{"schema_version":"papers/paper-detail-v1","title":"When Mean CE Fails: Median CE Can Better Track Language Model Quality","surface":"papers","opportunity_kernel":{"paper_id":"9cdd187e-8289-4d18-8085-954457245b61","title":"When Mean CE Fails: Median CE Can Better Track Language Model Quality","authors":["Hao Guo","Simon Dennis","Rivaan Patil","Kevin Shabahang"],"arxiv_id":"2605.24667v1","doi":null,"published_at":"2026-05-23T17:10:20.000Z","score_object":{"overall":{"value":0,"scale":"0-10","confidence":0.85,"confidence_reason":"Backfilled from persisted papers.viability_score.","model_version":"phase0-backfill-v1","pipeline_version":"phase0-kernel-v1","computed_at":"2026-05-27T16:35:14.424Z","fresh_until":"2026-06-26T16:35:14.424Z","is_stale":false,"source_count":1,"missingness":[]},"technical":{"value":1.4,"scale":"0-10","confidence":0.55,"confidence_reason":"Backfilled from paper_extraction_scorecards.reconstruction_score.","model_version":"phase0-backfill-v1","pipeline_version":"phase0-kernel-v1","computed_at":"2026-05-27T16:36:20.732Z","fresh_until":"2026-06-10T16:36:20.732Z","is_stale":false,"source_count":1,"missingness":["reproducibility_results.reproducibility_score","deployability_scores.score"]},"commercial":{"value":0,"scale":"0-10","confidence":0.75,"confidence_reason":"Backfilled from persisted commercial_flags and repo availability.","model_version":"phase0-backfill-v1","pipeline_version":"phase0-kernel-v1","computed_at":"2026-05-27T16:35:14.424Z","fresh_until":"2026-06-26T16:35:14.424Z","is_stale":false,"source_count":1,"missingness":[]},"market":{"value":1,"scale":"0-10","confidence":0.45,"confidence_reason":"Heuristic market score materialized from paper metadata, repo availability, and deployability.","model_version":"phase0-backfill-v1","pipeline_version":"phase0-kernel-v1","computed_at":"2026-05-27T16:52:40.065Z","fresh_until":"2026-06-10T16:52:40.065Z","is_stale":false,"source_count":1,"missingness":[]},"team":{"value":3.6,"scale":"0-10","confidence":0.28,"confidence_reason":"Heuristic fallback from paper author count.","model_version":"phase0-backfill-v1","pipeline_version":"phase0-kernel-v1","computed_at":"2026-05-27T16:35:14.424Z","fresh_until":"2026-06-10T16:35:14.424Z","is_stale":false,"source_count":1,"missingness":["paper_author_affiliations"]},"methodology":{"value":3.9,"scale":"0-10","confidence":0.82,"confidence_reason":"Backfilled from paper_extraction_scorecards.total_score.","model_version":"phase0-backfill-v1","pipeline_version":"phase0-kernel-v1","computed_at":"2026-05-27T16:36:20.732Z","fresh_until":"2026-06-26T16:36:20.732Z","is_stale":false,"source_count":2,"missingness":[]}},"evidence_receipt":{"freshness":"fresh","proof_status":"unverified","repo_status":"missing","references_count":0,"source_count":3,"coverage":0.5,"missingness":["repo_url","references","proof_status"],"unresolved_unknowns":["proof verification has not been recorded yet"],"last_verification_at":"2026-05-27T16:52:40.065Z"},"lineage_hash":"e964245dbe832cfd9db7d3681ce55339967f2d2937e496094a2a90c391567d55"},"distribution":null,"replication_evidence":[],"author_dna":[]}