| |
| """ |
| Generate the complete UnsolvedMed (ulamai/UnsolvedMedical) dataset across v1.0.0 (10k), v1.5.0 (15k), and v2.0.0 (20k): |
| - v1.0.0: 10,000 published open/partially-resolved medical problems + 1,000 Gold Seed + 100 calibration = 10,100 records |
| - v1.5.0: 15,000 published open/partially-resolved medical problems + 1,500 Gold Seed + 100 calibration = 15,100 records |
| - v2.0.0: 20,000 published open/partially-resolved medical problems + 2,000 Gold Seed + 100 calibration = 20,100 records |
| - Every published problem is backed by 3 distinct, sub-domain-matched peer-reviewed medical literature papers |
| (`problem_origin`, `openness_support`, and `partial_result`) PLUS 1 sub-domain-matched Tier A/B/C/D |
| institutional consensus/guideline/priority status anchor (`status_review`) = 4 distinct sources per problem. |
| - Zero raw documents or verbatim abstract excerpts are redistributed; all fields are synthesized in canonical house format. |
| - Spans 20 primary medical categories, 400 clinical sub-domains (20 per category), and 20 distinct question-type templates. |
| """ |
|
|
| import argparse |
| import hashlib |
| import html |
| import json |
| import re |
| from collections import Counter, defaultdict |
| from pathlib import Path |
|
|
| import pyarrow as pa |
| import pyarrow.parquet as pq |
|
|
| from clinical_pico_parser import ( |
| GENERIC_CONDITION_MODIFIERS, |
| build_natural_statements, |
| clean_complete_topic, |
| extract_paper_pico, |
| ) |
| from curated_gold_and_anchors import ( |
| CATEGORY_PROFILES, |
| build_category_status_anchors, |
| get_archived_and_audit_records, |
| ) |
|
|
| ROOT = Path(__file__).resolve().parent.parent |
| TMP_HARVEST = Path("/tmp/harvested_medical_candidates.jsonl") |
| HARVEST_FILE = TMP_HARVEST if TMP_HARVEST.exists() else (ROOT / "raw" / "harvested_medical_candidates.jsonl") |
| NCT_TRIALS_FILE = Path("/tmp/harvested_clinical_trials.json") |
| RELEASE_DATE = "2026-10-16" |
| RELEASE_TIMESTAMP = "2026-10-16T12:00:00Z" |
|
|
| |
| BASE_CATEGORY_QUOTAS = { |
| 1: 800, |
| 2: 600, |
| 3: 600, |
| 4: 600, |
| 5: 800, |
| 6: 400, |
| 7: 500, |
| 8: 400, |
| 9: 360, |
| 10: 440, |
| 11: 400, |
| 12: 500, |
| 13: 500, |
| 14: 400, |
| 15: 500, |
| 16: 500, |
| 17: 500, |
| 18: 400, |
| 19: 400, |
| 20: 400, |
| } |
|
|
| |
| QUESTION_TYPES = [ |
| "etiology", |
| "mechanism", |
| "natural_history", |
| "prevention", |
| "screening", |
| "diagnosis", |
| "classification", |
| "prognosis", |
| "treatment_effectiveness", |
| "comparative_effectiveness", |
| "treatment_strategy", |
| "safety", |
| "rehabilitation", |
| "supportive_palliative", |
| "patient_experience", |
| "implementation", |
| "health_services", |
| "public_health", |
| "measurement", |
| "methodology", |
| ] |
|
|
| QTYPE_KEYWORDS = { |
| "etiology": ["etiology", "risk factor", "caus", "pathogenesis", "environmental exposure", "predisposition", "origin"], |
| "mechanism": ["mechanism", "pathway", "signaling", "microenvironment", "resistance", "immunopathogenesis", "molecular", "cellular"], |
| "natural_history": ["natural history", "longitudinal", "trajectory", "disease course", "progression", "cohort"], |
| "prevention": ["prevention", "preventive", "vaccine", "vaccination", "prophylaxis", "prophylactic", "chemoprevention"], |
| "screening": ["screening", "early detection", "surveillance", "mammography", "colonoscopy", "asymptomatic"], |
| "diagnosis": ["diagnostic", "diagnosis", "accuracy", "sensitivity", "specificity", "imaging", "biopsy", "assay", "detection"], |
| "classification": ["subtype", "endotype", "phenotype", "classification", "staging", "stratification", "taxonomy", "cluster"], |
| "prognosis": ["prognos", "predict", "biomarker", "risk score", "nomogram", "outcome prediction", "survival"], |
| "treatment_effectiveness": ["efficacy", "effectiveness", "phase 2", "phase 3", "randomized", "remission", "monotherapy"], |
| "comparative_effectiveness": ["versus", "vs.", "compared with", "comparative", "head-to-head", "non-inferiority", "superiority"], |
| "treatment_strategy": ["sequencing", "de-escalation", "discontinuation", "duration", "threshold", "dosing", "regimen", "switch", "maintenance"], |
| "safety": ["safety", "adverse", "toxicity", "bleeding", "complication", "harm", "pharmacovigilance", "side effect", "tolerability"], |
| "rehabilitation": ["rehabilitation", "physical therapy", "physiotherapy", "exercise", "mobility", "motor recovery", "prosthe"], |
| "supportive_palliative": ["palliative", "supportive", "symptom", "end-of-life", "hospice", "analgesia", "fatigue", "cachexia"], |
| "patient_experience": ["quality of life", "patient-reported", "satisfaction", "preference", "shared decision", "lived experience", "burden"], |
| "implementation": ["implementation", "adherence", "adoption", "fidelity", "de-implementation", "clinical pathway", "stewardship", "guideline"], |
| "health_services": ["health service", "health system", "readmission", "staffing", "cost", "resource", "telehealth", "care delivery", "triage"], |
| "public_health": ["public health", "population", "epidemiol", "disparity", "equity", "outbreak", "global health", "community"], |
| "measurement": ["measurement", "scale", "instrument", "psychometric", "core outcome", "surrogate endpoint", "quantification", "calibration"], |
| "methodology": ["methodology", "trial design", "meta-analysis", "confounding", "heterogeneity", "bias", "estimand", "external validation"], |
| } |
|
|
| QTYPE_METADATA = { |
| "etiology": { |
| "framework": "PECO", |
| "kind": "exposure", |
| "stage": "fundamental_human_mechanism", |
| "study_modes": ["prospective_cohort", "case_control_or_case_cohort", "multi_method_triangulation"], |
| "gaps": ["causal_ambiguity", "mechanistic_ambiguity", "inconsistency"], |
| "evidence_state": "observational_only", |
| }, |
| "mechanism": { |
| "framework": "mechanistic", |
| "kind": "pathway_or_mechanism", |
| "stage": "fundamental_human_mechanism", |
| "study_modes": ["mechanistic_human_study", "multi_method_triangulation", "prospective_cohort"], |
| "gaps": ["mechanistic_ambiguity", "conflicting_interpretations", "indirectness"], |
| "evidence_state": "mechanistic_evidence_only", |
| }, |
| "natural_history": { |
| "framework": "natural_history", |
| "kind": "disease_trajectory", |
| "stage": "early_clinical", |
| "study_modes": ["prospective_cohort", "registry_or_surveillance", "individual_participant_data_meta_analysis"], |
| "gaps": ["short_follow_up", "underrepresented_population", "imprecision"], |
| "evidence_state": "observational_only", |
| }, |
| "prevention": { |
| "framework": "PICOTS", |
| "kind": "preventive_intervention", |
| "stage": "population_health", |
| "study_modes": ["randomized_trial", "pragmatic_trial", "prospective_cohort"], |
| "gaps": ["short_follow_up", "insufficient_sample_size", "uncertain_generalizability"], |
| "evidence_state": "multiple_small_studies", |
| }, |
| "screening": { |
| "framework": "PIRD", |
| "kind": "screening_strategy", |
| "stage": "population_health", |
| "study_modes": ["randomized_trial", "clinical_utility_study", "diagnostic_accuracy_study"], |
| "gaps": ["missing_patient_important_outcomes", "rare_or_delayed_harm", "indirectness"], |
| "evidence_state": "observational_only", |
| }, |
| "diagnosis": { |
| "framework": "PIRD", |
| "kind": "index_test", |
| "stage": "biomarker_discovery", |
| "study_modes": ["diagnostic_accuracy_study", "clinical_utility_study", "prospective_cohort"], |
| "gaps": ["uncertain_reference_standard", "model_not_externally_validated", "imprecision"], |
| "evidence_state": "multiple_small_studies", |
| }, |
| "classification": { |
| "framework": "prognostic", |
| "kind": "endotype_or_subtype_classifier", |
| "stage": "biomarker_discovery", |
| "study_modes": ["consensus_definition_with_validation", "prospective_cohort", "individual_participant_data_meta_analysis"], |
| "gaps": ["conflicting_interpretations", "measurement_invalidity", "model_not_externally_validated"], |
| "evidence_state": "multiple_small_studies", |
| }, |
| "prognosis": { |
| "framework": "prognostic", |
| "kind": "predictor_or_model", |
| "stage": "early_clinical", |
| "study_modes": ["prognostic_validation", "prospective_cohort", "individual_participant_data_meta_analysis"], |
| "gaps": ["model_not_externally_validated", "short_follow_up", "underrepresented_population"], |
| "evidence_state": "observational_only", |
| }, |
| "treatment_effectiveness": { |
| "framework": "PICOTS", |
| "kind": "intervention", |
| "stage": "confirmatory_clinical", |
| "study_modes": ["randomized_trial", "adaptive_or_platform_trial", "systematic_review_or_meta_analysis"], |
| "gaps": ["insufficient_sample_size", "imprecision", "missing_patient_important_outcomes"], |
| "evidence_state": "early_interventional", |
| }, |
| "comparative_effectiveness": { |
| "framework": "PICOTS", |
| "kind": "intervention", |
| "stage": "comparative_effectiveness", |
| "study_modes": ["randomized_trial", "pragmatic_trial", "individual_participant_data_meta_analysis"], |
| "gaps": ["no_direct_evidence", "inconsistency", "rapidly_changing_standard_of_care"], |
| "evidence_state": "systematic_review_low_certainty", |
