# Context Pack: run_20260430_032930_188986_dynamic_role_switching_20260430_094833_robustness

## Candidate

- candidate_id: dynamic_role_switching_20260430_094833
- candidate_kind: strategy
- candidate_status: archived
- status_reason: Archived by reseed rebuild; not in canonical 20260430 reseed pool.
- git_ref: faebc78a52031975fa3e831a09dd27018736026e
- stage: robustness
- status: failed

## Overall Result

- games: 0
- win_rate: 0.0
- tie_rate: 0.0
- loss_rate: 0.0
- points_per_game: 0.0
- avg_score: 0.0
- score_std: 0.0
- error_count: 0
- timeout_count: 0
- promotion_score: 0.35

## Layout-Wise Summary

| Layout | Games | Win Rate | PPG | Avg Score | Error Rate | Timeout Rate | Promotion | Eligible | Active Rank | Champion |
|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|

## League Decision

- No league decision event recorded for this run.

## Opponent-Wise Summary

| Opponent | Games | Win Rate | PPG | Avg Score |
|---|---:|---:|---:|---:|

## Observed Weakness Heuristics

- No deterministic weakness heuristic was triggered.

## Request

```json
{
  "candidate": {
    "agent_ref": "your_best",
    "candidate_id": "dynamic_role_switching_20260430_094833"
  },
  "git_ref": "faebc78a52031975fa3e831a09dd27018736026e",
  "layouts": [
    "tinyCapture",
    "defaultCapture",
    "alleyCapture",
    "mediumCapture",
    "strategicCapture"
  ],
  "max_workers": 3,
  "notes": "Top-K robustness probe for current best dynamic_role_switching planning candidate before further code changes.",
  "num_games_per_pair": 2,
  "opponent_selector": {
    "candidate_kinds": [
      "strategy"
    ],
    "eligible_statuses": [
      "promoted",
      "active",
      "champion"
    ],
    "include_baseline": true,
    "include_global_best": false,
    "mode": "layout_top_k",
    "opponents": [],
    "ranking_metric": "promotion_score",
    "top_k": 5
  },
  "run_name": "dynamic_role_switching_20260430_094833_topk_probe",
  "stage": "robustness"
}
```

## Next Codex Task

Propose one controlled revision targeting the weakest layout or highest error/timeout source, then rerun smoke or candidate_eval as appropriate.