GPT Researcher — Jev context filtering

What it does

Jev rates usefulness, not truth. Without a TypeSafe key, automatic mode uses local keyword ranking. A successful Jev filter can still discard every passage; the report writer then declines to write if no context remains.

How you can use it

If your project already uses GPT Researcher, ask a developer to compare its Jev and keyword filters on questions with known useful passages. Jev rates each passage on four levels. Check what the final report misses as well as what the filter keeps; usefulness does not establish truth.

Automatic mode selects Jev when a TypeSafe access key is present, sending the question and scored passages to that outside service. Errors handled by the filter use keyword ranking, but a successful empty selection does not. Plan for missing evidence rather than asking the writing model to invent it.