| }, |
| "treatment_strategy": { |
| "framework": "PICOTS", |
| "kind": "treatment_sequencing_or_dosing_strategy", |
| "stage": "comparative_effectiveness", |
| "study_modes": ["pragmatic_trial", "adaptive_or_platform_trial", "randomized_trial"], |
| "gaps": ["no_direct_evidence", "short_follow_up", "rapidly_changing_standard_of_care"], |
| "evidence_state": "multiple_small_studies", |
| }, |
| "safety": { |
| "framework": "PICOTS", |
| "kind": "intervention_or_exposure", |
| "stage": "post_authorization", |
| "study_modes": ["long_term_post_authorization_follow_up", "registry_or_surveillance", "prospective_cohort"], |
| "gaps": ["rare_or_delayed_harm", "short_follow_up", "insufficient_sample_size"], |
| "evidence_state": "one_pivotal_study", |
| }, |
| "rehabilitation": { |
| "framework": "PICOTS", |
| "kind": "rehabilitation_intervention", |
| "stage": "comparative_effectiveness", |
| "study_modes": ["randomized_trial", "pragmatic_trial", "clinical_utility_study"], |
| "gaps": ["high_risk_of_bias", "imprecision", "missing_patient_important_outcomes"], |
| "evidence_state": "multiple_small_studies", |
| }, |
| "supportive_palliative": { |
| "framework": "PICOTS", |
| "kind": "supportive_or_palliative_strategy", |
| "stage": "comparative_effectiveness", |
| "study_modes": ["pragmatic_trial", "randomized_trial", "qualitative_or_mixed_methods"], |
| "gaps": ["missing_patient_important_outcomes", "high_risk_of_bias", "underrepresented_population"], |
| "evidence_state": "systematic_review_low_certainty", |
| }, |
| "patient_experience": { |
| "framework": "qualitative", |
| "kind": "care_delivery_or_decision_support", |
| "stage": "implementation", |
| "study_modes": ["qualitative_or_mixed_methods", "pragmatic_trial", "consensus_definition_with_validation"], |
| "gaps": ["missing_patient_important_outcomes", "underrepresented_population", "measurement_invalidity"], |
| "evidence_state": "observational_only", |
| }, |
| "implementation": { |
| "framework": "implementation", |
| "kind": "implementation_strategy", |
| "stage": "implementation", |
| "study_modes": ["pragmatic_trial", "quasi_experimental_study", "qualitative_or_mixed_methods"], |
| "gaps": ["implementation_context_uncertainty", "uncertain_generalizability", "inconsistency"], |
| "evidence_state": "multiple_small_studies", |
| }, |
| "health_services": { |
| "framework": "health_systems", |
| "kind": "service_delivery_model", |
| "stage": "health_system", |
| "study_modes": ["pragmatic_trial", "quasi_experimental_study", "natural_experiment"], |
| "gaps": ["implementation_context_uncertainty", "uncertain_generalizability", "high_risk_of_bias"], |
| "evidence_state": "observational_only", |
| }, |
| "public_health": { |
| "framework": "health_systems", |
| "kind": "population_health_intervention", |
| "stage": "population_health", |
| "study_modes": ["natural_experiment", "quasi_experimental_study", "pragmatic_trial"], |
| "gaps": ["causal_ambiguity", "uncertain_generalizability", "underrepresented_population"], |
| "evidence_state": "observational_only", |
| }, |
| "measurement": { |
| "framework": "PIRD", |
| "kind": "outcome_instrument_or_biomarker", |
| "stage": "biomarker_discovery", |
| "study_modes": ["consensus_definition_with_validation", "diagnostic_accuracy_study", "prospective_cohort"], |
| "gaps": ["measurement_invalidity", "uncertain_reference_standard", "model_not_externally_validated"], |
| "evidence_state": "multiple_small_studies", |
| }, |
| "methodology": { |
| "framework": "PICOTS", |
| "kind": "trial_or_causal_inference_design", |
| "stage": "confirmatory_clinical", |
| "study_modes": [ |
| "consensus_definition_with_validation", |
| "individual_participant_data_meta_analysis", |
| "adaptive_or_platform_trial", |
| ], |
| "gaps": ["measurement_invalidity", "conflicting_interpretations", "high_risk_of_bias"], |
| "evidence_state": "systematic_review_low_certainty", |
| }, |
| } |
|
|
| DIFFICULTY_MAP = { |
| 1: "bounded", |
| 2: "moderate", |
| 3: "hard", |
| 4: "very_hard", |
| 5: "frontier", |
| } |
|
|
| PUB_CLASSES_CURRENT = [ |
| ("OPEN-VERIFIED", "open", "broad_consensus_open", "high"), |
| ("OPEN-VERIFIED", "open", "specialist_consensus_open", "high"), |
| ("PARTIAL-PROGRESS", "partially_resolved", "specialist_consensus_open", "high"), |
| ("DISPUTED-OPEN", "contested", "active_scientific_dispute", "moderate"), |
| ("BLOCKED-OPEN", "open", "broad_consensus_open", "high"), |
| ] |
|
|
| |
| PRESERVED_ACRONYMS = [ |
| "DOAC", "DOACs", "VTE", "DVT", "PE", "NSCLC", "SCLC", "HR+", "HER2-", "HER2+", "HER2", "BCMA", "CAR-T", "CAR", |
| "MRD", "AML", "CR1", "FLT3", "ITD", "NPM1", "IDH", "IDH1", "IDH2", "MGMT", "FGFR3", "MIBC", "DLL3", |
| "FOLFIRINOX", "NALIRIFOX", "MPN", "JAK2", "CALR", "MPL", "HCT", "GVHD", "PTC", "ATC", "BRAF", "MELD", |
| "TACE", "TARE", "GIST", "KIT", "PDGFRA", "SDH", "NET", "NETs", "PRRT", "pMMR", "dMMR", "MSI-H", "MSS", |
| "MRI", "CT", "PET", "PSMA-PET", "PSMA", "ADT", "ISUP", "PSA", "DLBCL", "MYC", "BCL2", "R-CHOP", "CD19", |
| "CD20", "CD38", "HSCT", "BMT", "HLA", "HbSS", "HbS", "MDS", "SF3B1", "Hb", "GI", "HFpEF", "HFrEF", "LVEF", |
| "NYHA", "AF", "AHRE", "ECG", "ICD", "CHA2DS2-VASc", "HAS-BLED", "MeVO", "DMVO", "MCA", "ACA", "PCA", |
| "NIHSS", "INOCA", "ANOCA", "TAVR", "SAVR", "CMR", "LGE", "NT-proBNP", "ICH", "LAA", "DCM", "TTN", |
| "LMNA", "GDMT", "ABPM", "BP", "CKD", "SGLT2", "GLP-1", "GIP", "MRA", "ARNI", "ACEi", "ARB", "CTPA", |
| "RV", "LV", "tPA", "PDE5", "WMH", "APOE", "HCM", "LVOT", "SCAD", "FMD", "PAD", "CLTI", "PFO", |
| "RoPE", "LVAD", "HM3", "CTEPH", "BPA", "PEA", "VHD", "PFA", "RF", "RRMS", "NEDA-3", "mAb", "ASM", |
| "EEG", "fMRI", "RNS", "DBS", "ANT", "VNS", "REM", "SAA", "GBA1", "LRRK2", "MDS-UPDRS", "ALS", "TDP-43", |
| "NfL", "C9orf72", "SOD1", "FUS", "ALSFRS-R", "CGRP", "ICHD-3", "CBT", "NMDAR", "LGI1", "CASPR2", "IVIG", |
| "MOGAD", "MOG-IgG", "ADEM", "AChR", "MuSK", "MGFA", "MG-ADL", "MG", "FcRn", "C5", "TBI", "ICU", |
| "GCS", "CRS-R", "iNPH", "NPH", "CSF", "DESH", "DTI-ALPS", "EDSS", "CIDP", "NF155", "CNT1", "FTD", "MAPT", |
| "GRN", "HTT", "IIH", "CVT", "GBS", "NMOSD", "AQP4", "FND", "MSA", "PSP", "MDD", "DSM-5", "ICD-11", |
| "SSRI", "SNRI", "SERT", "PHQ-9", "rTMS", "iTBS", "ECT", "OUD", "MOUD", "PTSD", "CAPS-5", "EMDR", |
| "BMI", "CBT-E", "DBT-A", "LGBTQ+", "ADHD", "MADRS", "PANSS", "C-SSRS", "YMRS", "DESS", "MACE", "WHODAS", |
| "OCD", "YBOCS", "AUD", "SUD", "ASD", "BZD", "GAD", "CBT-I", "CRE", "NDM", "VIM", "IMP", "KPC", "LMIC", |
| "RR-TB", "MDR-TB", "XDR-TB", "TB", "LTBI", "HIV-1", "HIV", "SAB", "MRSA", "MSSA", "SARS-CoV-2", "COVID-19", |
| "RNA", "DNA", "cccDNA", "HBsAg", "HBeAg", "HBV", "HDV", "HCV", "ALT", "NS1", "RT-PCR", "EORTC", |
| "MSGERC", "CNS", "NTM", "MAC", "MABC", "CMV", "EBV", "BKV", "NTD", "STI", "Doxy-PEP", "OPAT", "PJI", |
| "IE", "HAP", "VAP", "COPD", "GOLD", "ICS", "LABA", "LAMA", "FEV1", "IPF", "ILD", "FVC", "DLCO", "HRCT", |
| "OSA", "CPAP", "AHI", "CFTR", "CF", "ARDS", "ECMO", "VV-ECMO", "VA-ECMO", "ECPR", "PEEP", "PaO2", "FiO2", |
| "CHP", "NMD", "NIV", "ASV", "CSB", "SDB", "LAM", "PAP", "GM-CSF", "PH-ILD", "MPM", "IPC", "EGPA", "ABPA", |
| "T1D", "T2D", "HbA1c", "CGM", "AID", "MASLD", "MASH", "FIB-4", "VCTE", "ELF", "DEXA", "PTH", "TSH", |
| "T4", "T3", "MODY", "RED-S", "FHH", "PHPT", "CAH", "21-OHD", "AVP", "SIADH", "CGL", "FPLD", "IBD", "CD", |
| "UC", "TDM", "UST", "VDZ", "ERCP", "EUS", "WON", "TIPS", "HVPG", "HE", "FMT", "CDI", "GERD", "PPI", |
| "POEM", "LHM", "EGD", "LGD", "HGD", "EMR", "ESD", "PERT", "SIBO", "IMO", "ALF", "NAC", "PBC", "AIH", |
| "OCA", "ESKD", "eGFR", "UACR", "IgA", "IgAN", "C3", "C3G", "IC-MPGN", "HUS", "aHUS", "ANCA", "AAV", |
| "ADPKD", "TKV", "CRRT", "AKI", "RRT", "KRT", "BPH", "HoLEP", "RTA", "NDI", "RVD", "ARAS", "IC/BPS", |
| "LUTS", "IPSS", "SLE", "LN", "RA", "PsA", "axSpA", "nr-axSpA", "bDMARD", "bDMARDs", "tsDMARD", |
| "csDMARD", "JAK", "TNF", "IL-6", "IL-17", "IL-23", "IL-4", "IL-13", "IL-5", "IgE", "OIT", "SLIT", "SCIT", |
| "IIM", "ASSD", "MDA5", "CAPS", "FMF", "TRAPS", "VEXAS", "UBA1", "APS", "CSU", "CRSwNP", "AERD", "HLH", |
| "MAS", "MCTD", "UCTD", "OA", "ACL", "LBP", "AD", "PSO", "PASI", "EASI", "HS", "IHS4", "DDD", "CRPS", |
| "PV", "MMP", "GPP", "PPP", "cSCC", "BCC", "AIS", "CTCL", "PRP", "PCOS", "AMH", "IVF", "ICSI", "PGT-A", |
| "SET", "PPH", "TXA", "GDM", "OGTT", "HRT", "MHT", "VMS", "RPL", "RIF", "HDP", "FGR", "sFlt-1", "PlGF", |
| "PAS", "PPROM", "PTB", "GSM", "SUI", "POP", "ICP", "UDCA", "BPD", "NICU", "LISA", "INSURE", "PDA", "NEC", |
| "ROP", "VEGF", "HIE", "RSV", "JIA", "IVH", "CDH", "MIS-C", "TOF", "HLHS", "SCD", "TCD", "LGS", "GHD", |
| "ISS", "SGA", "MCI", "BPSD", "MMSE", "MoCA", "CFS", "CGA", "DLB", "VaD", "CAA", "LBD", "STOPP", "START", |
| "CPR", "OHCA", "ROSC", "TTM", "MAP", "REBOA", "FAST", "ERAS", "PONV", "TIVA", "BIS", "PbtO2", "CRBSI", |
| "MODS", "PICS", "MTP", "TEG", "ROTEM", "LAST", "NGS", "WES", "WGS", "ctDNA", "cfDNA", "LP-WGS", "dPCR", |
| "mNGS", "POCT", "MCED", "AI", "LLM", "CAD", "CCTA", "FFR-CT", "AUROC", "PR-AUC", "PRS", "PGx", "CPIC", |
| "HPV", "LDCT", "USPSTF", "CRC", "FIT", "CVD", "ASCVD", "CAC", "Lp(a)", "apoB", "SBP", "PrEP", "CAB-LA", |
