Before you build a scraper: a public-data CSV scope card that keeps one-offs honest

Most “quick scrapes” fail the same way: unclear fields, no row cap, and silent nulls after pagination.

Before you open Playwright (or hire someone), write this scope card for any public source:

1. URL pattern — exact pages or API endpoints (include pagination / sitemap).
2. Public-only check — no login wall, no CAPTCHA farm, robots.txt / ToS allow automated fetch.
3. Fields (≤12) — name each column; drop “nice to have” extras.
4. Row budget — e.g. ≤5,000 rows. Caps force honesty about coverage.
5. Dedup key — what makes two rows the same (id, url, …).
6. Delivery — CSV plus a one-page schema.md with row count, null rates, and the source pattern.

Prefer official JSON/API when it exists (?$limit= on open portals often beats HTML parse). Flatten nested fields once; note what you dropped.

Update (Oct 2026): the OpsPacket Public Data Pull service mentioned in the original version of this post is retired and no longer for sale. The scope card above still stands on its own.