Workflows

Nothing in this system asks a vendor a question while you wait. Every value you see was written earlier by one of the jobs below, which is why the question to ask of any number is how old it is.

The data model records the owning job for every column. These pages describe those jobs.

Pages

• Scheduled jobs. What is on a timer, what it writes, and which of them are actually firing.
• Imports. Card text, printings, sets, precons, products and the rules corpus, and which vendor owns which.
• Tagging and embeddings. The passes that turn imported text into concept associations and vectors.

If you are looking for how to get your own data in, that is importing a collection. For what the price jobs produce and how to read a price timestamp, see where a price comes from.

What is not here

These pages cover every job that writes data you can read. Four kinds of script are deliberately left out, because none of them changes what the app shows:

• Evaluation and audit. The suites that score the AI against known answers, and a pass that critiques recorded pipeline traces. Described where they matter, under grounding and limits.
• One-time backfills. Scripts written to correct historical rows once and not run again.
• Repository tooling. Build-time asset generation, environment checks, and the prose linter for these pages.
• Shared helpers. Index and vacuum handling used by the import jobs rather than run on their own, described in imports where they affect what a rebuild does.

Reading a timestamp

Every imported value carries the time it was written, and that timestamp is the honest answer about freshness. A schedule says what was meant to happen. A job that has not run looks exactly like one that has, apart from its timestamp.