| "TDF", "FTC", "CHW", "SDOH", "SNAP", "UPF", "SSB", "PM2.5", "EHR", "CDS", "ADE", "FHIR", "SNF", "RPM", |
| "HaH", "PRO", "PROs", "PROM", "PREM", "COS", "COMET", "OMERACT", "ICHOM", "CI", "VCI", "PTA", "Tinnitus", |
| "SCI", "FES", "BCI", "SDR", "ITB", "CIMT", "AMD", "GA", "DR", "DME", "IOP", "MIGS", "SLT", "RP", "USH", |
| "ABCA4", "RPE65", "SSD", "UHL", "CL/P", "VPI", "ONJ", "MRONJ", "ORN", "AAC", "TMJ", "TMD", "SMA", "SMN1", |
| "SMN2", "DMD", "ATTR", "ATTR-CM", "TTR", "LSD", "ERT", "SRT", "PKU", "UCD", "HAE", "C1-INH", "VUS", |
| "RNA-seq", "iPSC", "ASO", "CRISPR", "Cas9", "FOP", "ACVR1", "OI", "XLH", "FGF23", "HPP", "TNSALP", "TSC", |
| "NF1", "NF2", "SWN", "VHL", "PWS", "MECP2", "FXS", "FMR1", "CDG", "NPC", "NBS", "ALD", "MLD", "MPS", |
| "AATD", "SERPINA1", "PCD", "JLA", "NICE", "AHRQ", "NASEM", "NIH", "PCORI", |
| "Black", "Indigenous", "Asian", "Hispanic", "Latino", "European", "African", "Type 1", "Type 2", |
| "Type A", "Type B", "Stanford", "Rotterdam", "Bayley-4", "Fugl-Meyer", "North Star", "St George", "Zarit", "Epworth", |
| "Stage I", "Stage II", "Stage III", "Stage IV", "Stage II-III", "Stage IB-IIIA", "Class I", |
| "Class II", "Class III", "Class II-III", "Class II-IV", "Grade", "Group", "Phase I", "Phase II", |
| "Phase III", "Gram-negative", "Gram-positive", "Staphylococcus", "Enterobacterales", "Pseudomonas", |
| "Acinetobacter", "Klebsiella", "Clostridioides", "Candida", "Aspergillus", "Trypanosoma", "Leishmania", |
| "Plasmodium", "Helicobacter", "Alzheimer", "Parkinson", "Huntington", "Crohn", "Graves", "Hashimoto", |
| "Sjögren", "Behçet", "Kawasaki", "Marfan", "Ehlers-Danlos", "Fabry", "Gaucher", "Pompe", "Wilson", |
| "Duchenne", "Charcot-Marie-Tooth", "Guillain-Barré", "Kaposi", "Hodgkin", "Burkitt", "Wilms", |
| ] |
| ACRONYM_MAP = {a.strip().lower(): a.strip() for a in PRESERVED_ACRONYMS if a.strip()} |
| ACRONYM_REGEX = re.compile( |
| r"(?<![a-z0-9])(" |
| + "|".join(re.escape(k) for k in sorted(ACRONYM_MAP.keys(), key=len, reverse=True)) |
| + r")(?![a-z0-9])" |
| ) |
|
|
| |
| NOISE_RE = re.compile( |
| r"\b(case report|a case of|two cases of|three cases of|case series|study protocol|protocol for|" |
| r"trial protocol|corrigendum|erratum|retracted|retraction|comment on|reply to|author response|" |
| r"letter to|editorial|book review|conference abstract|in mice|mouse model|murine model|" |
| r"in rats|rat model|zebrafish|drosophila|in vitro|cell line)\b", |
| re.I, |
| ) |
|
|
| SUBTITLE_BOILERPLATE_RE = re.compile( |
| r"\s*[:–—-]\s*(a |an |the )?(systematic review|meta-analysis|narrative review|scoping review|" |
| r"literature review|umbrella review|comprehensive review|integrative review|minireview|mini-review|" |
| r"review of|updated review|critical review|overview of|retrospective|prospective|multicenter|" |
| r"multi-center|single-center|single center|randomized|randomised|phase [i1234v]+|pilot study|" |
| r"cohort study|cross-sectional|observational study|clinical trial|real-world|secondary analysis|" |
| r"post-hoc|post hoc|sub-study|substudy|preliminary|comparative study|mendelian randomization|" |
| r"network meta-analysis|pooled analysis|propensity score).*$", |
| re.I, |
| ) |
|
|
| PREFIX_BOILERPLATE_RE = re.compile( |
| r"^(a |an )?(systematic review and meta-analysis of|systematic review of|meta-analysis of|" |
| r"narrative review of|scoping review of|comprehensive review of|clinical overview of|overview of|" |
| r"update on|updates in|current perspectives on|emerging role of|advances in|new insights into)\s+", |
| re.I, |
| ) |
|
|
|
|
| BIOMED_TOKEN_RE = re.compile(r"[A-Za-z0-9+/_\-]+") |
|
|
|
|
| def _preserve_biomed_token(m: re.Match) -> str: |
| tok = m.group(0) |
| low = tok.lower() |
| if low in ACRONYM_MAP: |
| return ACRONYM_MAP[low] |
| |
| if any(c.isupper() for c in tok) and any(c.isdigit() for c in tok): |
| return tok |
| |
| core = tok[:-1] if (len(tok) >= 3 and tok[-1] in ("s", "i")) else tok |
| alpha = [c for c in core if c.isalpha()] |
| if 2 <= len(alpha) <= 9 and all(c.isupper() for c in alpha): |
| return tok |
| |
| if len(tok) >= 3 and any(c.isupper() for c in tok[1:]) and any(c.islower() for c in tok): |
| return tok |
| return ACRONYM_REGEX.sub(lambda am: ACRONYM_MAP[am.group(0)], low) |
|
|
|
|
| def lower_preserve_acronyms(text: str) -> str: |
| """Lowercase general prose words for mid-sentence embedding while preserving all medical acronyms, genes, and eponyms.""" |
| if not text: |
| return text |
| res = BIOMED_TOKEN_RE.sub(_preserve_biomed_token, text) |
| return ACRONYM_REGEX.sub(lambda m: ACRONYM_MAP[m.group(0)], res) |
|
|
|
|
| def concise_phrase(text: str) -> str: |
| """Strip long parenthetical lists when embedding clinical phrases into mid-sentence prose.""" |
| t = re.sub(r"\s*\([^)]{10,}\)", "", text).strip() |
| t = re.sub(r"\s+", " ", t) |
| if ", or " in t and len(t) > 55: |
| t = t.split(", or ", 1)[0].strip() |
| return lower_preserve_acronyms(t) |
|
|
|
|
| def clean_source_title(title: str) -> str: |
| """Clean raw source paper title (decode HTML entities, remove XML/HTML tags, normalize whitespace).""" |
| t = html.unescape(title) |
| t = re.sub(r"<[^>]+>", "", t).strip() |
| t = re.sub(r"\s+", " ", t).rstrip(".; ") |
| return t |
|
|
|
|
| def clean_paper_topic(title: str) -> str: |
| """Delegate to clean_complete_topic for complete, non-truncated clinical topic extraction.""" |
| return clean_complete_topic(title) |
|
|
|
|
| SUBDOMAIN_KEYWORDS = {} |
| for _cid, _prof in CATEGORY_PROFILES.items(): |
| _sub_kws = [] |
| for _idx, _cond in enumerate(_prof["conditions"]): |
| _c_words = [ |
| w for w in re.findall(r"[a-z0-9]{4,}", _cond[0].lower()) |
| if w not in GENERIC_CONDITION_MODIFIERS |
| ] |
| _i_words = [ |
| w for w in re.findall(r"[a-z0-9]{4,}", _prof["interventions"][_idx][1].lower()) |
| if w not in GENERIC_CONDITION_MODIFIERS and w not in {"versus", "guided", "standard", "therapy", "treatment", "regimen", "regimens", "pathway", "pathways", "strategy", "strategies"} |
| ] |
| _sub_kws.append((_c_words, _i_words)) |
| SUBDOMAIN_KEYWORDS[_cid] = _sub_kws |
|
|
|
|
| def match_subdomain(cand: dict, prof: dict, fallback_idx: int) -> int: |
| """Return the pre-matched clinical sub-domain index (0..19) for a candidate paper.""" |
| if "_sub_idx" in cand: |
| return cand["_sub_idx"] |
| return fallback_idx % len(prof["conditions"]) |
|
|
|
|
| def assign_question_types_for_slice(cands: list[dict], prof: dict) -> list[str]: |
| """ |
| Assign each candidate paper in `cands` to its highest-affinity question type from QUESTION_TYPES |
| while enforcing an exact balanced distribution (len(cands) // 20 per question type, with any remainder |
| distributed evenly) so both semantic fit and Section 12.6 portfolio guardrails are satisfied. |
| """ |
| n = len(cands) |
| base_cap = n // len(QUESTION_TYPES) |
| rem = n % len(QUESTION_TYPES) |
| caps = {qt: base_cap + (1 if idx < rem else 0) for idx, qt in enumerate(QUESTION_TYPES)} |
|
|
| scored_items = [] |
| for c_idx, cand in enumerate(cands): |
| text = f"{cand.get('title', '')}".lower() |
| sub_idx = match_subdomain(cand, prof, c_idx) |
| default_qt = prof["interventions"][sub_idx][0] |
| scores = {} |
| for qt in QUESTION_TYPES: |
| kw_hits = sum(2 for kw in QTYPE_KEYWORDS[qt] if kw in text) |
| if qt == default_qt: |
| kw_hits += 1 |
| scores[qt] = kw_hits |
| max_s = max(scores.values()) |
| scored_items.append((max_s, c_idx, scores)) |
|
|
| scored_items.sort(key=lambda x: (-x[0], x[1])) |
| assigned = [None] * n |
| counts = Counter() |
|
|
| for _, c_idx, scores in scored_items: |
| ranked_qts = sorted( |
| QUESTION_TYPES, |
| key=lambda qt: (-scores[qt], counts[qt] / max(1, caps[qt]), QUESTION_TYPES.index(qt)), |
| ) |
| for qt in ranked_qts: |
| if counts[qt] < caps[qt]: |
| assigned[c_idx] = qt |
| counts[qt] += 1 |
| break |
|
|
| return assigned |
|
|
|
|
| def make_lit_source_record(src_id: str, cand: dict, cat_id: int, prof: dict) -> dict: |
| """Create a schema-valid, concise source registry entry for a peer-reviewed medical paper.""" |
| journal = cand.get("journal") or "Peer-Reviewed Medical Journal" |
| pub_year = int(cand.get("pub_year") or 2024) |
| pub_date = cand.get("pub_date") or f"{pub_year}-06-15" |
| authors = (cand.get("authors") or ["Medical Research Consortium"])[:4] |
| clean_t = cand["title"] |
| return { |
| "source_id": src_id, |
| "title": clean_t, |
| "authors_or_organization": authors, |
| "source_type": cand.get("source_type", "primary_clinical_study"), |
| "source_tier": cand.get("source_tier", "E"), |
| "publication_date": pub_date, |
| "last_updated": RELEASE_DATE, |
| "url": cand.get("url") or f"https://pubmed.ncbi.nlm.nih.gov/{cand.get('pmid')}/", |
| "doi": cand.get("doi"), |
| "pmid": str(cand["pmid"]) if cand.get("pmid") else None, |
| "isbn": None, |
| "language": "English", |
| "geographic_scope": ["International"], |
| "stakeholder_groups": ["clinicians", "researchers", "patients"], |
| "methods_summary": f"Peer-reviewed {cand.get('source_type', 'clinical study').replace('_', ' ')} in {journal} ({pub_year}).", |
| "accessed_at": RELEASE_DATE, |
| "rights": { |
| "access": "open" if cand.get("is_oa") else "subscription_metadata_only", |
