Open-source agentic AI for transaction monitoring. Just ask Jooby.What's Jube? →
jube
  • Documentation
  • Configuration
  • Exhaustive Adaptation
  • Exhaustive Data Sources

Exhaustive data sources

Every ExhaustiveSearchInstance has a DataSourceId (Jube.Dto.ExhaustiveSearchInstance.ExhaustiveDataSource) deciding what Jube.Engine/Exhaustive/Stages/SampleStage.cs samples from when the instance trains:

  • Archive (0) — the default. The instance samples from the entity analysis model’s own archived transaction history, the same data it will recall against in production.
  • DemoDataset (1) — MockDataLoader generates a synthetic dataset for the model instead. Useful for a walkthrough or a sales demonstration where no real transaction history exists yet. This per-instance choice replaced the old global UseMockDataExhaustive environment variable, which is gone from the code and the configuration surface.
  • UploadedDataset (2) — the instance trains against a confirmed CSV or JSON file uploaded through the dataset panel; see Exhaustive uploaded datasets for the upload/map/confirm flow.

Preservation impact

An export carries an uploaded dataset’s rows (ExhaustiveInstanceCsv/ExhaustiveInstanceJson in the preservation payload) alongside the owning ExhaustiveSearchInstance whenever the export’s Exhaustive option is set — the same option that carries the rest of the Exhaustive tree. The instance’s ExhaustiveInstanceCsvId and ExhaustiveInstanceJsonId foreign keys are remapped to whatever id the dataset gets on import, exactly like every other Exhaustive foreign key.

If the instance’s DataSourceId is UploadedDataset but the dataset itself was never exported (for example a dataset still mid-upload and not yet confirmed when the export ran), the import cannot resolve the missing foreign key. Rather than fail the whole import over one instance’s unavailable dataset, the importer nulls ExhaustiveInstanceCsvId/ExhaustiveInstanceJsonId and resets DataSourceId back to Archive on that instance, so it imports cleanly and falls back to sampling from the archive the next time it trains.