Where each tier of data lives and how much of it there is.
Create or upgrade every table, and import a blob
_catalog.json catalogue into the relational store. Run it once after
attaching a database to a deployment. The market-data CSVs are never touched — they
stay in blob storage. Re-import overwrites the SQL catalogue from the blob one even
when SQL already has rows.
Convert any dataset still on a legacy blob layout to chunked, gzipped blobs. Every dataset stored from now on is written that way already, so once the pass has completed this reports nothing left to convert and touches blob storage not at all.
Refresh the datasets the Telegram watches read, send what needs sending,
then spend whatever time is left refreshing the rest of the catalogue. This is the
job a scheduler fires at /api/telegram/tick; a run is idempotent, so a
watch that has already spoken about a bar stays quiet unless its situation changed.
The user signs in at the terminal with these credentials.
“New password” means the account still has the password it was created or reset with.
Optimise lets the account run a parameter sweep on this server rather than on its own PC. A sweep runs the backtest hundreds of times, so it is off for every account until you turn it on here, and the budget shown under Deployment caps each run and each account's day. The cell reports what the account has spent of today's allowance; it refreshes at 00:00 UTC.