Assumption Rigor Audit: AI Skill for Product Management
Assumption Rigor Audit is an AI skill that audits how a team validated its assumptions, from method fit to evidence strength. Every finding names the violated principle, carries a severity rating, and comes with a concrete fix.
npx skills add Uxcel-Lab/product-skills --skill pm-assumption-rigor-audit
What this skill does
Turns "we validated it" claims into a structured rigor audit that shows whether the evidence supports the decision.
- Checks the riskiest assumption was tested first, not the comfortable one.
- Tests whether each assumption is phrased as a falsifiable statement with a real fail condition.
- Matches the test method to the decision's stakes: no production builds for prototype questions.
- Separates real behavior from stated intent, so "would you use this?" never passes as proof.
- Verifies success criteria were set before the test, not after results arrived.
- Interrogates samples, duration, and significance before a claim counts as evidence.
- Confirms disproven assumptions actually change the plan, with results documented.
- Rates every finding by decision damage and ends with the top three fixes.
When to use it
Use it when you need an AI assistant to:
- audit a discovery plan, experiment, or validation approach;
- pressure-test a "we validated it" claim before a big investment;
- check whether the riskiest assumption got tested;
- match test fidelity to the stakes of the decision;
- separate real evidence from stated intent and anecdotes;
- validate an AI-generated assumption test plan before running it.
What this audit checks
| Category | What it catches |
|---|---|
| Riskiest assumption | Comfortable or easy-to-test beliefs tested first, the load-bearing assumption never named |
| Falsifiable phrasing | Vague beliefs with no metric, threshold, or fail condition |
| Method fidelity | Production builds for prototype questions, big bets resting on hallway surveys |
| Behavior over intent | "Would you use this?" answers treated as evidence |
| Pre-set success criteria | Pass/fail thresholds decided after results are in |
| Evidence quality | Tiny samples, one-day windows, moved goalposts, anecdotes read as patterns |
| Learning loop | Disproven assumptions that change nothing, undocumented results |
How the skill works
1
Establish stakes
The skill asks what decision the validation supports and how costly or reversible it is.
2
Test the evidence
It checks which assumption got tested, by what method, against what pre-set bar.
3
Deliver ranked findings
It rates each issue by decision damage, gives the fix, and lists the top three priorities.
Try it with
Related skills
1.0.0
MIT
Product Management
Audits
