{"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=Qwen2.5&bench=MBPP","json_url":"https://assignee.net/benchmarks/evidence.json?model=Qwen2.5&bench=MBPP","model":"Qwen2.5","benchmark":"MBPP","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":2026,"basis":"public source URLs, source titles, and reported publication years in this evidence cluster"},"reported_range":{"min_score_pct":38.6,"max_score_pct":85.2},"spread_pp":46.6,"severity":"HIGH","entries":[{"model":"Qwen2.5","benchmark":"MBPP","score_pct":85.2,"source_title":"ReflexiCoder: Teaching Large Language Models to Self-Reflect on Generated Code and Self-Correct It via Reinforcement Learning","source_url":"https://arxiv.org/abs/2603.05863","source_domain":"arxiv.org","year":2026},{"model":"Qwen2.5","benchmark":"MBPP","score_pct":69.2,"source_title":"Qwen2.5 Technical Report","source_url":"http://arxiv.org/abs/2412.15115v2","source_domain":"arxiv.org","year":2024},{"model":"Qwen2.5","benchmark":"MBPP","score_pct":38.6,"source_title":"HumanEval Pro and MBPP Pro: Evaluating Large Language Models on Self-invoking Code Generation","source_url":"http://arxiv.org/abs/2412.21199v2","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."]}