-lean-canvas, pm-prototype