The honest answers.

Completed, certified UTC days of market data: day-partitioned Parquet files — order books, trades, market metadata and settlement price streams — one file per venue / channel / scope / day, each with a SHA256 manifest. You buy access, download the files, and can verify every byte.

The datasets page lists every venue with its live stage: regulated prediction exchanges, global CLOB venues, an on-chain prediction game, the crypto reference exchange (spot and futures), and the oracle stream that settles crypto up/down markets. Each venue page documents every channel to the column.

Standard Apache Parquet. Anything that reads Arrow reads the archive — pandas, polars, DuckDB, Spark, R, ClickHouse. No SDK required; the API is four GET endpoints with bearer auth.

A completed UTC day becomes available after its nightly certification passes. We sell certified history, not a live feed — and we never promise delivery faster than the measured certification time.

Before a day is sold it passes five ordered gates: coverage gates against expected market windows, on-chain trade reconciliation, artifact audits (schema, timestamp walls, value domains, manifest counts and hashes), durable-mirror verification, and a gapless promotion gate that refuses day D unless D−1 is released. A day that fails any gate is recovered and re-run.

Every file ships with a manifest carrying its SHA256, row count and schema fingerprint. Re-hash the file yourself and compare. The verification page walks through it — it is three shell commands.

The day goes red and is not sold. We recover it and re-run the full chain. If a correction is ever applied to released data, it appears in the public corrections ledger with its reason — never a silent rewrite.

When a venue provably never listed an expected window, we record it as venue_unlisted backed by collection-time evidence, and the gap is disclosed in the data rather than patched over. Venue silence and our failure are never conflated.

The day its own data passes certification. Venues shown as "certifying" or "collecting" are being captured daily, but their history only becomes sellable from their own certified start — never backfilled from uncertified collection. Their dataset pages show dated collection totals in the meantime.

Yes. Coverage requests are prioritized for paying customers and scoped explicitly on Enterprise. Tell us what you need and we will tell you honestly whether and when we can certify it.

Yes. The catalog, availability index and full schema docs are open without a key, and signing up gets you 2 file downloads per day from the full archive, free forever — enough to evaluate real data and integrate your whole pipeline, including hash verification, before paying.

Every key has a daily download budget that resets at 00:00 UTC. Authorized requests count; failed auth does not. Exceeding the budget returns 429 daily_budget_exceeded and never silently charges. The pricing page maps each tier’s budget to what it means in archive-days per day.

Internal research, modeling, backtesting and derived aggregate results are covered. Redistribution of the data itself — whole or partial, raw or identifiably derived — requires a separate written agreement. See the data license.

Negotiated budgets or bulk object-storage sync, invoicing, an SLA, priority coverage requests, and redistribution licensing where appropriate. If you are syncing the whole archive daily, this is the tier built for it.