# KREF Watch — llms.txt

> Searchable aggregation of Kentucky state & local campaign-finance filings
> (Kentucky Registry of Election Finance public records), 2016–present.
> 28 elections, ~537,000 contribution records, ~$466M.
> An independent project of The Lexington Times (https://lexingtonky.news).
> Not affiliated with the Registry.

## Open JSON data API (static, CORS-open, no auth, no key)

All site data is plain JSON under `https://krefwatch.com/data/`:

- `/data/elections.json` — index of elections (date, type, totals). Election keys are `YYYY-MM-DD` (e.g. `2026-05-19` = May 2026 primary).
- `/data/candidates-index.json` — every candidate across all elections (name, slug, office, location, total). Use this to find a candidate's slug + election date.
- `/data/e/<date>/summary.json` — KPIs, topCandidates, topDonors (named donors only), topSelfFunders, topEmployers, byOffice, byContributionType, monthly series, selfFundingTotal, unitemizedTotal.
- `/data/e/<date>/races.json` — every race (office + location) with per-candidate totals.
- `/data/e/<date>/donors.json` — per-donor aggregates with per-recipient breakdown (large file, up to ~6 MB).
- `/data/e/<date>/employers.json` — employer-normalized rollup (canonical aliases merge "UK"/"University of Kentucky", "Commonwealth of Ky" variants, law-firm suffixes, etc.).
- `/data/e/<date>/candidates/<slug>.json` — full itemized contributions for one candidate. Row fields: d=donor key, n=name, a=amount, t=type, m=mode, c=city, s=state, z=zip, e=employer, o=occupation, r=receipt date, k=in-kind description.

Typical flows:
- "Who raised the most in X election?" → `/data/e/<date>/summary.json` → `topCandidates`
- "How much has <candidate> raised?" → `/data/candidates-index.json` (find slug+date) → `/data/e/<date>/candidates/<slug>.json`
- "What has <person/company> donated?" → `/data/e/<date>/donors.json`, filter client-side
- "Which employers' people gave the most?" → `/data/e/<date>/employers.json`

Donor keys prefixed `candidate-self-`, `unitemized-`, `anonymous-`, `cash-unnamed-`, `unnamed-` are synthetic (self-funding, small-dollar bundles, unnamed cash) — not named people.

## Agent guide

- `https://krefwatch.com/skill.md` — task-oriented guide for AI agents using this data.

## Caveats (important for accurate reporting)

- Filings lag: pre-election report deadlines mean recent giving may not appear yet.
  Dataset build date is `generatedAt` in `/data/elections.json`.
- Donor identity is grouped by name; same-name people can merge, misspellings can split.
- Employer values are self-reported and normalized heuristically.
- Amounts are as-filed (refunds included where present). Verify against KREF originals
  (https://secure.kentucky.gov/kref/publicsearch) before publication.
- Federal races (US House/Senate/President) are NOT here — that's FEC data, not KREF.

## Source code

Fully open source (pipeline + frontend): https://github.com/paul-codes-1/krefwatch

## Contact

editor@lexingtonky.news · The Lexington Times · https://lexingtonky.news