| # HebOCRBench v1.0.0 compact baseline results |
|
|
| This publication contains exact aggregate `metrics.json` files and exact official `modern-score.json` files from completed, blind Modern Hebrew baseline runs. Every source run artifact is bound by SHA-256 and byte size in each `result.json`. |
|
|
| | Result | Model | Status | Evaluated items | Runner/API failures | |
| |---|---|---:|---:|---:| |
| | `surya2` | Surya OCR 2 | `non_conformant` | 34267 | 0 | |
| | `tesseract` | Tesseract | `non_conformant` | 34267 | 0 | |
|
|
| ## Separate extensions and diagnostics |
|
|
| Extension metrics are reported per track only. Synthetic diagnostics are not rankable, and no extension is blended into the Modern headline or a combined score. |
|
|
| | Result | Model | Real extensions | Synthetic diagnostics | Evaluated items | Runner/API failures | |
| |---|---|---:|---:|---:|---:| |
| | `surya2` | Surya OCR 2 | 3 | 2 | 2264 | 0 | |
| | `tesseract` | Tesseract | 3 | 2 | 2264 | 0 | |
|
|
| ## Deliberate compactness boundary |
|
|
| This pack is **not self-contained**. It does not include raw predictions, page-level records, error details, HTML reports, CSV summaries, run manifests, source images, gold annotations, organizer ID maps, or key material. Therefore it cannot by itself recompute a score. Its attestations can verify the hashes of complete artifacts when those artifacts are obtained separately. |
|
|
| No aggregate score combines separate Hebrew families or extension tracks. |
|
|
