pith. sign in

Recoverable Identifier

arXiv:2604.23811 · detector doi_compliance · incontrovertible · 2026-05-19 22:44:19.665415+00:00

advisory doi_compliance recoverable_identifier

DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.5281/zenodo.19558452.The) was visible in the surrounding text but could not be confirmed against doi.org as printed.

Paper page Integrity report arXiv Try DOI

Evidence text

using algorith- mic family-consistency filters and nuisance-fit stratifica- tion (172 recoverable, 153 usable, 74 poor, 74 catas- trophic). This benchmark is algorithmically curated, not hand-cleaned, making the evaluation harder but more reproducible than sanitized sets used in prior work [4]. The second isRRUFF-325, a deterministic downstream subset from the same curation pipeline that retains only the usable and recoverable nuisance-fit strata. We use this stricter slice for calibration and topological analysis so that the most extreme nuisance regimes do not domi- nate those controlled measurements. C. Decoder Comparison (RRUFF-473) Table IV and Fig. 3 summarize the decoder compari- son. The system has two distinct operating points.Best Top-1:Stage-2b(larger uniform pretraining plus fine- tuning) with the fused decoder (α= 0.50) at 16.70%, benefiting from larger uniform pretraining that increased geometric rigidity in the split path.Best Top-5:Stage- 2a(the earlier fine-tuned checkpoint) with the Bayesian 6 auxiliary head at 52.22%, retaining the softer joint rank- ing that the larger uniform stage partially traded away. Fusion improves Top-1 for both checkpoints, confirming that the split and auxiliary paths provide complementary information. FIG. 3. Decoder comparison on the RRUFF-473 benchmark. TABLE IV. Top-kaccuracy (%) on the full RRUFF-473 benchmark. Model Decoder Top-1 Top-3 Top-5 Stage-2aSplit 14.80 28.96 34.04 Bayesian aux 14.59 36.15 52.22 Fused (α=0.25) 15.64 

Evidence payload

{
  "printed_excerpt": "using algorith- mic family-consistency filters and nuisance-fit stratifica- tion (172 recoverable, 153 usable, 74 poor, 74 catas- trophic). This benchmark is algorithmically curated, not hand-cleaned, making the evaluation harder but more r",
  "reconstructed_doi": "10.5281/zenodo.19558452.The",
  "ref_index": 1,
  "resolved_title": null,
  "verdict_class": "incontrovertible"
}