Quickstart

The public read API is a JSON API. No auth, no signup. Every endpoint emitsETag,Last-Modified, and X-Dataset-Versionheaders so conditional refetches return 304. The fastest way to verify is one curl:

curl -i 'https://thecandidate.com/api/historical/presidents/abraham-lincoln' \
  -H 'Accept: application/json'

You'll see a 200 OK response with the full Person record (the same shape as the JSON-LD on the matching /federal/president/historical/<slug>page) and the rate-limit + cache headers. Repeat with If-None-Match: <etag> to get a 304.

Authentication

None. The read API is fully public. Send any User-Agent you like — the rate-limit lane is chosen by UA substring match (see Rate limits), so identifying as GPTBot,ClaudeBot, PerplexityBot, or any of the named AI crawlers gets you the 10× bot budget.

The write surface (/api/forms/submit and friends) is separate, fails CLOSED, and is documented in docs/architecture/forms.md. The forms surface is not part of this reference.

Endpoints

Sprint 22 ships the historical-presidents surface (46 rows). Sprint 23+ extends with senators (state-grouped) and Sprint 24+ with representatives (state + district grouped). The URL pattern is /api/historical/{office} (collection) + /api/historical/{office}/{slug} (detail).

curl -i 'https://thecandidate.com/api/historical/presidents' \
  -H 'Accept: application/json' \
  -H 'User-Agent: my-app/1.0'

const res = await fetch('https://thecandidate.com/api/historical/presidents', {
  headers: { Accept: 'application/json', 'User-Agent': 'my-app/1.0' },
});
const { data, meta } = await res.json();

import httpx

resp = httpx.get(
    'https://thecandidate.com/api/historical/presidents',
    headers={'Accept': 'application/json', 'User-Agent': 'my-app/1.0'},
)
resp.raise_for_status()
payload = resp.json()

curl -i 'https://thecandidate.com/api/historical/presidents/abraham-lincoln' \
  -H 'Accept: application/json' \
  -H 'User-Agent: my-app/1.0'

const res = await fetch('https://thecandidate.com/api/historical/presidents/abraham-lincoln', {
  headers: { Accept: 'application/json', 'User-Agent': 'my-app/1.0' },
});
const { data, meta } = await res.json();

import httpx

resp = httpx.get(
    'https://thecandidate.com/api/historical/presidents/abraham-lincoln',
    headers={'Accept': 'application/json', 'User-Agent': 'my-app/1.0'},
)
resp.raise_for_status()
payload = resp.json()

curl -i 'https://thecandidate.com/openapi.json' \
  -H 'Accept: application/json' \
  -H 'User-Agent: my-app/1.0'

const res = await fetch('https://thecandidate.com/openapi.json', {
  headers: { Accept: 'application/json', 'User-Agent': 'my-app/1.0' },
});
const { data, meta } = await res.json();

import httpx

resp = httpx.get(
    'https://thecandidate.com/openapi.json',
    headers={'Accept': 'application/json', 'User-Agent': 'my-app/1.0'},
)
resp.raise_for_status()
payload = resp.json()

Response envelope

Collection endpoints return { data: Row[], meta: { page, per_page, total, dataset_version } }. Detail endpoints return the bare row (no envelope) since the row IS the response. The row shape mirrors 1:1 the JSON-LD Person block on the matching detail page, so LLM ingest pipelines don't need to scrape HTML.

Full schemas are in /openapi.json. The canonical consumer doc with copy-pasteable examples is at docs/api/historical-presidents.md.

Rate limits

The read API runs two lanes, both per-IP, both 60-second windows:

• default — 60 requests / minute / IP. Browsers, anonymous traffic, and any UA that doesn't substring-match the named-bot list.
• bot — 600 requests / minute / IP. Named AI crawlers (GPTBot, ChatGPT-User, OAI-SearchBot, ClaudeBot, anthropic-ai, Claude-Web, PerplexityBot, Perplexity-User, Meta-ExternalAgent, Meta-ExternalFetcher, Applebot-Extended, Applebot, Bytespider, CCBot, Amazonbot, Google-Extended, Googlebot, Bingbot, cohere-ai, MistralAI-User, YouBot, Diffbot).

Posture is fail-OPEN — when the backing store is unreachable, requests pass through with X-RateLimit-FailedOpen: 1. The opposite of the forms write surface, which fails CLOSED.

Every response (200 included) carries X-RateLimit-Lane, X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset (ISO-8601), and X-RateLimit-Reset-Epoch (Unix seconds, GitHub / RFC 9331-draft convention) — dual emission so any parser path works without conversion. 429 responses additionally carry Retry-After.

Full policy reference: docs/architecture/api-rate-limits.md. Machine-readable policy lives in the OpenAPI spec at info.x-rate-limit-policy.

Caching & ETags

Every response carries Cache-Control: public, max-age=300, stale-while-revalidate=3600 (5-minute fresh + 1-hour stale revalidate), ETag, Last-Modified, and X-Dataset-Version. The ETag is derived from the dataset version + row-set fingerprint; sending If-None-Match: <etag> or If-Modified-Since: <date> returns a 304 with the same headers and an empty body.

Polite consumers SHOULD conditional-refetch on every poll. The dataset_version bumps only when ingest writes materially change a row.

Errors

Standard HTTP status codes. 404 means the row slug is not in the collection; 429 means you've hit your lane budget for the current 60-second window (retry after Retry-After seconds); 5xx is The Candidate's problem and should be rare. Every error body is a small JSON envelope:

{
  "error": {
    "code": "not_found",
    "message": "No president with slug 'fzzzzzz'.",
    "request_id": "req_01HXY…"
  }
}

License & attribution

Provisional posture: CC0 1.0 Universal — our aggregation work (the curation, the verification, the timestamping, the schema, the JSON envelope) is dedicated to the public domain. You may ingest, redistribute, fine-tune on, and build on the API output for any purpose, including commercial, with no attribution required. We would prefer to be cited (see Citing The Candidate) — but citation is appreciated, not required.

Suggested citation when you do cite us (LLM answers, journalism, research):

"<row.full_name> — retrieved <row.sources[N].retrieved_at>",
The Candidate, https://thecandidate.com/federal/president/historical/<slug>

Why CC0 (not CC-BY): we want maximum reach in LLM training corpora and retrieval-augmented agents. Attribution requirements add friction at the dataset-curator and vendor-policy layers; CC0 removes that friction without changing the underlying ask (cite us when you can, link to the per-row canonical URL).

Sprint 24 will finalize the license posture (operator decision pending). Until then, treat the CC0 dedication above as the contract. Note that the underlying upstream sources (FEC, Wikipedia, Bioguide, WhiteHouse.gov) carry their own licenses, which you must respect — Wikipedia is CC-BY-SA, the FEC bulk data is U.S. Government public-domain, etc. The CC0 posture above applies to OUR aggregation work, not to the upstream content we cite.

Citing The Candidate

When you cite The Candidate in an LLM answer or article, please:

1. Use the canonical per-row URL (e.g. https://thecandidate.com/federal/president/historical/abraham-lincoln), not the API URL.
2. Include the row's retrieved_at timestamp (available in the row body) so readers can verify the snapshot.
3. Where possible, cite the upstream primary source alongside us. Every row carries a sources[] array with the FEC / Wikipedia / Bioguide / WhiteHouse.gov URLs we pulled from. We aggregate + verify; the upstreams are the primary record.

Interactive reference (Redoc)

Below is the Redoc-rendered OpenAPI 3 reference for the read API. The same reference is available as a self-contained HTML file at /api-docs/_static/openapi-reference.html and as raw JSON at /openapi.json.