LAST RESORT — arbitrary READ-ONLY SQL against the Cardano databases (cardano, cardano_flow). The ada_* tools are the PRIORITY; use this ONLY when none can answer. No query optimizer here: filter on the per-key tables and NEVER JOIN big tables (use IN (SELECT …)); never GLOBAL IN (it reads the whole table). Runs read-only ClickHouse SQL. SQL reference: https://clickhouse.com/docs/reference/home This data runs on an older release of the engine: some functions and syntax in that reference (map() and toJSONString() among them) do not exist here — when a statement is refused for an unknown function or setting, use the forms shown in this description. DATA MODEL: UTXO model with native tokens — EVERY table has ONE ROW PER ASSET: currency_id (native ADA = 7029218, token_index = -1) and value UInt64 in RAW units. ALWAYS group by currency_id and divide by dictGetFloat64('currency','divider',toUInt64(currency_id)) (1e6 for ADA; 1, i.e. raw units, for a token without registered decimals and for an unknown id); a plain sum(value) adds lovelace to token units and is meaningless, and sum() of large token values can overflow — sum toFloat64(value). TOKENS: token identity via the currency dict: 'symbol' / 'name' (a ticker or the hex asset name; '-' for an unknown id), 'asset_id' (fingerprint asset1…), 'decimals', 'properties' (YAML text; policy id = extract(dictGetString('currency','properties',toUInt64(currency_id)), 'policyId: ([0-9a-f]{56})')). Tickers are not unique — match tokens by fingerprint or policy id. A dictGet in WHERE runs for EVERY row read (no short-circuit), so filter by address / tx / currency_id first. Addresses are plain strings (addr1… bech32 lowercase, stake1… reward accounts, Byron base58 Ae2… / Ddz… case-sensitive); a script reward account (stake17… on explorers) is mostly recorded in its stake1u… spelling, whose checksum differs — look such an account up with ada_address_profile or ada_sent_from_address, which match both spellings. stake_key = the staking credential of a base address — on enterprise addresses (addr1v… / addr1w…) the column holds the payment credential re-encoded as stake1u…, which is NOT a staking key. Tx id: tx_id_bin = unhex('<64hex>'), output hex(tx_id_bin). Time = tx_time. Date-times come back as 'YYYY-MM-DD hh:mm:ss' UTC; select formatDateTime(tx_time, '%FT%TZ') for ISO-8601 (2026-09-21T15:04:18Z), the shape the other tools return. TABLES: db cardano, partitioned by month on tx_date (add a plain tx_date bound whenever you can — inside … OR … it prunes nothing): inputs_by_address / outputs_by_address (key address; inputs carry output_tx_id_bin / output_index of the spent output and source Enum input/withdrawal/stake_deregistration; outputs carry direction — only not_change / unknown on ordinary outputs: change is NEVER labeled, so not_change includes change back to the payer), inputs_by_tx / outputs_by_tx / transactions (fee_value, deposit_value, metadata) / mints (key tx_id_bin — pass (tx_date, tx_id_bin) IN (…) when the date is known, otherwise every month is probed; to reuse one lookup in several UNION ALL branches write (tx_date, tx_id_bin) IN identity((SELECT groupArray((tx_date, tx_id_bin)) FROM cardano.transactions WHERE tx_id_bin IN (…))), which runs once; toInt64(value) is negative for a burn), blocks (key block). Deposits are already rows: a stake registration is an output to the stake1… address (direction stake_registration), other protocol deposits an output with an empty address (deposit), a refund an input (stake_deregistration) — per transaction ADA inputs = outputs + fee_value and, per token, inputs + mints = outputs; do not add deposit_value on top. db cardano_flow: tx_inputs / tx_outputs keyed (block, tx_index), not partitioned — the cheap way to expand many transactions: (block, tx_index) IN (SELECT block, tx_index FROM cardano.inputs_by_address WHERE address = '…') (output_type commission / spend — never change; the commission row has an empty address). This copy is INCOMPLETE: it misses some transactions (a few hours of blocks in March 2023 among them) and most outputs of a token minted in the same transaction — use cardano.*_by_tx for exact per-transaction totals. address_transfers (AggregateFunction columns → -Merge; its amount columns ADD ALL ASSETS together and its stake_key is empty — use it for counts only). Address labels: none in this data — use labels_for_addresses(chain='cardano'). LIMITS: read-only; returns the rows of your query. 64-bit integers come back as JSON numbers, exact only up to 2^53: return a hash such as cityHash64(...) or any other 64-bit id as toString(x). A result of more than 25,000 rows is not cut off - the whole query fails with an error - so aggregate (GROUP BY, count) or add a LIMIT of at most 25,000.