claude-jev-funnel
Claude Code plugin and CLI for the Jev funnel: judge items in bulk, settle confident ones in code, send only uncertain ones onward.
Overview
claude-jev-funnel packages one pattern, the Jev funnel, as a Claude Code skill and a zero-dependency Python CLI. Code collects items such as reviews, logs, diffs, tickets or warnings; jev.py sends them in bulk with typed yes/no, pick-one or rubric questions; confident answers (YES, NO, ACT, CONFIDENT) are handled in code with no further model call; and only the uncertain band (UNCERTAIN, REVIEW, ABSTAIN, UNSURE) goes to Claude or a human.
It frames Jev as closer to a programmable if for judgments plain code cannot make. It cites list pricing and the official parallel-questions cookbook, reports real usage from every response, and includes a live benchmark of batch mode in the repository.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Yes/no judgment per item | YES · NO · UNCERTAIN |
| Choice | Pick-one judgment per item | — |
| Score | Rubric position per item | — |
How it works
- Collect items in code
- Send them in bulk with typed questions
- Handle confident answers in code
- Route the uncertain band to Claude or a human
Strengths
- Most items never reach the LLM
- Zero dependencies
- Usage reported per response
Caveats
- Band thresholds need tuning per task
Getting started
Install the plugin in Claude Code or run skills/jev/scripts/jev.py directly with TYPESAFE_API_KEY set.At a glance
- Type
- Plugin
- Status
- beta
- Language
- Python
- Runtime
- Claude Code skill + CLI
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Claude Code plugin + zero-dependency CLI for TypeSafe's Jev: judge items in bulk with calibrated yes/no, pick-one and rubric answers; handle the confident ends in code, review only the uncertain band.
logicrw/awesome-jev-projects
Frequently asked questions
What is the funnel?
Bulk-judge items, settle confident ends in code, and send only uncertain ones onward.
Does it need dependencies?
No, the CLI uses only the standard library.
Who reviews uncertain items?
Claude or a human.
Related projects
Research demo where Jev answers plain questions about text and a small learned head adds new questions from human feedback.
ProjectresearchPythonAgent skill that tells you when you need an eval and how to build it, with an opt-in lane routing closed-label residue to Jev.
BenchmarkbetaPythonPre-registered audits of AI judges that measure how many decisions a judge can take alone at a given error rate, with calibration.
BenchmarkbetaPythonpytest-style tool that pins fixtures and expected Jev answers, then fails a model upgrade when answers flip or confidence drops.
ProjectbetaPythonWrite-up testing whether Jev beats small fast LLM tiers on public and private classification, as a classifier or a filter.
ProjectresearchPythonHarness that runs Jev on all 19,646 TMMLU+ Traditional Chinese questions as four-way Choices, reporting accuracy and calibration.
ProjectresearchPython