It is the other half of this tool: bulk history from the exchanges' own archives, planned before it downloads anything, verified by checksum, resumable through a ledger. The design is specific because the archives were surveyed rather than assumed:

• data.binance.vision — S3 XML listings, 1 000 keys a page, sha256 sidecars, UTC days. Its metrics and bookDepth datasets exist only as daily files and fundingRate only as monthly ones.
• Binance's spot CSVs switched from millisecond to microsecond stamps on 2025-01-01; its futures CSVs did not. Spot and futures aggTrades also disagree on header presence, column count and how they spell a boolean.
• public.bybit.com — HTML indexes, gzip, fractional unix seconds, and values in scientific notation.
• okx.com/cdn/okex — no listing at all, CRLF line endings, and its “day” runs on UTC+8: one UTC day needs two files, both trimmed.
• hyperliquid-archive — S3 requester-pays, which bills the caller, so it needs an explicit opt-in before a single byte moves.

Until it is built, a gap stays a gap: Subscriptions reports one rather than filling it.