What problem does it solve?
This Skill helps design zicato evaluation boards that produce meaningful score differences between champion and challenger harnesses, avoiding all-pass, all-fail, constant-drift, and non-differentiating evaluations.
Core Features & Use Cases
- Behavioral Coverage: Select entries that exercise distinct regression-sensitive behaviors rather than repeating similar inputs.
- Discriminating Evaluation: Tune expectations, continuous scores, weights, and sensitive-band difficulty so tournaments can identify genuine improvements.
- Entry Strategy: Choose appropriately among single-turn, scripted multi-turn, and emulated multi-turn tests while accounting for cost, determinism, and emulator noise.
- Board Controls: Apply board-wide drift suppression and judge-only evaluation appropriately, and prevent emulator collusion through distinct callables.
- Use Case: When a zicato evolution loop runs successfully but promotes nothing, use this Skill to diagnose dead-weight entries and redesign the board around measurable behavior differences.
Quick Start
Ask the zicato board design skill to create a discriminating evaluation board for the harness behaviors you need to improve, including suitable entry types, expectations, weights, and drift settings.
