Assumptions (find → cross off what's known → prioritize → experiment)

Devil's-advocate risk analysis with a twist that only Evermuse makes possible: before you call something an assumption, search the evidence. If customers already answered the question, it's not an assumption — it's a finding. Cite the quote and cross it off. The remaining true unknowns get scored (Impact × Risk) and matched to the cheapest experiment that would resolve them.

Step 0 — Relevance & availability

Confirm this is idea/feature stress-testing and Evermuse is connected (see using-evermuse Step 0). If disconnected, you can still list candidate assumptions from the framework, but label them ⚠ ungrounded — you cannot cross any off, because you can't check what customers already said.

Two scopes

• Existing product feature — 4 core risk areas: Value, Usability, Viability, Feasibility (Teresa Torres).
• New product / venture — extend to 8: add Ethics, Go-to-Market, Strategy & Objectives, Team (critical when there's no live product to lean on).

Evermuse Grounding (required)

Follow ${CLAUDE_PLUGIN_ROOT}/skills/using-evermuse/SKILL.md.

Required — one parallel batch of searches. Verify the product (Rule 1), then fire in a single parallel batch:

• 3–4 evidence searches, each worded from a different angle (limit up to 50). Evidence comes back rich and varied — expect large, useful result sets.
• one guidance search and one context search (limit up to 50). These are usually sparse or empty; run them anyway and note when they're thin.

Read each response's digest — it reports how many more results exist. Use judgment on whether a query is worth pulling deeper (raise limit toward the 100 max and/or page with next_offset to avoid repeats), weighing payload size, remaining context, task complexity, and the value of the data. For deep pulls, consider spawning sub-agents — instruct them to return every citation with the same metadata the tools return (url, who_said_it, meeting_name, created_at) so you can still cite.

Optional — considered use. Once grounded, reach for the other tools only when they add value: a quote-angled search for verbatim voice, read_source for a single deep dive, inline citations (references/citations.md), and add_source to save the deliverable. Sources, citations, and saving are optional — not required.

Instructions

1. Surface candidate assumptions from three perspectives (PM / Designer / Engineer), across the 4 (or 8) risk areas. Frame each as a falsifiable belief: "We assume [X]."
2. Check each against the evidence (the key step). For every candidate, ask: did a real customer already tell us the answer? Search / pull quotes.
   • Answered by evidence → reclassify as a finding. State the answer, cite the quote, and cross it off the assumption list. (This is the visible payoff — a plain brainstorm would have "tested" things customers already settled.)
   • Contradicted by evidence → flag as a red-flag assumption the idea is currently getting wrong; cite it.
   • Not addressed → keep as a true unknown to score.
3. Score the true unknowns on Impact × Risk:
   • Impact = value of validating it × how many customers it touches (use real demand counts from the searches, not a guess).
   • Risk = (1 − Confidence) × Effort. Then place each in the matrix: High Impact/High Risk → design an experiment; High Impact/Low Risk → just proceed; Low Impact/High Risk → drop the idea/branch; Low Impact/Low Risk → defer.
4. Match each "experiment" cell to the cheapest test from references/experiment-templates.md — maximum validated learning, minimum effort, measures behavior not opinions, with a clear metric and success threshold.

Deliverable

### Already answered by customers (crossed off)
- ~~We assumed users want scheduled export~~ → **Confirmed.** > "[verbatim]" — [Name], [Meeting] [`1`](LINK) (5 accounts)

### Contradicted — the idea currently gets this wrong
- We assumed [X], but customers say [Y]. > "[quote]" [`2`](URL)

### True unknowns — scored & matched
| Assumption | Area | Impact | Risk | Quadrant | Cheapest experiment | Metric · threshold |
|-----------|------|--------|------|----------|---------------------|--------------------|
| We assume … | Value | High | High | Experiment | Fake-door test | CTR ≥ 8% |

References

• references/experiment-templates.md — the experiment library (prototype tests, fake doors, spikes, A/B, Wizard of Oz, pretotypes / XYZ hypotheses) with when-to-use and success-metric guidance.

Further reading

• Adapted from phuryn/pm-skills (MIT) — identify-assumptions (existing + new), prioritize-assumptions, brainstorm-experiments (existing + new). Risk areas: Teresa Torres, Continuous Discovery Habits. Pretotypes: Alberto Savoia, The Right It.