data-discovery

Find and vet a government source before adding any number to the dashboard. Use when sourcing a new KPI, comparator, Government Order, or Status-Paper figure — turns "I need a number" into "I have an

SKILL.md

GitHub

Data discovery

Every figure on this dashboard must trace to an official government source (see docs/data/source-policy.md). This skill is how you find one and confirm it before writing a fixture.

Steps

  1. Check what we already catalogued. Read the per-domain inventory in docs/data/ (e.g. fiscal.md, health.md) and the typed registry data/sources.ts. The source may already be listed.
  2. Go to the primary body, not coverage. Prefer, in order: the issuing department (*.kerala.gov.in / *.gov.in), a constitutional/statutory body (RBI, CAG, ECI, Census, MoSPI), the Legislative Assembly (niyamasabha.org), then data.gov.in. A newspaper or aggregator is only a lead to the underlying official document — never the source of record.
  3. Get a resolvable URL to the exact document/page (a PDF, a table, a dataset). Note the precise table/figure number — provenance must be specific (Table 3.1b, not "the report").
  4. Record retrieval. Capture the retrievedAt/lastRefreshed timestamp (ISO with +05:30).
  5. Verify it passes the gate. Run deno task check:sources. If the host isn't on the government allowlist, you have the wrong source — go back to 2.

Reliability ladder (best → unacceptable)

Issuing department → constitutional/statutory body → Assembly → national open-data portal → (stop: below this line is not a source of record) → think-tank / aggregator (PRS, OpenBudgets, Dataful) → newspaper → Wikipedia / blog / social.

Output

A { source, sourceUrl, retrievedAt } triple ready to drop into the fixture, plus the exact table/figure citation. If you cannot find an official source, the number does not ship — open a tracking note instead.

Related