egorfedorov Auto Balancer
Use this skill to run controlled parameter tuning loops with deterministic validation gates.
Workflow
- Define balancing contract.
- Declare target metrics, tolerances, hard constraints, and stop conditions.
- Declare which parameters are allowed to move and their bounds.
- Establish baseline and iteration plan.
- Record baseline metrics before tuning.
- Apply small, traceable parameter changes per iteration.
- Track config hash/version for each run.
- Run balance loop.
- Execute simulation/evaluation runs.
- Compare observed metrics to targets and compute deltas.
- Keep only changes that improve objective without violating hard constraints.
- Validate gate conditions.
- Check each metric against tolerance range.
- Fail immediately on hard-constraint breaches.
- Require minimum run count before final pass.
- Prepare sign-off handoff.
- Return final parameter set, metric table, and failed/passed gates.
- Include patch plan and exact verification commands.
Commands
python3 scripts/validate_balance_runs.py \
--input <path/to/balance_runs.json> \
--spec <path/to/target_spec.json>
Treat non-zero exits as blocker results.
Output Contract
Return:
Target Contract: metrics, tolerances, and constraints.Run Summary: baseline, best run, and final run deltas.Gate Results: pass/fail per metric and per hard constraint.Patch Plan: exact files/params to update.Residual Risks: unresolved drift or instability concerns.
References
references/workflow.md: balancing process and iteration order.references/metric-rules.md: tolerance and hard-constraint rules.references/signoff-template.md: balancing sign-off template.
Execution Rules
- Keep balancing changes bounded and reversible.
- Keep hard constraints non-negotiable.
- Keep baseline comparison in every report.
- Flag non-convergent loops as blockers.