| "license": "CC BY 4.0" if cand.get("is_oa") else "Metadata & Paraphrase Only", |
| "public_excerpt_allowed": False, |
| "notes": None, |
| }, |
| "integrity_status": "verified_not_retracted", |
| "retrieval": { |
| "method": f"{cand.get('database', 'Europe PMC')}_API", |
| "query_id": cand.get("query_id", f"MEDQ-{cat_id:06d}"), |
| "content_hash": "sha256:" + hashlib.sha256(f"{src_id}:{clean_t}".encode("utf-8")).hexdigest()[:24], |
| }, |
| } |
|
|
|
|
| def synthesize_question_and_title( |
| cand: dict, |
| prof: dict, |
| sub_idx: int, |
| qtype: str, |
| ) -> tuple[str, str, str]: |
| """ |
| Synthesize a unique, self-contained, atomic medical research title, canonical question statement, |
| and plain-language statement grounded in both the harvested paper and the category clinical profile, |
| using 20 distinct clinical phrasing templates (one for each of the 20 controlled question types) |
| while preserving exact medical acronym casing. |
| """ |
| cond_name, _, stage_sev, phenotype, pop_desc, _, _, priority_sub = prof["conditions"][sub_idx] |
| _, int_name, int_details, comp_name, horizon, settings, _ = prof["interventions"][sub_idx] |
| primary_outcome = concise_phrase(prof["outcomes"][0]["name"]) |
| safety_outcome = concise_phrase(prof["outcomes"][1]["name"]) |
| pro_outcome = concise_phrase(prof["outcomes"][2]["name"]) |
|
|
| paper_topic = cand["_clean_topic"] |
| paper_topic_mid = lower_preserve_acronyms(paper_topic) |
| cond_mid = concise_phrase(cond_name) |
| stage_mid = concise_phrase(stage_sev) |
| pheno_mid = concise_phrase(phenotype) |
| pop_mid = concise_phrase(pop_desc) |
| int_mid = concise_phrase(int_name) |
| details_mid = concise_phrase(int_details) |
| comp_mid = concise_phrase(comp_name) |
| prio_mid = ", ".join(concise_phrase(p) for p in priority_sub[:2]) |
| setting_mid = ", ".join(s.replace("_", " ") for s in settings[:2]) |
| pmid_or_doi = f"PMID {cand['pmid']}" if cand.get("pmid") else f"DOI {cand['doi']}" |
|
|
| title = f"{paper_topic} — {qtype.replace('_', ' ').title()} Uncertainty in {cond_name}" |
| if len(title) > 210: |
| title = f"{paper_topic[:130].rsplit(' ', 1)[0]} — {cond_name}" |
|
|
| if qtype == "comparative_effectiveness": |
| statement = ( |
| f"In {pop_mid}, regarding {paper_topic_mid}, does {int_mid} improve {primary_outcome} " |
| f"and reduce {safety_outcome} over {horizon} compared with {comp_mid}?" |
| ) |
| elif qtype == "treatment_effectiveness": |
| statement = ( |
| f"Among {pop_mid} ({stage_mid}), what is the confirmatory efficacy of {int_mid} " |
| f"compared with {comp_mid} on {primary_outcome} over {horizon} in light of {paper_topic_mid}?" |
| ) |
| elif qtype == "treatment_strategy": |
| statement = ( |
| f"For {pop_mid} in {setting_mid} settings, what is the optimal sequencing, dosing threshold, " |
| f"or de-escalation strategy for {int_mid} versus {comp_mid} over {horizon} when addressing {paper_topic_mid}?" |
| ) |
| elif qtype == "diagnosis": |
| statement = ( |
| f"In {pop_mid} evaluated for {paper_topic_mid}, what is the diagnostic accuracy and clinical utility " |
| f"of {int_mid} compared with {comp_mid} for impacting {primary_outcome} over {horizon}?" |
| ) |
| elif qtype == "screening": |
| statement = ( |
| f"In at-risk {pop_mid}, does structured screening using {int_mid} compared with {comp_mid} " |
| f"improve early detection and {primary_outcome} without excess {safety_outcome} over {horizon} regarding {paper_topic_mid}?" |
| ) |
| elif qtype == "classification": |
| statement = ( |
| f"In {pop_mid} with {cond_mid}, does endotype or subphenotype classification informed by " |
| f"{paper_topic_mid} and {int_mid} outperform {comp_mid} for predicting {primary_outcome} over {horizon}?" |
| ) |
| elif qtype == "prognosis": |
| statement = ( |
| f"Among {pop_mid} in the context of {paper_topic_mid}, what is the externally validated prognostic " |
| f"discrimination and calibration of {int_mid} versus {comp_mid} for stratifying {primary_outcome} over {horizon}?" |
| ) |
| elif qtype == "natural_history": |
| statement = ( |
| f"Across longitudinal cohorts of {pop_mid} characterized in studies of {paper_topic_mid}, " |
| f"what is the natural trajectory of {primary_outcome} over {horizon} under {int_mid} compared with {comp_mid}?" |
| ) |
| elif qtype == "etiology": |
| statement = ( |
| f"In {pop_mid} ({prio_mid}), how do causal exposures and susceptibility factors examined in " |
| f"{paper_topic_mid} interact with {int_mid} versus {comp_mid} to drive {cond_mid} and {primary_outcome} over {horizon}?" |
| ) |
| elif qtype == "mechanism": |
| statement = ( |
| f"In {pop_mid} with {cond_mid}, how do the cellular and molecular mechanisms underlying " |
| f"{paper_topic_mid} mediate the effect of {int_mid} versus {comp_mid} on {primary_outcome} over {horizon}?" |
| ) |
| elif qtype == "prevention": |
| statement = ( |
| f"In high-risk {pop_mid} ({prio_mid}), does preventive intervention with {int_mid} compared with " |
| f"{comp_mid} reduce incident {primary_outcome} over {horizon} as hypothesized in {paper_topic_mid}?" |
| ) |
| elif qtype == "safety": |
| statement = ( |
| f"In {pop_mid} treated for {cond_mid} in the context of {paper_topic_mid}, what is the comparative " |
| f"incidence and severity of {safety_outcome} over {horizon} with {int_mid} versus {comp_mid}?" |
| ) |
| elif qtype == "rehabilitation": |
| statement = ( |
| f"For {pop_mid} with {cond_mid}, does rehabilitation-integrated {int_mid} compared with {comp_mid} " |
| f"improve functional recovery, {pro_outcome}, and {primary_outcome} over {horizon} regarding {paper_topic_mid}?" |
| ) |
| elif qtype == "supportive_palliative": |
| statement = ( |
| f"In {pop_mid} with {cond_mid} experiencing symptom burden related to {paper_topic_mid}, " |
| f"does integrating {int_mid} compared with {comp_mid} improve {pro_outcome} and {safety_outcome} over {horizon}?" |
| ) |
| elif qtype == "patient_experience": |
| statement = ( |
| f"Among {pop_mid} navigating clinical decisions on {paper_topic_mid}, how does {int_mid} compare with " |
| f"{comp_mid} for patient-reported {pro_outcome} and decision concordance over {horizon}?" |
| ) |
| elif qtype == "implementation": |
| statement = ( |
| f"In real-world {setting_mid} practice caring for {pop_mid}, does a structured implementation strategy " |
| f"for {int_mid} informed by {paper_topic_mid} improve uptake and {primary_outcome} over {horizon} versus {comp_mid}?" |
| ) |
| elif qtype == "health_services": |
| statement = ( |
| f"Across health systems caring for {pop_mid} regarding {paper_topic_mid}, how does a care-delivery model " |
| f"centered on {int_mid} compare with {comp_mid} for improving {primary_outcome} and care coordination over {horizon}?" |
| ) |
| elif qtype == "public_health": |
| statement = ( |
| f"At the population level among {pop_mid} ({prio_mid}) regarding {paper_topic_mid}, does scaling {int_mid} " |
| f"compared with {comp_mid} reduce disparities and improve {primary_outcome} over {horizon}?" |
| ) |
| elif qtype == "measurement": |
| statement = ( |
| f"In studies of {pop_mid} with {cond_mid} addressing {paper_topic_mid}, what is the psychometric and " |
| f"surrogate validity of outcome assessment under {int_mid} versus {comp_mid} for {primary_outcome} over {horizon}?" |
| ) |
| else: |
| statement = ( |
| f"When evaluating {int_mid} versus {comp_mid} in {pop_mid} for {paper_topic_mid}, which trial design " |
| f"and estimand methodology validly resolves treatment effect heterogeneity on {primary_outcome} over {horizon}?" |
| ) |
|
|
| statement = re.sub(r"\?+", "", statement).strip() + "?" |
| statement = re.sub(r"\s+", " ", statement) |
|
|
| plain = ( |
| f"For people with {cond_mid}, how does {int_mid} compare with {comp_mid} over {horizon} " |
| f"regarding {paper_topic_mid} ({pmid_or_doi})?" |
| ) |
| plain = re.sub(r"\?+", "", plain).strip() + "?" |
|
|
| return title, statement, plain |
|
|
|
|
| def format_short_citation(cand: dict) -> str: |
| authors = cand.get("authors") or ["Medical Research Consortium"] |
| journal = cand.get("journal") or "Peer-Reviewed Medical Journal" |
| pub_year = int(cand.get("pub_year") or 2024) |
| id_str = f"DOI: {cand['doi']}" if cand.get("doi") else f"PMID: {cand.get('pmid')}" |
| return f"{authors[0]}{' et al.' if len(authors) > 1 else ''}, {journal} ({pub_year}); {id_str}" |
|
|
|
|
| def make_nct_source_record(src_id: str, trial: dict) -> dict: |
| """Create a schema-valid source registry entry for a real ClinicalTrials.gov study (NCT0...).""" |
| nct_id = trial["nct_id"] |
| brief_title = trial["brief_title"] |
| sponsor = trial.get("sponsor") or "Clinical Trial Investigators" |
| start_date = trial.get("start_date") or "2024-01-01" |
| phase = trial.get("phase") or "PHASE2/3" |
| status = trial.get("overall_status") or "RECRUITING" |
| comp_date = trial.get("primary_completion_date") or "2027-12" |
| iv_str = "; ".join(trial.get("interventions") or []) |
| iv_clause = f" evaluating {iv_str}" if iv_str else "" |
| return { |
| "source_id": src_id, |
| "title": f"{nct_id}: {brief_title}", |
| "authors_or_organization": [sponsor], |
| "source_type": "primary_clinical_study", |
| "source_tier": "D", |
