Skip to main content
Follow a guide end to end for one job: pull a bounded history, export files, check a window before use, run a job inside your limits, evaluate fit, or move a read off a native venue API. Each guide starts from one market and one window, and ends with output you can reproduce.

Before you start

  • Know the record you need: a snapshot, historical rows, a file export, a freshness decision, or a replay.
  • Name one venue and market type, and one symbol or market identity. Keep the original symbol spelling; Venues and market types shows the namespace for each.
  • Load an API key from the runtime environment for authenticated market-data routes. The public /v1/symbols discovery route does not need a key. REST authentication covers the request header, and each endpoint page lists its final parameters and response fields.
  • Keep each run bounded. Start with one market, one route, and one short window, and widen only after the first run has a clear result and request IDs.

Pull and check data

Pull historical market data

Page through one bounded window with cursors, and save the request IDs and quality decision with the rows.

Check one window before use

Read freshness, coverage, and incidents for the same market and window before a backtest or alert depends on it.

Export historical data

Create one bounded Data Catalog export selection and keep a traceable order record for the files.

Prepare data for a reproducible backtest

Freeze a window, check it, and pull it through REST or WebSocket replay with a run manifest.

Plan and evaluate

Limits and throughput

Run one finite job with explicit worker, page, retry, and stop budgets.

Evaluate a market data API

Decide whether an API returns the record one job needs, with route, quality, capacity, and rights evidence.

Migrate from a native venue API

Move one historical order-book read to 0xArchive and keep execution and account calls where they are.

First bounded request for coding agents

Give a coding agent one exact route, a bounded window, and a stop condition.

What every guide records

Every guide ends with the same record beside its output: the method, path, and query; the venue and market type; the symbol; the response status; the request IDs; and the quality or coverage decision that controls the next action. A second engineer should be able to repeat the same bounded call from that record without guessing. Keep the response body separate from the record so recorded inputs are never mistaken for returned data. If a request fails, the Errors page lists each error_code and what to do next. A stale or incomplete quality result means narrowing, delaying, or marking the output, never filling in missing rows.

Next step

Start with Pull historical market data for your first bounded pull.
Last modified on October 6, 2026