| { |
| "composition": { |
| "categories": { |
| "generalized-cpp-global": 6 |
| }, |
| "families": { |
| "clock": 1, |
| "complex-numbers": 1, |
| "crypto-square": 1, |
| "grade-school": 1, |
| "kindergarten-garden": 1, |
| "perfect-numbers": 1 |
| } |
| }, |
| "counts": { |
| "available_shadow": 6, |
| "monitor": 6, |
| "train": 6, |
| "validation": 0 |
| }, |
| "files": { |
| "grpo_train": "grpo/train.jsonl", |
| "train_monitor": "eval/train_monitor.jsonl", |
| "validation": "eval/validation.jsonl" |
| }, |
| "kind": "aider-polyglot-cpp-shadow-grpo", |
| "profile": "generalized-cpp-grpo20", |
| "prompt_sha256": [ |
| "561456e93280959b222cfd8448e3bf9d61e67cf94fc736971f5d0ef640023e22", |
| "5cff42213c939d6eb582c942757b3afabaebb3a17785353daa700061bd644411", |
| "637564e470c9acb73295c3ec06eed35945da0657603a3a92be6d32c8f08aec3c", |
| "ac9afa3138675597a6d0e4cdaecb39ba1a3d9cbcabf8cd6db561e6af39967bcd", |
| "d13512a0c70692bb7a479ff545d2d9742d1ed592becbd6bdaae82ffa23dd3c14", |
| "eb7a2adfd2333fbc6ee975fdff2f6737df430e2a66cfb0d0494e4482fd5a9998" |
| ], |
| "run_id": "generalized-cpp-kernel-grpo20-spot-20260829-083214-retry1", |
| "schema_version": 5, |
| "source_manifest_kind": "aider-polyglot-cpp-shadow-rubrics", |
| "source_manifest_sha256": "369f0218313c2240a1073d99b1a9350b2ea54830a7ecc144ca0be2f88289444c", |
| "source_root": "official:Aider-AI/polyglot-benchmark@7e0611e77b54e2dea774cdc0aa00cf9f7ed6144f/cpp/exercises/practice", |
| "source_tree_sha256": "c5609e64371cbbd7f8412f9b250edf267ab996642ad189d76e15655e4ba94192", |
| "split_contract": { |
| "imitation_role": "none", |
| "monitor": "training-task trend monitor only", |
| "official_26": "explicitly authorized training target", |
| "official_task_id_overlap": [ |
| "clock", |
| "complex-numbers", |
| "crypto-square", |
| "grade-school", |
| "kindergarten-garden", |
| "perfect-numbers" |
| ], |
| "reference_answers_packaged": false, |
| "source_reference_answers_packaged": false, |
| "train": "manifest-declared executable shadow tasks", |
| "validation": "training-task monitor fallback; not checkpoint-selection evidence" |
| } |
| } |
|
|
