Fast and balanced answer different questions
The pinned tests give solo drafts their own stage and stands-alone question. A first accepted isolated draft is not equivalent to comparing all completed plans.
Inspect the supporting source ↗REPOSITORY ANALYSIS
jev-planner
A multi-agent planning coordinator: generative agents draft and review plans, while Jev scores them and selects whether to review again or choose a finalizer.
E2 · source reviewedPublic material or source evidence was reviewed. A fixed-version source review requires checking a pinned revision; this is not a reproduction, performance endorsement, or safety guarantee.
- Stars
- 0
- Forks
- 0
- License
- MIT
- Last commit
- 2026-09-23
What Jev does here
Jev evaluates completeness, feasibility, risk coverage, stronger plan, finalizer and whether another pass is needed; code controls the fast, balanced and ultra workflows.
Makes review rounds and finalizer choices explicit instead of letting an unconstrained text response drive the planning loop.
Fixed-version documentation and implementation material were inspected on 2026-10-24. No planning agents or paid calls were executed. Source tests validate question construction, not the quality of final plans.
What task does this project fit?
Coordinate several planning agents without asking Jev to write the plans.
Material analysed：
From input to action
- Agents draft independently; Jev evaluates completeness, feasibility, risk coverage and whether another review is warranted.
- Mode and round limits decide whether to cross-review.
- Jev chooses a finalizer; a generative agent merges plans unless an existing plan can stand alone.
Implementation details that matter
Constrained judgments still need workflow policy
jev.ts maps planner questions to Choice, Score and Noul. It returns structured verdicts; mode, review-round caps and finalizer overrides remain outside the model.
Inspect the supporting source ↗When to choose it
Useful when multiple drafts expose real tradeoffs. Select fast for early acceptance and balanced/ultra for broader review, accepting that more agent calls change cost and latency.
How to inspect it yourself
Compare solo, drafts and reviews cases in jev.test.ts, then trace to toTypeSafe in jev.ts. Check that the offered finalizer choices correspond to the configured agent set rather than arbitrary model-generated names.
Limits of this review
- No planning agents or paid calls were executed.
- Source tests validate question construction, not the quality of final plans.