Research API reference

Create a key in the dashboard, send it with one request, and keep the returned request ID with your research log. The example asks for the public SEC filing chronology for Apple; it does not require a market-data license.

export DATACEDAR_API_KEY="dc_..."

curl --request GET \
  --url "https://api.datacedar.com/v3/stocks/AAPL/filings?form=10-Q&limit=5" \
  --header "X-API-Key: $QUARTERTRACE_API_KEY"

Use the production origin configured for your account. Local development defaults to the API origin in QT_API_BASE_URL; the Next.js application proxies authenticated dashboard requests without becoming the data authority.

Send one key through X-API-Key or Authorization: Bearer. Do not send both. Account keys begin with dc_, are displayed once when created, and are stored as hashes by DataCedar.

# Header form
X-API-Key: dc_...

# Bearer form
Authorization: Bearer dc_...

Successful list responses return data, pagination, and meta. A stream may also include coverage or availability when the distinction between empty, missing, and restricted matters.

{
  "data": [{ "symbol": "AAPL", "form": "10-Q" }],
  "pagination": {
    "count": 1,
    "limit": 5,
    "has_more": false,
    "next_cursor": null
  },
  "meta": {
    "request_id": "req_...",
    "timestamp": "2026-07-17T20:15:00Z",
    "schema_version": "v3"
  }
}

Unknown JSON fields may be added in a backwards-compatible release. Clients should ignore fields they do not recognize and must not rely on object key order.

List endpoints use opaque cursors. Pass the returned next_cursor unchanged into the next request. Do not parse it, synthesize it, or combine it with a different filter set.

GET /v3/stocks/AAPL/documents?limit=100
GET /v3/stocks/AAPL/documents?limit=100&cursor=eyJ...

# Continue until pagination.has_more is false.

Cursor ordering is deterministic for the same filters and eligible data snapshot. If an acquisition run adds newly eligible records during a long backfill, restart the export with a saved as-of cutoff when exact snapshot consistency is required.

DataCedar does not overload one timestamp. Depending on the stream, a record can carry economic-period, event, publication, filing, retrieval, market-session, and known-at times.

The catalog below is the implemented v3 stock-research surface. A route can be implemented but legitimately return unavailable when no permitted source is active.

Bar timeframes accepted by the current validator are 1Min, 5Min, 15Min, 1Hour, and 1Day. Public market availability is a separate rights decision.

Default acquisition cadence

Cadence is configurable per deployment. These defaults describe how often the durable scheduler attempts each source; they are not a guarantee that a provider has published a new record.

Error responses use a stable machine code, a human-readable message, optional field details, and the same request ID emitted to sanitized service logs.

{
  "error": {
    "code": "INVALID_SYMBOL",
    "message": "Symbol contains unsupported characters.",
    "details": { "field": "symbol" },
    "request_id": "req_..."
  }
}

Every rights-sensitive market or transcript row carries a license class. Public requests return only records eligible for that environment and account. Internal validation data remains filtered even if a customer plan would otherwise allow a longer history window.

Decodo may transport a brittle public request, but it does not grant copyright, exchange, database, storage, or redistribution rights. Acquisition adapters and serving entitlements are reviewed independently.

Limits protect the shared service and define the maximum eligible market-history window. They do not create access to a stream whose source rights or coverage are unavailable.

For batch work, page sequentially within each symbol, cap concurrency, cache immutable document responses, and retry only transient status codes. Contact us before designing a sustained workload near the Desk ceiling.

A reproducible result saves more than the response body. Record the request URL and filters, as-of cutoff, response request IDs, pagination cursors or export manifest, coverage state, schema version, source run IDs, and the code revision that produced the analysis.

Raw provider objects are content-addressed and written to Cloudflare R2 before normalization. They are recovery evidence, not an unfiltered public download tier. Customer-facing data still passes normalization, coverage, plan, and rights checks.