Download outputs/scope_audit.json from SabaPivot/repro-differentiable-conformal-training-for-llm-reasoning-factuality: direct link, hf CLI and curl.
- Browser
- Download file 913 Bytes
-
https://huggingface.co/spaces/SabaPivot/repro-differentiable-conformal-training-for-llm-reasoning-factuality/resolve/main/outputs/scope_audit.json
- Command line
-
hf download hf://spaces/SabaPivot/repro-differentiable-conformal-training-for-llm-reasoning-factuality/outputs/scope_audit.json
-
curl -L -o scope_audit.json https://huggingface.co/spaces/SabaPivot/repro-differentiable-conformal-training-for-llm-reasoning-factuality/resolve/main/outputs/scope_audit.json
913 Bytes
| { | |
| "arxiv_id": "2604.20098v1", | |
| "claim_1": "exact source table arithmetic; 141.1% retention improvement, but DCF misses the 97% target by 0.45 percentage points", | |
| "claim_2": "exact source table arithmetic; 61.3% retention improvement and 99.15% coverage exceeds the 99% target", | |
| "claim_3": "Theorem 3.1 itself proves soft nonconformity-score convergence. Soft-quantile recovery is stated separately in Section 3.3 and empirically validated in Section 4.2.", | |
| "claim_4": "finite graph follows the Appendix-A.2 coupled temperature schedule and recovers the hard prediction set", | |
| "claim_5": "all eight Table-2 rows independently recomputed from confusion counts", | |
| "claim_6": "exact source cascade and an ancestor-removal destructive control", | |
| "large_scale_scope": "No LLM scoring, MATH/FELM model training, cross-validation, or dataset evaluation is performed or claimed.", | |
| "paper_id": "XfndtVLIub" | |
| } | |