interview-synthesizerlisted

# Interview Synthesizer Transcripts in, patterns out — every pattern earned by verbatim evidence, or it doesn't ship. ## Verification gates (defined first; output is blocked until all pass) - **G1 — Evidence floor:** every pattern cites ≥2 verbatim quotes. From ≥2 different transcripts when more than one exists; a single-transcript pattern is labeled `single-source`. - **G2 — Zero invented quotes:** every quoted string matches the source transcript character-for-character (contiguous span; trimming only at the ends). The gate check is mechanical: search each output quote in the input text — any miss fails the gate. - **G3 — Attribution:** every quote carries transcript + speaker (e.g. `[T2-Marco]`). ## Steps 1. **Index.** Label each transcript (T1, T2, …) and each speaker. If the user supplied zero transcripts, stop: ask for them — never synthesize from nothing. 2. **Extract candidates.** Read across transcripts for repeated pains, workarounds, desires, and objections. A candidate is anything mentioned by ≥2 sources — or a strong single-source signal, tracked separately. 3. **Evidence pass.** For each candidate, collect exact quote spans (≤40 words each — quote minimally, cite precisely). Copy, never retype from memory. 4. **Demote or drop.** Candidates that can't produce 2 verbatim quotes become `Hypothesis (insufficient evidence: N quote)` at the bottom of the output, or are dropped. Never pad evidence to promote a favorite. 5. **Gate pass.** Run G1–G3 on the draft: su