{"slug": "diagnosing-fine-grained-inconsistency-classification-in-financial-disclosure", "title": "Diagnosing Fine-Grained Inconsistency Classification in Financial Disclosure Text", "summary": "A new study from arXiv finds that fine-tuned compact encoders match or exceed large language models in classifying fine-grained inconsistencies in financial disclosure text. Using a 5,940-instance snapshot of the SBID-FD benchmark, a fine-tuned 300M encoder achieved 61.9% accuracy, compared with 61.5% for LoRA-adapted Qwen3.5-9B and 61.3% for GPT-5.4. Supplying gold evidence spans improved the encoder to 65.3%, while automatically predicted spans recovered only part of that gain, highlighting evidence localization as a key bottleneck.", "body_md": "arXiv:2607.26368v1 Announce Type: new\nAbstract: Financial disclosures contain numerical claims, temporal statements, entity references, policy commitments, and risk descriptions that may conflict in qualitatively different ways. Detecting a conflict is only the first step: review workflows may also need to determine its type, since numerical, temporal, referential, factual, and normative inconsistencies require different evidence and downstream checks. We study this problem as fine-grained inconsistency classification. Using a fixed 5,940-instance snapshot of SBID-FD, a synthetic financial-disclosure benchmark with 11 inconsistency labels and paired reference evidence spans, we compare frozen embedding classifiers, fine-tuned encoders, evidence-augmented classifiers, prompted large language models, and LoRA-adapted generative models under a shared evaluation protocol. A fine-tuned 300M encoder reaches 61.9% accuracy, compared with 61.5% for a LoRA-adapted Qwen3.5-9B model and 61.3% for GPT-5.4. Because these systems differ in architecture, supervision, training objective, and input format, we interpret this as a practical efficiency result for compact supervised encoders rather than a controlled conclusion about model scale. Supplying gold evidence spans improves the fine-tuned encoder to 65.3%, whereas automatically predicted spans recover a meaningful but incomplete share of that gain, indicating that localization quality remains a bottleneck. Class-level analyses show that Referential inconsistencies are especially sensitive to localization quality, while Factual and Logical inconsistencies remain difficult even when the relevant evidence is provided. Together, the oracle, distractor, and per-class analyses separate localization errors from residual type-discrimination errors, indicating that progress requires both stronger evidence extraction and better reasoning over closely related inconsistency categories.", "url": "https://wpnews.pro/news/diagnosing-fine-grained-inconsistency-classification-in-financial-disclosure", "canonical_source": "https://www.machinebrief.com/news/diagnosing-fine-grained-inconsistency-classification-in-fina-njec", "published_at": "2026-07-30 04:00:00+00:00", "updated_at": "2026-07-30 04:32:30.613437+00:00", "lang": "en", "topics": ["natural-language-processing", "artificial-intelligence", "machine-learning", "large-language-models"], "entities": ["arXiv", "SBID-FD", "Qwen3.5-9B", "GPT-5.4"], "alternates": {"html": "https://wpnews.pro/news/diagnosing-fine-grained-inconsistency-classification-in-financial-disclosure", "markdown": "https://wpnews.pro/news/diagnosing-fine-grained-inconsistency-classification-in-financial-disclosure.md", "text": "https://wpnews.pro/news/diagnosing-fine-grained-inconsistency-classification-in-financial-disclosure.txt", "jsonld": "https://wpnews.pro/news/diagnosing-fine-grained-inconsistency-classification-in-financial-disclosure.jsonld"}}