Skip to main content
Engines says what an engine is and how to choose one. Use this version with "engine": {"name": "jev", "version": "1.14.0"}. This version never changes. Your judgment uses exactly this version until you change it.

​
Limits

Each request to this version carries one or more questions and the compiled context of one document, which is what the judgment’s context recipe selects from the document and its related documents. Probabilities come in steps of 0.01. A bool’s p is one of 101 values: 0, 0.01, 0.02 and so on, up to 1. The probabilities of a choice or a score also come in steps of 0.01, and Brussle then scales them so that they add up to 1. Many documents can have the same p, and the correction gives each p one corrected number, so it cannot separate them. A correction with features can, because documents with the same p can have different features. You pay for judging on this version per judgment, by size class, times the weight of its engine. The weight of jev is 1, so a standard judgment on this version counts as 1 judgment.

​
Changes and retirement

An exact version of Jev never changes, so it has no epochs. Each answer records the version in engine_version, for example 1.14.0.
  • Retirement. If we retire an exact version, we send an email to the owners and admins of each organization that judged with it in the last 60 days.
  • When TypeSafe AI stops serving a version. Judging with that version fails with engine_version_unavailable. We never move a judgment to a different engine version for you.

​
Data handling

We send your data to the engine provider’s API in the cases below, each to the same provider on the same terms.
  • Judging. For each document that this engine judges, we send the document’s compiled context and the question of each judgment.
  • Recipe tuning. If your plan includes recipe tuning, we send smaller versions of the stored compiled contexts of answers that have recent outcomes, with the question of each judgment, and use the results to test cheaper context recipes.
  • A judgment moved to this engine. When you move a judgment to this engine, we send the stored compiled contexts of its answers that have recent outcomes, as they were stored, with the question of each judgment, and use the results to fit its correction on this engine.
  • Alternatives. If your plan includes alternatives, we send stored compiled contexts of your labelled answers to this engine with the question and criteria of an alternative, and use the results to score that alternative on your outcomes. A recipe alternative can also send contexts that we compile from today’s documents. A namespace’s contexts come to this engine only when the namespace uses this engine, when an org admin sets alternatives_engines to the organization’s engines and another namespace uses it, or when you name this engine as an alternative. We bill these calls as judging.
After you delete a document, we do not send its stored contexts again. The data processing agreement lists each subprocessor that your data passes through, and each subprocessor’s terms in that agreement set how long it keeps your data (see where your data goes).

​
Conformance

The conformance suite measures the quality of an engine version by asking it a fixed set of questions. Each question is about one document, and people labelled its correct answer. This version ran the suite before it became active, and while it is active, GET /engines returns the results of its latest run with the run’s date. Neither suite shows how well the engine answers your own questions, so let your outcomes compare engines before you choose.

​
The relational suite

The relational suite checks the engine’s quality with questions about documents that the engine reads with their related documents. This version never changes, so it runs the suite once, before it becomes active. GET /engines does not return its results.