marketing-measure-learnlisted

# marketing-measure-learn — score, then steer Measure and learn always run as a pair: first get the numbers (a deterministic API + math step), then reflect on them (agent judgement). Do them in order. ## Step 1 — MEASURE (deterministic) ```bash studio marketing measure --channel <name> --comments-n 60 ``` Fetches views/likes/comments (+ retention & subs gained if the analytics scope is granted), computes a **virality composite** (log-damped view-velocity + retention + engagement + sub-conversion), ranks every video into a **percentile within this channel's own portfolio**, and tags each `win` (≥P75) / `loss` (≤P25) / `neutral` / `cold-start`. Writes back to the journal and drops `08_stats.json` + `08_comments.json` into each run dir. Watch for: - **Wait for watch time** — measuring same-day gives noise. 48–72h+ minimum. - **Cold-start (<10 deployed):** percentiles are meaningless; every outcome is `cold-start`. - **Retention/subs** need one extra OAuth scope; fetched best-effort, the loop runs fine without. See [`../marketing-guru/references/analytics.md`](../marketing-guru/references/analytics.md). - **Scoring weights** (0.5 velocity / 0.2 retention / 0.2 engagement / 0.1 subs) are slated to be re-tuned to a retention-first order per research finding F-SI9 — see [`../marketing-guru/references/scoring.md`](../marketing-guru/references/scoring.md) and [`docs/20-research/self-improving-loop.md`](../../../docs/20-research/self-improving-loop.md). ## Step 1.5 — SNAPS