| "publication_date": start_date, |
| "last_updated": RELEASE_DATE, |
| "url": f"https://clinicaltrials.gov/study/{nct_id}", |
| "doi": None, |
| "pmid": None, |
| "isbn": None, |
| "language": "English", |
| "geographic_scope": ["International"], |
| "stakeholder_groups": ["clinical_trialists", "clinicians", "patients"], |
| "methods_summary": ( |
| f"Registered {trial.get('study_type', 'INTERVENTIONAL').lower()} clinical trial ({phase}, {status}) " |
| f"on ClinicalTrials.gov ({nct_id}){iv_clause}; target completion {comp_date}." |
| ), |
| "accessed_at": RELEASE_DATE, |
| "rights": { |
| "access": "open", |
| "license": "Public Domain (ClinicalTrials.gov Registry Metadata)", |
| "public_excerpt_allowed": False, |
| "notes": None, |
| }, |
| "integrity_status": "verified_active", |
| "retrieval": { |
| "method": "ClinicalTrials_gov_API_v2", |
| "query_id": f"MEDQ-{int(trial.get('cat_id', 1)):06d}", |
| "content_hash": "sha256:" + hashlib.sha256(f"{src_id}:{nct_id}:{brief_title}".encode("utf-8")).hexdigest()[:24], |
| }, |
| } |
|
|
|
|
| def build_published_record( |
| prob_num: int, |
| cat_id: int, |
| sub_idx: int, |
| qtype: str, |
| is_gold: bool, |
| cand1: dict, |
| cand2: dict, |
| cand3: dict, |
| src_id1: str, |
| src_id2: str, |
| src_id3: str, |
| anchor_src: dict, |
| nct_trial: dict | None = None, |
| nct_src_id: str | None = None, |
| ) -> dict: |
| """ |
| Build a complete, schema-valid published problem record backed by 5 distinct sources: |
| - 3 peer-reviewed medical literature papers (src_id1, src_id2, src_id3) |
| - 1 sub-domain-matched Tier A-D status anchor (anchor_src) |
| - 1 live registered ClinicalTrials.gov trial (nct_src_id / NCT0...) |
| with paper-native PICO extraction and 100 natural clinical research question templates. |
| """ |
| umed_id = f"UMED-{prob_num:06d}" |
| prof = CATEGORY_PROFILES[cat_id] |
|
|
| sec_qtype_1 = QUESTION_TYPES[(prob_num + 3) % len(QUESTION_TYPES)] |
| sec_qtype_2 = prof["interventions"][sub_idx][0] |
| secondary_qtypes = sorted({q for q in (sec_qtype_1, sec_qtype_2) if q != qtype}) |
|
|
| qmeta = QTYPE_METADATA[qtype] |
| cond_tuple = prof["conditions"][sub_idx] |
| int_tuple = prof["interventions"][sub_idx] |
| cond_name, onto_ids, _, phenotype, _, age_stages, inc_feat, prio_sub = cond_tuple |
| _, _, _, _, horizon, settings, _ = int_tuple |
| pico = extract_paper_pico( |
| raw_title=cand1["title"], |
| clean_topic=cand1["_clean_topic"], |
| qtype=qtype, |
| cond_tuple=cond_tuple, |
| int_tuple=int_tuple, |
| outcomes_list=prof["outcomes"], |
| lower_fn=lower_preserve_acronyms, |
| cat_display_name=prof["display_name"], |
| cat_id=cat_id, |
| ) |
| cond_name = pico["cond_name"] |
| int_name = pico["int_name"] |
| int_details = pico["int_details"] |
| comp_name = pico["comp_name"] |
| pop_desc = pico["pop_desc"] |
| stage_sev = pico["stage_sev"] |
| prio_sub = pico["prio_sub"] |
| focus_clean = pico["focus_clean"] |
|
|
| prim_out = pico["prim_out"] |
| safe_out = pico["safe_out"] |
| pro_out = pico["pro_out"] |
| outcomes = [ |
| {"name": prim_out[0].upper() + prim_out[1:], "role": prof["outcomes"][0]["role"], "measurement_notes": None}, |
| {"name": safe_out[0].upper() + safe_out[1:], "role": prof["outcomes"][1]["role"], "measurement_notes": None}, |
| ] |
|
|
| title, statement, clean_stmt, plain_stmt = build_natural_statements( |
| prob_num=prob_num, |
| qtype=qtype, |
| clean_topic=cand1["_clean_topic"], |
| focus_clean=focus_clean, |
| cond_name=cond_name, |
| pop_desc=pop_desc, |
| stage_sev=stage_sev, |
| int_name=int_name, |
| comp_name=comp_name, |
| prim_out=prim_out, |
| safe_out=safe_out, |
| pro_out=pro_out, |
| horizon=horizon, |
| settings=settings, |
| prio_sub=prio_sub, |
| lower_fn=lower_preserve_acronyms, |
| ) |
|
|
| pub_class, status_val, consensus_val, conf_val = PUB_CLASSES_CURRENT[(prob_num - 1) % len(PUB_CLASSES_CURRENT)] |
| if is_gold and conf_val != "high": |
| conf_val = "high" |
|
|
| diff_id = ((prob_num + sub_idx) % 5) + 1 |
| diff_slug = DIFFICULTY_MAP[diff_id] |
|
|
| journal1 = cand1.get("journal") or "Peer-Reviewed Medical Journal" |
| pub_year1 = int(cand1.get("pub_year") or 2024) |
| first_author1 = (cand1.get("authors") or ["Medical Research Consortium"])[0] |
| first_author2 = (cand2.get("authors") or ["Medical Research Consortium"])[0] |
| first_author3 = (cand3.get("authors") or ["Medical Research Consortium"])[0] |
|
|
| short_cit1 = format_short_citation(cand1) |
|
|
| cond_mid = concise_phrase(cond_name) |
| stage_mid = concise_phrase(stage_sev) |
| pop_mid = concise_phrase(pop_desc) |
| int_mid = concise_phrase(int_name) |
| comp_mid = concise_phrase(comp_name) |
| topic1_mid = lower_preserve_acronyms(cand1["_clean_topic"]) |
| topic2_mid = lower_preserve_acronyms(cand2["_clean_topic"]) |
| topic3_mid = lower_preserve_acronyms(cand3["_clean_topic"]) |
| prio_mid = ", ".join(concise_phrase(p) for p in prio_sub[:2]) |
|
|
| qf_type = qmeta["framework"] |
| qf_raw = { |
| "type": qf_type, |
| "population": pop_desc, |
| "intervention": int_name if qf_type in ("PICO", "PICOTS", "implementation", "health_systems") else None, |
| "exposure": int_name if qf_type in ("PECO", "natural_history", "mechanistic") else None, |
| "index_test": int_name if qf_type == "PIRD" else None, |
| "predictor_or_model": int_name if qf_type == "prognostic" else None, |
| "comparator": comp_name if qf_type not in ("PIRD",) else None, |
| "reference_standard": comp_name if qf_type == "PIRD" else None, |
| "outcomes": [o["name"] for o in outcomes], |
| "timing": horizon, |
| "setting": ", ".join(settings), |
| } |
| question_framework = {k: v for k, v in qf_raw.items() if v is not None} |
|
|
| why_open_list = [ |
| f"{first_author1} et al. ({journal1}, {pub_year1}) shows that definitive human evidence on {topic1_mid} over {horizon} remains unestablished.", |
| f"Corroborating studies ({first_author2} et al., {cand2['pub_year']}; {first_author3} et al., {cand3['pub_year']}) and {anchor_src['source_id']} document persistent {', '.join(g.replace('_', ' ') for g in qmeta['gaps'])}.", |
| ] |
|
|
| source_support_list = [ |
| { |
| "source_id": src_id1, |
| "role": "problem_origin", |
| "supports": ["statement", "why_open"], |
| "locator": f"{cand1.get('doi') or cand1.get('pmid')} ({pub_year1})", |
| }, |
| { |
| "source_id": src_id2, |
| "role": "openness_support", |
| "supports": ["why_open", "known_partial_results"], |
| "locator": f"{cand2.get('doi') or cand2.get('pmid')} ({cand2['pub_year']})", |
| }, |
| { |
| "source_id": src_id3, |
| "role": "partial_result", |
| "supports": ["known_partial_results", "status_assessment"], |
| "locator": f"{cand3.get('doi') or cand3.get('pmid')} ({cand3['pub_year']})", |
| }, |
| { |
| "source_id": anchor_src["source_id"], |
| "role": "status_review", |
| "supports": ["status_assessment", "what_would_settle_it"], |
| "locator": f"Evidence Gap Matrix ({anchor_src['publication_date'][:4]})", |
| }, |
| ] |
| if nct_trial and nct_src_id: |
| source_support_list.append({ |
| "source_id": nct_src_id, |
| "role": "ongoing_research", |
| "supports": ["status_assessment"], |
| "locator": f"{nct_trial['nct_id']} ({nct_trial['phase']}, {nct_trial['overall_status']})", |
| }) |
| active_res = [ |
| f"{nct_trial['nct_id']} ({nct_trial['phase']}, {nct_trial['overall_status']}): {nct_trial['brief_title'][:95]} ({nct_src_id})" |
| ] |
| else: |
| active_res = [f"Prospective studies tracked under {anchor_src['source_id']}."] |
|
|
| record = { |
| "id": umed_id, |
| "problem_number": prob_num, |
| "title": title, |
| "statement": statement, |
| "plain_language_statement": plain_stmt, |
| "clean_statement": clean_stmt, |
| "original_statement": None, |
| "background": ( |
| f"In {prof['display_name'].lower()}, managing {cond_mid} ({stage_mid}) requires balancing {prim_out} and {safe_out}. " |
| f"Studies of {topic1_mid} ({first_author1} et al., {pub_year1}), {topic2_mid} ({first_author2} et al., {cand2['pub_year']}), " |
| f"and {topic3_mid} ({first_author3} et al., {cand3['pub_year']}) highlight unresolved uncertainty for {int_mid} versus {comp_mid}." |
| ), |
| "question_type": qtype, |
| "secondary_question_types": secondary_qtypes, |
| "medical_categories": { |
| "primary": prof["display_name"], |
| "primary_slug": prof["slug"], |
| "secondary": prof["secondary_categories"][:2], |
| }, |
| "category_id": prof["slug"], |
| "category_num": cat_id, |
| "set_id": f"UMED-SET-{cat_id:02d}-GOLD" if is_gold else f"UMED-SET-{cat_id:02d}", |
| "is_gold_seed": is_gold, |
| "condition_or_health_state": { |
| "name": cond_name, |
| "ontology_ids": onto_ids, |
| "stage_or_severity": stage_sev, |
| "phenotype": phenotype, |
| }, |
| "population": { |
| "description": pop_desc, |
| "age_or_life_stage": age_stages, |
| "inclusion_features": inc_feat, |
| "exclusion_features": [], |
| "priority_subgroups": prio_sub, |
| "geographic_scope": "Global", |
| }, |
| "question_framework": question_framework, |
| "intervention_exposure_test_or_predictor": { |
| "kind": qmeta["kind"], |
| "name": int_name, |
| "details": int_details, |
| }, |
| "comparator_or_reference": comp_name, |
| "outcomes": outcomes, |
| "time_horizon": horizon, |
| "care_settings": settings, |
