# Pre-registration — Cold Start Data, N2-A-001

*Section 4 of the episode research document, frozen before any data was pulled and
reproduced here verbatim. The as-run method — and every deviation from this plan —
is in `README.md`.*

## 4. Sample design (frozen before the pull)

Cohort: all public repos created 2023-01-01 → 2025-06-30 gaining **≥5,000 stars within 60 days of creation** (also report ≥10,000/30d hyper-viral subset so thresholds visibly aren't doing the work). Outcomes at T+6 and T+12 after crossing day.

- **Primary discovery: GH Archive WatchEvent counts** (BigQuery `githubarchive.day.2023*` … or streamed hourly files) — re-runnable by anyone, no token; sees later-deleted repos (the survivorship measurement).
- **Cross-check: Search API** `q=created:YYYY-MM-01..YYYY-MM-31+stars:>=5000` bucketed by month (1,000-result cap workaround; split any bucket ≥1,000). Search filters stars-*today*, not velocity — the delta between routes IS the survivorship story.
- **AI labeling post-hoc with a frozen rubric** (topics ∩ pre-registered list, keywords, borderline calls published); report full cohort AND AI subset.