TypeSafe Jev

Navigator’s prompt-time classifiers can defer to TypeSafe’s Jev , a typed-decision model that returns calibrated probabilities instead of text. One request per prompt, eight questions, decisive answers override the keyword heuristics, everything else falls back. Off by default because the prompt text is sent to the API.

Setup, config keys, measured accuracy and the privacy boundary are on the configuration page: Typed Prompt Judge.

Quick version:

export TYPESAFE_API_KEY=...          # or write it to ~/.config/typesafe/api_key (chmod 600)

Then say “enable judge” and run python3 hooks/nav_hook_lib/judge.py --check from the plugin root.