Configuration

Local settings stored in threatlens.conf. Saved changes apply to the next ingest, from here or the command line, without a restart.

Ollama

Server the LLM engine sends verification requests to. Batch size, context window and timeouts are in the file's [llm] section, which the calibration tool writes (python -m threatlens.extractors.calibration); saving here keeps them.

Server URL

Models

Test the connection above to pick from the models installed on that server.

Verifier Checks each extracted candidate.

Profiler Classifies the whole report's scope. Blank uses the verifier model.

Database

The db/ file the dashboard opens at startup. Currently viewing ; use the header dropdown to switch now.

Default database

Quality weights

How much each 0-5 dimension counts toward an indicator's 0-100 quality score. Each between 0 and 1, and together they must total 1.00. Overrides the weights in the database's scoring_config table.

Total: -

Verifier agreement

Confidence multiplier applied when the Python and LLM verifiers agree or disagree to each degree. 1.00 means no discount.

Re-score existing indicators

Saved weights apply to newly ingested reports. To apply them to indicators already in , re-score them. This also refreshes timeliness against today's date.