v4 · architecture
Evaluator–Optimizer
generate → critique → revise, until it passes
Evaluator–Optimizer architecture
v4 evaluator–optimizerA generator produces a draft; a critic scores it (0–10) against a rubric and lists concrete issues; the generator revises. The loop repeats until the critic passes (score ≥ 8) or a max-rounds cap. Best for a single artifact you want iteratively improved — a draft, spec, or piece of analysis.
Generator
Critic
generate → critique → revise, until it passes
How it works
- 1A generator produces a first draft of the deliverable.
- 2A critic scores it 0–10 against a rubric and lists concrete issues.
- 3The generator revises, addressing each issue.
- 4The loop repeats until the critic passes (score ≥ 8) or a max-rounds cap.
When to use
A single artifact you want iteratively improved to a quality bar — a draft, spec, or snippet.
Trade-off
Cost grows with each round; a never-satisfied critic can burn the full round budget.
Agents
GeneratorProduces and revises the deliverable
CriticScores against a rubric and lists issues
Note
A demanding critic + a concrete rubric is the whole game. Vague rubrics give vague gains; specific, actionable issues drive real improvement round over round.