{ "description": "Schema for dsfasdfaserwer/Dataset \u2014 verified Python training examples", "fields": { "id": "unique example id (source-dataset + source row id)", "source_dataset / source_url / source_id / problem_source": "provenance of the original problem", "difficulty": "original difficulty label if any", "language": "always 'python' (Python-mandatory rule)", "instruction": "realistic human-style request: goal-oriented, underspecified, never names the algorithm/approach", "chain_of_thought": "solver-voice reasoning: restate goal -> analyze -> plan -> implementation decisions -> self-verification trace (per researched CoT best practices)", "code": "execution-VERIFIED Python solution (ran against the original test suite in a sandboxed subprocess)", "doc": "2-4 sentence clear explanation of what the code does + anything helpful", "tests": "up to 3 real test cases from the source dataset (input/expected_output/mode)", "verification": "how it was verified: method, style (stdin|fn), tests_passed/tests_total, whether a fix pass was applied", "token_counts": "tiktoken cl100k_base counts per field + total (total = instruction+cot+code+doc)", "context_bucket": "'16k' if total <= 16384 tokens else 'above_16k'" }, "splits": { "data/16k_context/": "examples with total <= 16384 tokens", "data/above_16k/": "examples with total > 16384 tokens" } }