PPO KAGGLE · 381 TASKS
These charts come from the 3-shard PPO + LoRA training we ran on free Kaggle T4s. Source data:
kaggle ran notebooks/shard {1,2,3}/training_kaggle{N}.json + the 381 scenarios in
rl-agent/scenarios/sim/{easy,medium,hard}/*.json, pre-bundled into
rl-agent/showcase_data.json by scripts/build_showcase_data.py.
Every visible number is computed from those files.
AWS scenario coverage · PPO mode
381 sim scenarios ship across 4 AWS service families plus generated chaos profiles.
Service-family coverage
| Family | ID prefix | What it teaches |
|---|
| Lambda | sim_easy_lambda_throttle* | Concurrency hits reserved cap; raise reserved-concurrency. |
| DynamoDB | sim_easy_ddb_throttle*, sim_hard_ddb_chain* | Hot partition + GSI; scale WCU vs. fix key skew. |
| API Gateway | sim_easy_apigw*, sim_med_apigw_lambda*, sim_hard_apigw_chain* | Stage-variable drift, integration timeouts, multi-stage cascade. |
| EventBridge | sim_easy_eb*, sim_med_eb_lambda* | Rule mis-routes, FailedInvocations, two-hop chain. |
| Step Functions | sim_med_sfn_lambda* | Catch vs. Retry semantics in execution history. |
| IAM | sim_hard_iam_chain* | Missing assume-role two services deep. |
Configuration
USE_AWS=true
AWS_REGION=us-east-1
AWS_PROFILE=incident-commander-rl
# Optional: pre-provisioned sandbox via scripts/aws_inventory.py --provision