LEGACY DATASET
These charts come from the kube-sre-gym-style heuristic + early notebook runs — the 11 hand-curated tasks in
rl-agent/scenarios/{easy,medium,hard}/*.json, recorded into
rl-agent/checkpoints/<run>/metrics.jsonl and
colab/logs/reward_breakdown_history.jsonl. They do not include the 381-task PPO Kaggle run.
Adversarial scenario designer
Designer is procedural fallback.
Source: training snapshot (5 scenarios). Showing last 5 scenarios.
Generated scenarios
| Base task | Faults | Difficulty | Red herrings |
|---|
| task3 | decimal_corruption, redis_pool_exhaustion | hard | ["postgres-vacuum", "redis-primary"] |
| task6 | tls_expiry, dns_resolution | hard | ["service-mesh", "coredns"] |
| task7 | configmap_race, cache_poisoning | hard | ["redis-cache", "gc-pause"] |
| task11 | liveness_regression, image_pull_backoff | hard | ["scheduler", "kubelet"] |
| task2 | payments_oom, kafka_lag | medium | ["broker-2", "zookeeper"] |
How it works
POST /adversarial/design
{
"base_task_id": "task2",
"difficulty": "hard",
"mix_faults": true,
"add_red_herrings": 2
}