| # Open Reason v1.3.8 release files
|
|
|
| Generated by `open-reason build --config all --seed 42`.
|
|
|
| | Configuration | Examples | Verified | Human-authored |
|
| | --- | ---: | ---: | ---: |
|
| | coding | 354 | 340 | 0 |
|
| | reasoning | 381 | 381 | 0 |
|
| | science | 368 | 368 | 0 |
|
| | mathematics | 648 | 648 | 0 |
|
| | human | 268 | 240 | 28 |
|
| | education | 337 | 103 | 0 |
|
| | core | 2348 | 2072 | 28 |
|
| | verified | 2072 | 2072 | 0 |
|
| | all | 2348 | 2072 | 28 |
|
|
|
| - `*.jsonl` — human-readable rows
|
| - `*.parquet` — Hugging Face / analytics
|
| - `core.*` — quality tiers S and A
|
| - `verified.*` — independently checked tier S
|
| - `statistics.md` — per-config tables
|
| - `manifest.yaml` — validation, dedup, contamination, and version
|
|
|
| Project license is Apache-2.0. Per-row `provenance.license_spdx` records upstream SPDX.
|
| Parquet and JSONL shards are local build products (gitignored except `data/sample/`).
|
| GitHub stores this catalog; Hugging Face receives shards on publish.
|
|
|
| **Open Reason does not use Reddit as a data source.**
|
|
|
| Do not mix `benchmarks/items.jsonl` into these files.
|
|
|
