{"schema":"https://assignee.net/schemas/benchmark-evidence-v1","schema_version":"1.0","contract_version":"benchmark-evidence-v1.0","contract_updated":"2026-06-01","schema_documentation":"https://assignee.net/schemas","changelog_url":"https://assignee.net/changelog","publisher":{"name":"Assignee Research","url":"https://assignee.net"},"html_url":"https://assignee.net/benchmarks/evidence?model=Llama-3.1-70B&bench=MATH","json_url":"https://assignee.net/benchmarks/evidence.json?model=Llama-3.1-70B&bench=MATH","model":"Llama-3.1-70B","benchmark":"MATH","source_count":3,"source_coverage":{"record_count":3,"distinct_source_count":3,"coverage_level":"MODERATE","basis":"distinct public paper URLs or titles in this evidence cluster"},"source_profile":{"source_url_count":3,"missing_source_url_count":0,"domains":["arxiv.org"],"year_min":2024,"year_max":2025,"basis":"public source URLs, source titles, and reported publication years in this evidence cluster"},"reported_range":{"min_score_pct":42.8,"max_score_pct":67.0},"spread_pp":24.2,"severity":"HIGH","entries":[{"model":"Llama-3.1-70B","benchmark":"MATH","score_pct":67.0,"source_title":"\"Give Me BF16 or Give Me Death\"? Accuracy-Performance Trade-Offs in LLM Quantization","source_url":"http://arxiv.org/abs/2411.02355v4","source_domain":"arxiv.org","year":2024},{"model":"Llama-3.1-70B","benchmark":"MATH","score_pct":60.0,"source_title":"Safe: Enhancing Mathematical Reasoning in Large Language Models via Retrospective Step-aware Formal Verification","source_url":"http://arxiv.org/abs/2506.04592v1","source_domain":"arxiv.org","year":2025},{"model":"Llama-3.1-70B","benchmark":"MATH","score_pct":42.8,"source_title":"Enhancing Reasoning Capabilities of LLMs via Principled Synthetic Logic Corpus","source_url":"http://arxiv.org/abs/2411.12498v2","source_domain":"arxiv.org","year":2024}],"interpretation":"This record groups score claims extracted from papers for the same model and benchmark label. A nonzero spread means the public literature reports different values for this cluster.","limitations":["Differences are not automatically errors.","Reported values may differ because of prompts, dataset versions, evaluation protocols, scoring rules, preprocessing, fine-tuning, or reporting conventions.","Source papers remain authoritative for their own claims."]}