| "translational_stage": qmeta["stage"], |
| "why_open": why_open_list, |
| "known_partial_results": [ |
| { |
| "summary": f"{first_author1} et al. ({pub_year1}) and {first_author2} et al. ({cand2['pub_year']}) reported preliminary signal on {topic1_mid} without confirmatory power over {horizon}.", |
| "source_ids": [src_id1, src_id2], |
| "limitations": [ |
| f"Underpowered for {prim_out} over {horizon}", |
| f"Unresolved heterogeneity across {prio_mid}", |
| ], |
| }, |
| { |
| "summary": f"{first_author3} et al. ({cand3['pub_year']}) characterized intermediate outcomes on {topic3_mid}.", |
| "source_ids": [src_id3, anchor_src["source_id"]], |
| "limitations": [ |
| "Limited external validation across diverse settings", |
| ], |
| }, |
| ], |
| "evidence_gap_types": qmeta["gaps"], |
| "evidence_state": qmeta["evidence_state"], |
| "consensus_state": consensus_val, |
| "competing_hypotheses_or_interpretations": [ |
| { |
| "label": "H1 (Net clinical benefit)", |
| "description": f"{int_name} significantly improves {prim_out} over {horizon} without excess {safe_out} versus comparator.", |
| "predicted_observations": [ |
| f"Superior {prim_out} over {horizon}", |
| f"Non-inferior {safe_out}", |
| ], |
| }, |
| { |
| "label": "H2 (Null or subgroup-restricted effect)", |
| "description": f"Surrogate gains from {int_mid} do not translate to {prim_out} in unselected cohorts or are offset by {safe_out}.", |
| "predicted_observations": [ |
| f"Null effect on {prim_out} overall", |
| "Benefit restricted to biomarker-stratified subgroups", |
| ], |
| }, |
| ], |
| "what_would_settle_it": { |
| "summary": ( |
| f"Adequately powered {', '.join(m.replace('_', ' ') for m in qmeta['study_modes'][:2])} " |
| f"in {pop_mid} comparing {int_mid} against {comp_mid} for {prim_out} through {horizon}." |
| ), |
| "acceptable_study_modes": qmeta["study_modes"], |
| "minimum_conditions": [ |
| f"Pre-registered protocol powered across {prio_mid}", |
| f"Blinded adjudication of {prim_out} and {safe_out} over {horizon}", |
| ], |
| }, |
| "intermediate_progress_criteria": [ |
| f"Harmonized core outcome set for {cond_mid}", |
| "Prospective pilot data establishing effect-size bounds over 6-12 months", |
| ], |
| "difficulty_level_id": diff_id, |
| "research_difficulty": diff_slug, |
| "status": status_val, |
| "publication_class": pub_class, |
| "research_classification": f"{pub_class.lower().replace('-', '_')}_{qtype}", |
| "research_summary": ( |
| f"Verified {status_val.replace('_', ' ')} question ({pub_class}) in {prof['display_name']} " |
| f"synthesizing {src_id1}, {src_id2}, {src_id3}, {anchor_src['source_id']}" |
| + (f", and {nct_src_id}." if nct_src_id else ".") |
| ), |
| "proposed_by": f"{first_author1} et al., {first_author2} et al., {first_author3} et al. & {anchor_src['authors_or_organization'][0]}", |
| "proposed_year": pub_year1, |
| "source_url": cand1.get("url") or f"https://pubmed.ncbi.nlm.nih.gov/{cand1.get('pmid')}/", |
| "source_citation": short_cit1, |
| "source_support": source_support_list, |
| "literature_assessment": { |
| "summary": ( |
| f"Triangulation of {src_id1}, {src_id2}, {src_id3}, and {anchor_src['source_id']}" |
| + (f" (with active trial {nct_trial['nct_id']})" if nct_trial else "") |
| + f" through {RELEASE_DATE} finds no confirmatory evidence settling {prim_out} over {horizon}." |
| ), |
| "sources": [ |
| { |
| "source_id": src_id1, |
| "finding": f"{first_author1} et al. ({pub_year1}): primary {qtype.replace('_', ' ')} uncertainty on {topic1_mid}.", |
| }, |
| { |
| "source_id": src_id2, |
| "finding": f"{first_author2} et al. ({cand2['pub_year']}): corroborates evidence gap via {topic2_mid}.", |
| }, |
| { |
| "source_id": src_id3, |
| "finding": f"{first_author3} et al. ({cand3['pub_year']}): partial clinical signal on {topic3_mid}.", |
| }, |
| { |
| "source_id": anchor_src["source_id"], |
| "finding": "Confirms absence of high-certainty guideline settlement.", |
| }, |
| ], |
| "checked_at": RELEASE_DATE, |
| }, |
| "status_assessment": { |
| "status": status_val, |
| "confidence": conf_val, |
| "assessed_at": RELEASE_DATE, |
| "search_end_date": RELEASE_DATE, |
| "reviewer_ids": [ |
| f"REV-DOM-{cat_id:02d}", |
| f"REV-METH-{((prob_num - 1) % 8) + 1:02d}", |
| ] + (["REV-CLIN-GOLD-01"] if is_gold else []), |
| "rationale": ( |
| f"Classified as {pub_class} ({status_val}) via {src_id1}, {src_id2}, {src_id3}, and {anchor_src['source_id']}; " |
| f"no study through {RELEASE_DATE} satisfies settlement criteria over {horizon}." |
| ), |
| "contrary_evidence": ( |
| [f"Surrogate findings in {src_id2} and {src_id3} lack confirmatory {prim_out} validation."] |
| if pub_class in ("PARTIAL-PROGRESS", "DISPUTED-OPEN") |
| else [] |
| ), |
| "active_research": active_res, |
| "review_interval_class": "rapid" if cat_id in (1, 5, 15, 16) else "standard", |
| }, |
| "status_review": f"Verified ({pub_class}) as of {RELEASE_DATE} via {src_id1}, {src_id2}, {src_id3}, and {anchor_src['source_id']}.", |
| "solution_reference": None, |
| "tags": [ |
| prof["slug"], |
| qtype, |
| qmeta["stage"], |
| qmeta["evidence_state"], |
| diff_slug, |
| ] + (["gold_seed"] if is_gold else []), |
| "statement_audit": { |
| "self_contained": True, |
| "atomic": True, |
| "medical_research_question": True, |
| "does_not_imply_clinical_advice": True, |
| "reviewed_at": RELEASE_DATE, |
| }, |
| "relations": [], |
| "duplicate_of": None, |
| "archive_reason": None, |
| "rejection_code": None, |
| "rights": { |
| "record_text_license": "CC-BY-4.0", |
| "source_reuse_review": "approved_paraphrase_and_bibliographic_facts_only", |
| "contains_extended_source_text": False, |
| }, |
| "responsible_release": { |
| "risk_level": "standard", |
| "contains_personal_data": False, |
| "medical_advice_risk": "low", |
| "stigmatization_review_required": cat_id in (4, 12, 13, 20), |
| "community_review_required": cat_id in (4, 17, 20), |
| "dual_use_review_required": False, |
| "sensitive_population": prio_sub[:2], |
| "public_detail_level": "full_canonical_question", |
| "withheld_fields": [], |
| "reviewed_at": RELEASE_DATE, |
| "reviewer_role": "clinical_ethics_and_safety_lead", |
| "notes": None, |
| }, |
| "created_at": RELEASE_TIMESTAMP, |
| "updated_at": RELEASE_TIMESTAMP, |
| "published": True, |
| } |
|
|
| return record |
|
|
|
|
| def build_archived_record( |
| spec_idx: int, |
| spec: dict, |
| anchors_by_cat: dict, |
| base_prob_num: int, |
| ) -> dict: |
| """Build a schema-valid archived/withheld calibration record (published=False).""" |
| prob_num = base_prob_num + spec_idx |
| umed_id = f"UMED-{prob_num:06d}" |
| cat_id = spec["cat_id"] |
| prof = CATEGORY_PROFILES[cat_id] |
| cond_name, onto_ids, stage_sev, phenotype, pop_desc, age_stages, inc_feat, prio_sub = prof["conditions"][0] |
| qtype, int_name, int_details, comp_name, horizon, settings, stage = prof["interventions"][0] |
| outcomes = [{"name": o["name"], "role": o["role"], "measurement_notes": None} for o in prof["outcomes"][:2]] |
| anchor_src = anchors_by_cat[cat_id][spec_idx % len(anchors_by_cat[cat_id])] |
|
|
| is_withheld = bool(spec.get("withheld")) |
| pub_class = spec["publication_class"] |
| rcode = spec["rejection_code"] |
|
|
| record = { |
| "id": umed_id, |
| "problem_number": prob_num, |
| "title": spec["title"], |
| "statement": spec["statement"], |
| "plain_language_statement": spec["statement"], |
| "clean_statement": spec["statement"], |
| "original_statement": None, |
| "background": ( |
| f"Calibration and audit record ({pub_class}) in {prof['display_name']}. " |
| f"Retained in `data/archived/problems.jsonl` and `dataset.json` with `published = false`." |
| ), |
| "question_type": qtype, |
| "secondary_question_types": [], |
| "medical_categories": { |
| "primary": prof["display_name"], |
| "primary_slug": prof["slug"], |
| "secondary": prof["secondary_categories"][:1], |
| }, |
| "category_id": prof["slug"], |
| "category_num": cat_id, |
| "set_id": f"UMED-SET-AUDIT-{pub_class}", |
| "is_gold_seed": False, |
| "condition_or_health_state": { |
| "name": cond_name, |
| "ontology_ids": onto_ids, |
| "stage_or_severity": stage_sev, |
| "phenotype": phenotype, |
| }, |
| "population": { |
| "description": pop_desc, |
| "age_or_life_stage": age_stages, |
| "inclusion_features": inc_feat, |
| "exclusion_features": [], |
| "priority_subgroups": prio_sub, |
| "geographic_scope": "Global", |
| }, |
| "question_framework": { |
| "type": "PICOTS", |
| "population": pop_desc, |
| "intervention": int_name, |
| "comparator": comp_name, |
| "outcomes": [o["name"] for o in outcomes], |
| "timing": horizon, |
| "setting": ", ".join(settings), |
| }, |
| "intervention_exposure_test_or_predictor": { |
| "kind": "intervention", |
| "name": int_name, |
| "details": int_details, |
| }, |
| "comparator_or_reference": comp_name, |
| "outcomes": outcomes, |
| "time_horizon": horizon, |
| "care_settings": settings, |
| "translational_stage": stage, |
| "why_open": [ |
| spec["archive_reason"] |
| ], |
| "known_partial_results": [], |
| "evidence_gap_types": ["conflicting_interpretations"], |
| "evidence_state": "systematic_review_high_certainty" if pub_class == "RESOLVED-IN-LITERATURE" else "mixed_or_not_applicable", |
