--- name: synthesise-validation description: Design and interpret customer, problem, demand, usability and willingness-to-pay experiments with explicit hypotheses, recruitment, thresholds and decision rules. license: MIT compatibility: Agent Skills-compatible agents; supporting workflow is loaded progressively. metadata: author: astisme --- # Synthesise Validation Reduce the most consequential uncertainty with the smallest credible experiment. Validation owns `validation-report.md`; a plan is not evidence that an experiment ran. ## Scope Customer, problem, solution, delivery and payment hypotheses; assumption ranking; interviews, smoke tests, prototypes, concierge pilots, paid pilots and transparent pre-sales; recruitment, sample, timeframe, cost, bias, thresholds and decision rules. ## Inputs A falsifiable hypothesis and target segment. If either is unclear, route to Audience or create an explicitly provisional hypothesis. ## Method 1. Write customer, problem, solution, delivery and payment hypotheses separately. 2. Rank assumptions by impact × uncertainty. 3. Choose the smallest experiment that can falsify the riskiest assumption. 4. Before execution, define segment, recruitment, method, sample, timeframe, cost, success/failure thresholds, bias and decision rule. 5. Match method to uncertainty: interviews for problem context; landing-page tests for qualified interest; concierge pilots for delivery; paid pilots or transparent pre-sales for willingness to pay; prototypes for usability. 6. Run the experiment and preserve raw observations with consent and provenance. 7. Separate observations from interpretation; note recruitment bias, sample limits and protocol deviations. 8. Decide proceed, iterate, pivot or stop. Explain remaining uncertainty and next test. ## Canonical output: validation-report.md Hypothesis register; prioritized assumptions; experiment protocol; recruitment/sample; predeclared thresholds; observed results; limitations/bias; decision and rationale; next experiment. ## Quality gates - Proposed and executed experiments are clearly distinguished. - Decision rules were defined before execution or deviations are disclosed. - Behavioural/payment evidence is weighted more strongly than hypothetical intent. - Sample and recruitment limitations are explicit. - Conclusions match evidence strength; no unsupported market-wide claims. ## Progressive methodology reference Load `references/validation-and-experiment-design.md` whenever performing substantive validation work. Use it to deepen the procedure, decision criteria and quality review; keep this `SKILL.md` as the concise entry point. The reference is additive and does not supersede shared contracts, approved decisions or the frozen v1.1.0 methodology. ## Shared evidence and handoff contract - Distinguish verified facts, user-provided claims, external research, analytical interpretations, hypotheses, recommendations and unknowns. Never silently promote one state into another. - Preserve evidence provenance and limitations: stable evidence ID, claim, type, source/locator, date, status, confidence, scope/segment, sample size where relevant, limitations and consuming artifacts. Reuse does not upgrade evidence strength. - This skill owns its canonical artifact. Other specialists may propose changes but must not silently overwrite it; cross-owner changes require owner/user approval. - Every handoff reports skill ID, project ID, objective, inputs used, outputs created, decisions proposed, decisions requiring approval, assumptions, evidence, limitations, next recommended skills and completion status. - Before completion, check required sections, output names, evidence support, explicit assumptions/limitations, ownership, project continuity and actionable next steps. State blockers rather than claiming completion. ## Handoff Pass validated/unvalidated hypotheses and evidence to Product Architecture. Product promises must not exceed what validation supports.