Retrieval and OCR quality alone do not guarantee reliable answers. How can systems allocate evidence across documents and account for document noise?
RAG & trustworthy knowledge services
Document understanding, retrieval, and trustworthy generation, from industrial OCR robustness to budgeted multi-document evidence allocation.
Combining retrieval, document structure, evidence allocation, and generation evaluation. Related papers cover OCR robustness for RAG, budgeted multi-document evidence allocation, and visual-document RAG.
InduOCRBench spans 12 industries, 570 documents, 3,402 pages, and 11 challenge types. BEAR evaluates hierarchical retrieval and multi-document reasoning under a shared generator and evidence budget.
Results & evaluation
InduOCRBench · Industrial-document robustness
Visual-style documents reach 82.9% OCR accuracy but 52.8% RAG answer accuracy, a 30.1-point gap. Table 2 compares the Avg scores of ten methods on general and industrial-document benchmarks.
| OCR method | OmniDocBench Avg ↑ | InduOCRBench Avg ↑ |
|---|---|---|
| PP-StructureV3 | 86.7 | 60.3 |
| MinerU2 | 75.5 | 66.5 |
| Doc2x | 84.6 | 61.6 |
| GPT-4o | 75.0 | 52.0 |
| Qwen3-VL-235B | 89.2 | 70.9 |
| Gemini-2.5 Pro | 88.0 | 74.5 |
| Deepseek-OCR | 87.0 | 61.5 |
| Hunyuan-OCR | 94.1 | 68.1 |
| MinerU2.5 | 90.7 | 72.5 |
| PaddleOCR-VL | 92.9 | 78.2 |
When Good OCR Is Not Enough · Table 2 / Figure 2 ↗

BEAR · Budgeted evidence allocation
DragonBall retrieval comparisons use a Qwen3-32B generator, with evidence contexts aligned to approximately 4K tokens where applicable; LongRefiner uses a different auxiliary model. Values are mean ± standard deviation over three runs. BEAR(docs) achieves 34.63 ± 2.17 EM and 48.02 ± 1.45 F1 on 2Wiki.
| Method | Recall % ↑ | Completeness % ↑ | Hallucination % ↓ | Irrelevance % ↓ |
|---|---|---|---|---|
| BM25 | 29.17 ± 1.06 | 35.13 ± 1.84 | 32.55 ± 2.40 | 32.25 ± 1.95 |
| BGE-M3 | 58.17 ± 1.91 | 59.56 ± 2.45 | 28.81 ± 2.26 | 11.54 ± 1.55 |
| RAPTOR | 31.93 ± 1.66 | 59.78 ± 2.70 | 26.50 ± 2.48 | 13.67 ± 1.29 |
| LongRefiner | 29.24 ± 2.32 | 43.86 ± 1.98 | 34.88 ± 2.25 | 21.26 ± 1.87 |
| HippoRAG2 | 68.80 ± 1.75 | 73.37 ± 2.24 | 19.55 ± 1.84 | 6.93 ± 0.93 |
| BEAR(docs) | 85.55 ± 0.47 | 91.86 ± 1.32 | 5.64 ± 1.02 | 2.40 ± 0.49 |
| BEAR(nodes) | 84.25 ± 0.84 | 88.64 ± 1.52 | 7.60 ± 1.12 | 3.71 ± 0.86 |

