XRP Ledger

Ten Parquet tables with 256 documented columns in total: transactions_tx, transfers_tx, payments_tx, offers_tx, nftoken_offers_tx, ripple_states_tx, escrows_tx, checks_tx, account_roots_tx and balances. Covering the most recent month to yesterday, refreshed daily.

delivered_value, never amount_value. On XRPL a partial payment sets amount_value as a requested ceiling rather than what moved, and delivers less. Summing amount_value overstates native XRP volume by roughly 27,000,000x on real data. The partial column flags those rows. This is the single most common way an XRPL volume figure goes wrong.

Yes, deliberately — failures are signal for fee and congestion analysis. transactions_tx carries result (the XRPL engine code) and success. Filter success = 1, equivalent to result = 'tesSUCCESS', for settled activity only.

XRPL encodes issued-currency codes longer than three characters as 40-character hex. For example 4D656F7752500000000000000000000000000000 decodes to MeowRP. Three-character codes such as XRP or ELS appear as-is. Decode hex to ASCII and strip trailing nulls to get the display symbol.

It depends on the table, and the difference matters. account_roots_tx stores balance and prev_balance as strings in drops — divide by 1,000,000 for XRP. The balances table stores them as floats already in XRP. The fee column in transactions_tx is also a string in drops. We kept each table faithful to its source rather than silently converting.

In ripple_states_tx the balance is signed from low_account's perspective: positive means low_account holds the asset, negative means high_account does. Take the absolute value and pick the holder by sign before aggregating issued-token supply.

Because one XRPL transaction can move value several ways. The direction column classifies each row as payment, fee, trade, nft_trade, mint or other. Filter on it rather than assuming one row per transaction, and note that fee rows will double-count if you add them to payment volume.

offers_tx carries both the current and prior state: compare pre_taker_gets_value and pre_taker_pays_value with taker_gets_value and taker_pays_value. DeletedNode rows keep the offer's final state rather than being zeroed, so a consumed offer still tells you what it was worth.

finish_after and cancel_after use the Ripple epoch, which starts 2000-01-01Z. Add 946684800 to convert to Unix time. Every other time column in the dataset is already Unix: tx_time in seconds, tx_date in milliseconds.

As flat Parquet files under a stable S3 layout, one prefix per table, with a JSON manifest listing every file and its sha256. Signed HTTPS links are emailed to you once the files are prepared.

Refreshed daily with T+1 latency. A purchase made today includes everything up to yesterday.

Yes. Every table links to a public sample file with real records, and the sample bucket holds full Parquet files you can load directly. No email address required.