Judge question review
Review a question before Jev answers it: two judgments in one, inputs no option fits, a rating hiding in a yes/no.
A badly designed question still gets a confident answer. Published tests found pick-one questions with no "none of these" option sending out-of-scope inputs to an option at 0.99 confidence. This judge reviews the parts of a question that need judgment: is it really two questions in one, would some realistic inputs fit none of its options, is it a rating asked as a yes/no. Each check was calibrated on questions from Evally's own judges and on deliberately broken ones; Evally's judge builder runs it when you press Check with Jev.
Try it live
Real calls to Jev. Pick an example or paste your own.
Pick an example or paste your own input, then run it.
Under the hood
Every question and rule, in the open. Fork it to change any of them.
What it asks
Reads question, options and data, then asks 3 questions in one call.
- Q1Two questions in oneYes / no
Does the
questionask for two or more separate judgments at once, so that one answer couldn't say which part was true? - Q2Inputs no option fitsYes / no
Picture the different inputs this
questionwill be asked about. Would some realistic ones fit none of theoptions, so that a wrong option would have to be picked? If theoptionsaren't a pick-one list, or one already means "other" or "none of these", answer no. - Q3A rating asked as yes/noYes / no
Does the
questionreally ask how much, how well or how likely, so it ought to be a rating instead of a yes/no? If theoptionsare a pick-one list or rating levels, answer no.
How it decides
Rules run in code, top to bottom. The first one that matches gives the verdict.
- 1If Two questions in one ≥ 0.8Rewrite
- 2If Two questions in one ≥ 0.7Improve
- 3If Inputs no option fits ≥ 0.75Improve
- 4If A rating asked as yes/no ≥ 0.6Improve
- ·DefaultGood to go
Call it from your code
Every template runs as-is with an API key, no setup: POST /api/v1/judges/question-check/run.
curl -X POST https://tryevally.com/api/v1/judges/question-check/run \ -H "Authorization: Bearer $EVALLY_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "input": { "question": "Is this ticket urgent and about billing?", "options": "Yes or no.", "data": "ticket: the customer'\''s message" }}'Seen in the wild
Community builds on Jev that use this pattern. Not affiliated with Evally.
More in Engineering
- Engineering
PR risk review
Is this pull request safe to merge, or does it need a careful human look?
Safe to mergeNeeds careful reviewChanges requested7 questions - Engineering
Semantic code lint
Lint code against conventions written in plain English, the rules a regex can't express.
CleanWarningsViolations5 questions - Engineering
Commit message check
Does the commit message match the staged diff, with no secrets or debug leftovers?
Looks goodAmendBlock5 questions - Engineering
PR labeler
Label pull requests by type and area, and flag what belongs in the changelog.
FeatureBug fixRefactor+64 questions - Engineering
Package risk check
Catch typosquats and hostile install scripts before an install runs them.
AllowReviewBlock7 questions - Engineering
Vulnerable code check
Flag injection, path traversal, hardcoded secrets and other common holes in a change.
CleanReviewVulnerable4 questions - Engineering
Should this be a judge?
Paste a prompt from your code. Find out whether it's really a decision a judge would answer faster and cheaper.
Use a judgeSplit itKeep the LLM4 questions