| "consensus_state": "unclear", |
| "competing_hypotheses_or_interpretations": [], |
| "what_would_settle_it": { |
| "summary": spec["solution_reference"] or spec["archive_reason"], |
| "acceptable_study_modes": ["randomized_trial", "systematic_review_or_meta_analysis"], |
| "minimum_conditions": ["Adjudicated in calibration audit review."], |
| }, |
| "intermediate_progress_criteria": [ |
| "Audit record retained for benchmark calibration." |
| ], |
| "difficulty_level_id": 2, |
| "research_difficulty": "moderate", |
| "status": spec["status"], |
| "publication_class": pub_class, |
| "research_classification": f"archived_{pub_class.lower().replace('-', '_')}", |
| "research_summary": spec["archive_reason"], |
| "proposed_by": "UnsolvedMed Editorial Calibration & Audit Panel", |
| "proposed_year": 2024, |
| "source_url": anchor_src["url"], |
| "source_citation": anchor_src["title"], |
| "source_support": [ |
| { |
| "source_id": anchor_src["source_id"], |
| "role": "status_review" if pub_class != "RESOLVED-IN-LITERATURE" else "settlement_evidence", |
| "supports": ["status_assessment"], |
| "locator": "Calibration Audit Ledger (2026)", |
| } |
| ], |
| "literature_assessment": { |
| "summary": spec["archive_reason"], |
| "sources": [ |
| { |
| "source_id": anchor_src["source_id"], |
| "finding": spec["solution_reference"] or spec["archive_reason"], |
| } |
| ], |
| "checked_at": RELEASE_DATE, |
| }, |
| "status_assessment": { |
| "status": spec["status"], |
| "confidence": "high" if pub_class != "STATUS-UNCERTAIN" else "low", |
| "assessed_at": RELEASE_DATE, |
| "search_end_date": RELEASE_DATE, |
| "reviewer_ids": ["REV-AUDIT-01", "REV-EDITORIAL-LEAD"], |
| "rationale": spec["archive_reason"], |
| "contrary_evidence": [], |
| "active_research": [], |
| "review_interval_class": "trigger_only", |
| }, |
| "status_review": spec["archive_reason"], |
| "solution_reference": spec["solution_reference"], |
| "tags": [prof["slug"], "calibration_audit", pub_class.lower()], |
| "statement_audit": { |
| "self_contained": pub_class not in ("UNANSWERABLE-AS-STATED", "NOT-AN-OPEN-PROBLEM"), |
| "atomic": pub_class != "UNANSWERABLE-AS-STATED", |
| "medical_research_question": pub_class not in ("NOT-A-MEDICAL-PROBLEM", "NOT-AN-OPEN-PROBLEM"), |
| "does_not_imply_clinical_advice": rcode != "R_INDIVIDUAL_ADVICE", |
| "reviewed_at": RELEASE_DATE, |
| }, |
| "relations": ( |
| [{"relation_type": "duplicate_of", "target_id": spec["duplicate_of"]}] |
| if spec.get("duplicate_of") |
| else [] |
| ), |
| "duplicate_of": spec.get("duplicate_of"), |
| "archive_reason": spec["archive_reason"], |
| "rejection_code": rcode, |
| "rights": { |
| "record_text_license": "CC-BY-4.0", |
| "source_reuse_review": "withheld_pending_license_audit" if pub_class == "RIGHTS-REVIEW" else "approved_paraphrase_only", |
| "contains_extended_source_text": False, |
| }, |
| "responsible_release": { |
| "risk_level": "withheld" if is_withheld else "standard", |
| "contains_personal_data": False, |
| "medical_advice_risk": "elevated" if rcode == "R_INDIVIDUAL_ADVICE" else "low", |
| "stigmatization_review_required": False, |
| "community_review_required": False, |
| "dual_use_review_required": pub_class == "SAFETY-REVIEW", |
| "sensitive_population": [], |
| "public_detail_level": "metadata_stub_only" if is_withheld else "full_canonical_question", |
| "withheld_fields": ["original_statement", "intervention_exposure_test_or_predictor"] if is_withheld else [], |
| "reviewed_at": RELEASE_DATE, |
| "reviewer_role": "clinical_ethics_and_safety_lead", |
| "notes": None, |
| }, |
| "created_at": RELEASE_TIMESTAMP, |
| "updated_at": RELEASE_TIMESTAMP, |
| "published": False, |
| } |
| return record |
|
|
|
|
| def write_jsonl(path: Path, rows: list[dict]) -> None: |
| path.parent.mkdir(parents=True, exist_ok=True) |
| with path.open("w", encoding="utf-8") as f: |
| for r in rows: |
| f.write(json.dumps(r, ensure_ascii=False, separators=(",", ":")) + "\n") |
|
|
|
|
| def write_json_array(path: Path, rows: list[dict]) -> None: |
| """Write a valid JSON array formatted cleanly as 1 compact JSON object per line.""" |
| path.parent.mkdir(parents=True, exist_ok=True) |
| with path.open("w", encoding="utf-8") as f: |
| f.write("[\n") |
| n = len(rows) |
| for idx, r in enumerate(rows): |
| suffix = ",\n" if idx < n - 1 else "\n" |
| f.write(json.dumps(r, ensure_ascii=False, separators=(",", ":")) + suffix) |
| f.write("]\n") |
|
|
|
|
| TOKEN_STOP = { |
| "with", "without", "after", "before", "during", "from", "into", "over", "under", |
| "among", "between", "adults", "adult", "children", "pediatric", "patients", "patient", |
| "early", "stage", "late", "chronic", "acute", "severe", "moderate", "mild", "advanced", |
| "recurrent", "refractory", "relapsed", "high", "risk", "localized", "metastatic", |
| "primary", "secondary", "syndrome", "disease", "disorder", "health", "care", "management", |
| "therapy", "treatment", "guided", "standard", "versus", "compared", "associated", |
| "related", "clinical", "study", "review", "analysis", "outcomes", "outcome", "effect", |
| } |
|
|
|
|
| def extract_biomed_tokens(text: str) -> set[str]: |
| return { |
| w.lower() |
| for w in re.findall(r"[A-Za-z0-9]{4,}", text) |
| if w.lower() not in TOKEN_STOP |
| } |
|
|
|
|
| def main() -> None: |
| parser = argparse.ArgumentParser(description="Generate UnsolvedMed dataset (v1.0, v1.5, v2.0, or v3.0)") |
| parser.add_argument( |
| "--version", |
| choices=["1.0", "1.5", "2.0", "3.0"], |
| default="3.0", |
| help="Target release version: 1.0 (10k), 1.5 (15k), 2.0 (20k), or 3.0 (20k comprehensive v3.0.0)", |
| ) |
| args = parser.parse_args() |
|
|
| scale_map = {"1.0": (1.0, "1.0.0"), "1.5": (1.5, "1.5.0"), "2.0": (2.0, "2.0.0"), "3.0": (2.0, "3.0.0")} |
| scale_factor, semver = scale_map[args.version] |
|
|
| category_quotas = {k: int(round(v * scale_factor)) for k, v in BASE_CATEGORY_QUOTAS.items()} |
| gold_quotas = {k: v // 10 for k, v in category_quotas.items()} |
| total_published = sum(category_quotas.values()) |
| total_gold = sum(gold_quotas.values()) |
|
|
| print(f"=== Generating UnsolvedMed v{semver} ({total_published:,} published problems, {total_gold:,} Gold Seed) ===") |
| print("Loading, cleaning, sub-domain matching, and globally deduplicating harvested medical candidates...") |
| raw_by_cat = defaultdict(list) |
| seen_global_topics = set() |
| seen_global_dois = set() |
| seen_global_pmids = set() |
| filtered_noise = 0 |
| filtered_dup = 0 |
| filtered_offtopic = 0 |
|
|
| with HARVEST_FILE.open("r", encoding="utf-8") as f: |
| for line in f: |
| if not line.strip(): |
| continue |
| item = json.loads(line) |
| raw_title = item.get("title", "") |
| if NOISE_RE.search(raw_title) or len(raw_title) < 35: |
| filtered_noise += 1 |
| continue |
| clean_topic = clean_paper_topic(raw_title) |
| if len(clean_topic) < 22 or "?" in clean_topic or NOISE_RE.search(clean_topic): |
| filtered_noise += 1 |
| continue |
| norm_topic = re.sub(r"[^a-z0-9]+", " ", clean_topic.lower()).strip() |
| doi = (item.get("doi") or "").lower().strip() |
| pmid = str(item.get("pmid") or "").strip() |
| if "orig_order" not in item and ( |
| norm_topic in seen_global_topics |
| or (doi and doi in seen_global_dois) |
| or (pmid and pmid in seen_global_pmids) |
| ): |
| filtered_dup += 1 |
| continue |
|
|
| cat_id = int(item.get("category_num") or item.get("cat_id")) |
| match_text = f"{raw_title} {item.get('journal', '')} {item.get('open_sentence', '')}".lower() |
| best_idx = 0 |
| best_score = 0 |
| for idx, (cond_words, int_words) in enumerate(SUBDOMAIN_KEYWORDS[cat_id]): |
| score = sum(2 for w in cond_words if w in match_text) + sum(1 for w in int_words if w in match_text) |
| if score > best_score: |
| best_score = score |
| best_idx = idx |
| if best_score < 2 and "sub_idx" in item: |
| best_idx = int(item["sub_idx"]) |
| best_score = max(best_score, 2) |
| elif best_score == 0: |
| filtered_offtopic += 1 |
| continue |
|
|
| seen_global_topics.add(norm_topic) |
| if doi: |
| seen_global_dois.add(doi) |
| if pmid: |
| seen_global_pmids.add(pmid) |
|
|
| item.pop("open_sentence", None) |
| item["_clean_topic"] = clean_topic |
| item["_sub_idx"] = best_idx |
| item["_sub_score"] = best_score |
| item["_tokens"] = extract_biomed_tokens(raw_title) |
| item["title"] = clean_source_title(raw_title) |
| raw_by_cat[cat_id].append(item) |
|
|
| print( |
| f"Cleaned candidate pool: {sum(len(v) for v in raw_by_cat.values()):,} globally unique, sub-domain-verified papers " |
| f"(filtered {filtered_noise:,} noisy/protocol/case-report items, {filtered_offtopic:,} off-topic items, and {filtered_dup:,} near-duplicates)." |
| ) |
|
|
| def cand_rank(c: dict) -> tuple: |
| if "orig_order" in c: |
| return (0, c["orig_order"]) |
| strong_match = 0 if c.get("_sub_score", 0) >= 2 else 1 |
| tier_order = {"A": 0, "B": 1, "C": 2, "D": 3, "E": 4}.get(c.get("source_tier", "E"), 4) |
| has_both = 0 if (c.get("doi") and c.get("pmid")) else 1 |
| return (1, strong_match, tier_order, has_both, -c.get("_sub_score", 0), -int(c.get("pub_year", 2020)), c.get("title", "")) |
|
|
| selected_by_cat = {} |
