What problem does it solve? After running customer discovery interviews, teams struggle to turn scattered notes into disciplined learning: hypotheses go unupdated, single anecdotes get over-weighted, and nobody decides when to stop interviewing. This Skill reads per-interview debrief files against your working HYPOTHESES.md and QUESTIONS.md and walks you through evidence-cited updates one proposal at a time. ## Core Features & Use Cases - Three-tier update rule: Distinguishes stray voices (parked in a 'That's funny' watch section), patterns (which validate, tune, or disprove hypotheses), and revelations (single voices that legitimately reframe a belief). - One-proposal-at-a-time walk: Every change is proposed with quoted evidence from named debrief files, decided by the user, applied immediately, and recorded in a mandatory change log with frozen hypothesis/question numbering. - Closing verdict: Ends each synthesis with a committed continue (with focus and number), stop-and-act, or re-aim decision based on surprise rate and convergence. - Use Case: You have nine interview debriefs in an interviews/ folder. Run this Skill to validate H2, tune H3's pricing threshold, mark H7 disproved, add a new hypothesis with a matching interview question, and get a verdict to continue with five more freelancer interviews. ## Quick Start Ask the assistant to synthesize the interview debriefs in your interviews folder against HYPOTHESES.md and QUESTIONS.md and tell you whether to keep interviewing.
