NewTypeSafe AI Community Playground
A community playground for typed decisions across classification, code review, browser tasks and games.
Added to Jevfast
NewHow it uses Jev
For an eligible pull-request review unit, the code sends Jev the PR title and description, target hunk, relevant impact graph, activated policies, deterministic findings, and a flag that the analysis is mocked. Jev selects one fixed governance outcome. The code validates the returned choice and records its confidence and selected-choice probability when they are valid numbers.
What Jev decides
Illustrative governance outcome requestExample answers · not a recorded Jev response · Source ↗
Question 1 · outcome
YOUR APP
INSTRUCTION
Classify the target's specified focus using only the supplied graph, policies, and hunk. Choose one outcome; use human review or more context when uncertain. Do not override deterministic findings or decide whether to merge.
STATE
One complete, eligible review unit with a relevant hunk, filtered impact graph, activated policies, deterministic findings, and mockAnalysis set to true. The graph and analysis are explicitly mocked; this example does not represent analysis of the full repository.
JEV · CHOICE
- safe_change
- possible_breaking_change
- missing_test_coverage
- security_or_policy_risk
- needs_human_review
- needs_more_context
App workflow
- Select review units
Blocked analyses produce no Jev units. Otherwise, complete hunks in production or generated files, or hunks with repository-policy findings, are split into compatibility, test-coverage, or general-change focuses.
- Check the target
Before calling Jev, the code confirms that the hunk is complete and the requested unit belongs to the generated review units.
- Validate the outcome
The code checks that Jev returned a choice from the allowed governance outcomes, then records the unit, outcome, source, and valid confidence and probability values.
Why it is interesting
The example makes Jev one bounded part of a guarded review flow. Deterministic checks retain control of hard blocks and findings, while Jev can request more context or human review and cannot make merge decisions.