# Dataset sources and attribution

`sources.json` records source metadata for each retained subset, separately for capped train/validation and S1MB test. `dataset_metadata.json` records current split counts and upstream dataset declarations. Source URLs, pinned acquisition revisions, declared licenses, authors and unresolved license caveats are retained where recorded. A source-license review was performed on 2026-09-30; see `sources.json` for source records. No blanket relicensing is applied.

Historical source documents are in `_sources/capped-train/` and `_sources/s1mb/`. In particular, `_sources/s1mb/SOURCES.md` and `_sources/s1mb/sources.json` retain detailed S1MB citations. Their relative paths are relative to the original release; their counts are historical, not current v0 membership. Original synthetic generalization reviews are in `generalization/`. Preserve the original synthetic author, version and author-intended label status. Open-Jev rows were acquired through ZefanCai/Open-Jev; customer-control retains its unresolved upstream question-wording license caveat.

Private Hugging Face counterpart: [hotchpotch/bekko-system-one-dataset-v0](https://huggingface.co/datasets/hotchpotch/bekko-system-one-dataset-v0). Published [source metadata](https://huggingface.co/datasets/hotchpotch/bekko-system-one-dataset-v0/blob/7140f9efc9f3b7e9fe8080a6c0d9784246e6cf4d/sources.json) retains the upstream declarations and notices.

Current Laya-source subset names are recorded in `namespace-map.json`; historical source documents retain their original identifiers. `LocalLLaMA/typed-decisions` supplies the four synthetic workflows in `laya__typed_decisions`.

## Authored evaluation subsets

### s1mb-generalization-contextual-choice

Author: GPT-6-Astra. License: not separately declared. Author-intended, self-reviewed; not independently human-validated. See [generalization/s1mb-generalization-contextual-choice/REVIEW.md](generalization/s1mb-generalization-contextual-choice/REVIEW.md).

### s1mb-generalization-contextual-noul

Author: GPT-6-Astra. License: not separately declared. Author-intended, self-reviewed; not independently human-validated. See [generalization/s1mb-generalization-contextual-noul/REVIEW.md](generalization/s1mb-generalization-contextual-noul/REVIEW.md).

### s1mb-generalization-contextual-score

Author: GPT-6-Astra. License: not separately declared. Author-intended, self-reviewed; not independently human-validated. See [generalization/s1mb-generalization-contextual-score/REVIEW.md](generalization/s1mb-generalization-contextual-score/REVIEW.md).

### s1mb-generalization-diverse-choice

Author: GPT-6-Astra. License: not separately declared. Author-intended, self-reviewed; not independently human-validated. See [generalization/s1mb-generalization-diverse-choice/REVIEW.md](generalization/s1mb-generalization-diverse-choice/REVIEW.md).

### s1mb-generalization-diverse-noul

Author: GPT-6-Astra. License: not separately declared. Author-intended, self-reviewed; not independently human-validated. See [generalization/s1mb-generalization-diverse-noul/REVIEW.md](generalization/s1mb-generalization-diverse-noul/REVIEW.md).

### s1mb-generalization-diverse-score

Author: GPT-6-Astra. License: not separately declared. Author-intended, self-reviewed; not independently human-validated. See [generalization/s1mb-generalization-diverse-score/REVIEW.md](generalization/s1mb-generalization-diverse-score/REVIEW.md).