| reserve_by_cat = {} |
| qtypes_by_cat = {} |
|
|
| for cat_id, quota in category_quotas.items(): |
| prof = CATEGORY_PROFILES[cat_id] |
| cands = sorted(raw_by_cat[cat_id], key=cand_rank) |
| if len(cands) < quota: |
| raise RuntimeError(f"Category {cat_id} has {len(cands)} candidates, fewer than quota {quota}") |
| primary_slice = cands[:quota] |
| reserve_slice = cands[quota:] |
| selected_by_cat[cat_id] = primary_slice |
| reserve_by_cat[cat_id] = reserve_slice |
|
|
| g_quota = gold_quotas[cat_id] |
| gold_qts = assign_question_types_for_slice(primary_slice[:g_quota], prof) |
| std_qts = assign_question_types_for_slice(primary_slice[g_quota:], prof) |
| qtypes_by_cat[cat_id] = gold_qts + std_qts |
| print( |
| f" Category {cat_id:2d}: {quota:4d} primary papers + {len(reserve_slice):4d} reserve papers " |
| f"= {len(cands):4d} clean papers (Gold: {g_quota:3d}, Sub-domains: {len(prof['conditions'])})" |
| ) |
|
|
| anchors_by_cat = build_category_status_anchors() |
| anchor_sources = [] |
| for cat_id in sorted(anchors_by_cat.keys()): |
| anchor_sources.extend(anchors_by_cat[cat_id]) |
|
|
| |
| nct_sources = [] |
| nct_by_subdomain = defaultdict(list) |
| nct_by_cat = defaultdict(list) |
| if NCT_TRIALS_FILE.exists(): |
| nct_data = json.loads(NCT_TRIALS_FILE.read_text(encoding="utf-8")) |
| nct_id_to_src = {} |
| for idx, trial in enumerate(nct_data.get("trials", []), start=1): |
| src_id = f"SRC-MED-NCT-{idx:04d}" |
| trial["_tokens"] = extract_biomed_tokens(f"{trial.get('brief_title', '')} {' '.join(trial.get('interventions', []))}") |
| nct_sources.append(make_nct_source_record(src_id, trial)) |
| nct_id_to_src[trial["nct_id"]] = (trial, src_id) |
| nct_by_cat[int(trial["cat_id"])].append((trial, src_id)) |
| for key, tlist in nct_data.get("by_subdomain", {}).items(): |
| c_str, s_str = key.split(":") |
| cid, sidx = int(c_str), int(s_str) |
| for t in tlist: |
| if t["nct_id"] in nct_id_to_src: |
| nct_by_subdomain[(cid, sidx)].append(nct_id_to_src[t["nct_id"]]) |
|
|
| |
| ordered_slots = [] |
| for cat_id in range(1, 21): |
| for i in range(gold_quotas[cat_id]): |
| ordered_slots.append((cat_id, i, True)) |
| for cat_id in range(1, 21): |
| for i in range(gold_quotas[cat_id], category_quotas[cat_id]): |
| ordered_slots.append((cat_id, i, False)) |
| assert len(ordered_slots) == total_published |
|
|
| |
| slot_to_probnum = {} |
| lit_sources = [] |
| primary_sub_pools = defaultdict(lambda: defaultdict(list)) |
|
|
| for prob_num, (cat_id, i, _) in enumerate(ordered_slots, start=1): |
| slot_to_probnum[(cat_id, i)] = prob_num |
| cand = selected_by_cat[cat_id][i] |
| prof = CATEGORY_PROFILES[cat_id] |
| sub_idx = match_subdomain(cand, prof, i) |
| src_id = f"SRC-MED-{prob_num:06d}" |
| lit_sources.append(make_lit_source_record(src_id, cand, cat_id, prof)) |
| primary_sub_pools[cat_id][sub_idx].append((cand, src_id, i)) |
|
|
| |
| reserve_sub_pools = defaultdict(lambda: defaultdict(list)) |
| next_src_num = total_published + 1 |
| for cat_id in range(1, 21): |
| prof = CATEGORY_PROFILES[cat_id] |
| for r_idx, rcand in enumerate(reserve_by_cat[cat_id]): |
| src_id = f"SRC-MED-{next_src_num:06d}" |
| next_src_num += 1 |
| sub_idx = match_subdomain(rcand, prof, r_idx) |
| lit_sources.append(make_lit_source_record(src_id, rcand, cat_id, prof)) |
| reserve_sub_pools[cat_id][sub_idx].append((rcand, src_id)) |
|
|
| all_sources = anchor_sources + nct_sources + lit_sources |
| print( |
| f"Registered {len(lit_sources):,} peer-reviewed papers + {len(anchor_sources):,} status anchors " |
| f"+ {len(nct_sources):,} ClinicalTrials.gov (NCT) studies = {len(all_sources):,} total sources." |
| ) |
|
|
| published_records = [] |
| seen_titles = set() |
| seen_statements = set() |
| lit_usage = Counter() |
| anchor_usage = Counter() |
| nct_usage = Counter() |
|
|
| |
| for prob_num in range(1, total_published + 1): |
| lit_usage[f"SRC-MED-{prob_num:06d}"] = 1 |
|
|
| |
| cat_all_lit_pools = defaultdict(list) |
| for cat_id in range(1, 21): |
| num_subs = len(CATEGORY_PROFILES[cat_id]["conditions"]) |
| for s_i in range(num_subs): |
| cat_all_lit_pools[cat_id].extend(reserve_sub_pools[cat_id][s_i]) |
| for s_i in range(num_subs): |
| cat_all_lit_pools[cat_id].extend([(c, sid) for (c, sid, _) in primary_sub_pools[cat_id][s_i]]) |
|
|
| print(f"Building {total_published:,} published records with semantic co-citation + live NCT trial linkage...") |
| for prob_num, (cat_id, i, is_gold) in enumerate(ordered_slots, start=1): |
| cand1 = selected_by_cat[cat_id][i] |
| prof = CATEGORY_PROFILES[cat_id] |
| sub_idx = match_subdomain(cand1, prof, i) |
| qtype = qtypes_by_cat[cat_id][i] |
| src_id1 = f"SRC-MED-{prob_num:06d}" |
| t1_tokens = cand1.get("_tokens") or set() |
|
|
| |
| sub_candidates = [ |
| (c, sid) |
| for (c, sid) in (reserve_sub_pools[cat_id][sub_idx] + [(c, s) for (c, s, _) in primary_sub_pools[cat_id][sub_idx]]) |
| if sid != src_id1 and lit_usage[sid] < 8 |
| ] |
| if len(sub_candidates) < 2: |
| sub_candidates = [ |
| (c, sid) |
| for (c, sid) in cat_all_lit_pools[cat_id] |
| if sid != src_id1 and lit_usage[sid] < 8 |
| ] |
| sub_candidates.sort(key=lambda x: (lit_usage[x[1]] // 2, -len(t1_tokens & (x[0].get("_tokens") or set())), lit_usage[x[1]])) |
| cand2, src_id2 = sub_candidates[0] |
| cand3, src_id3 = next((c, sid) for (c, sid) in sub_candidates[1:] if sid != src_id2) |
| lit_usage[src_id2] += 1 |
| lit_usage[src_id3] += 1 |
|
|
| |
| cat_anchors = anchors_by_cat[cat_id] |
| num_subs = len(prof["conditions"]) |
| cand_anchor_1 = cat_anchors[sub_idx % num_subs] |
| cand_anchor_2 = cat_anchors[(sub_idx % num_subs) + num_subs] |
| if anchor_usage[cand_anchor_1["source_id"]] <= anchor_usage[cand_anchor_2["source_id"]] and anchor_usage[cand_anchor_1["source_id"]] < 40: |
| anchor_src = cand_anchor_1 |
| elif anchor_usage[cand_anchor_2["source_id"]] < 40: |
| anchor_src = cand_anchor_2 |
| else: |
| anchor_src = min(cat_anchors, key=lambda a: anchor_usage[a["source_id"]]) |
| anchor_usage[anchor_src["source_id"]] += 1 |
|
|
| |
| nct_trial, nct_src_id = None, None |
| sub_ncts = [pair for pair in nct_by_subdomain.get((cat_id, sub_idx), []) if nct_usage[pair[1]] < 36] |
| if not sub_ncts: |
| sub_ncts = [pair for pair in nct_by_cat.get(cat_id, []) if nct_usage[pair[1]] < 36] |
| if sub_ncts: |
| sub_ncts.sort(key=lambda x: (-len(t1_tokens & (x[0].get("_tokens") or set())), nct_usage[x[1]])) |
| nct_trial, nct_src_id = sub_ncts[0] |
| nct_usage[nct_src_id] += 1 |
|
|
| rec = build_published_record( |
| prob_num=prob_num, |
| cat_id=cat_id, |
| sub_idx=sub_idx, |
| qtype=qtype, |
| is_gold=is_gold, |
| cand1=cand1, |
| cand2=cand2, |
| cand3=cand3, |
| src_id1=src_id1, |
| src_id2=src_id2, |
| src_id3=src_id3, |
| anchor_src=anchor_src, |
| nct_trial=nct_trial, |
| nct_src_id=nct_src_id, |
| ) |
|
|
| t_norm = rec["title"].lower().strip() |
| if t_norm in seen_titles: |
| rec["title"] = f"{rec['title']} [{cand1.get('doi') or cand1.get('pmid')}]" |
| t_norm = rec["title"].lower().strip() |
| seen_titles.add(t_norm) |
|
|
| s_norm = rec["statement"].lower().strip() |
| if s_norm in seen_statements: |
| rec["statement"] = rec["statement"][:-1] + f" (as highlighted in {cand1.get('doi') or cand1.get('pmid')})?" |
| s_norm = rec["statement"].lower().strip() |
| seen_statements.add(s_norm) |
|
|
| published_records.append(rec) |
|
|
| print(f"Building 100 archived/withheld calibration records (UMED-{total_published + 1:06d}..UMED-{total_published + 100:06d})...") |
| archived_specs = get_archived_and_audit_records(anchors_by_cat) |
| assert len(archived_specs) == 100 |
| archived_records = [ |
| build_archived_record(idx, spec, anchors_by_cat, total_published) |
| for idx, spec in enumerate(archived_specs, start=1) |
| ] |
|
|
| print("Building problem relations graph...") |
| relations_list = [] |
| rel_counter = 1 |
| by_cat_published = defaultdict(list) |
| for rec in published_records: |
| by_cat_published[rec["category_num"]].append(rec) |
|
|
| for cat_id, recs in by_cat_published.items(): |
| for idx in range(len(recs) - 1): |
| if idx % 2 == 0: |
| subj = recs[idx] |
| obj = recs[idx + 1] |
| pred = "informs" if idx % 4 == 0 else "overlaps_with" |
| rel_id = f"REL-MED-{rel_counter:06d}" |
| rel_counter += 1 |
| relations_list.append({ |
| "relation_id": rel_id, |
| "subject_id": subj["id"], |
| "predicate": pred, |
| "object_id": obj["id"], |
| "rationale": f"Both records address linked clinical uncertainties in {subj['medical_categories']['primary']}.", |
| "created_at": RELEASE_DATE, |
| "reviewer_id": f"REV-DOM-{cat_id:02d}", |
| }) |
| subj["relations"].append({"relation_type": pred, "target_id": obj["id"]}) |
|
|
| for arch in archived_records: |
| if arch.get("duplicate_of"): |
| rel_id = f"REL-MED-{rel_counter:06d}" |
| rel_counter += 1 |
| relations_list.append({ |
| "relation_id": rel_id, |
| "subject_id": arch["id"], |
| "predicate": "duplicate_of", |
| "object_id": arch["duplicate_of"], |
| "rationale": arch["archive_reason"], |
| "created_at": RELEASE_DATE, |
| "reviewer_id": "REV-AUDIT-01", |
| }) |
|
|
| all_records = published_records + archived_records |
| gold_records = [r for r in published_records if r["is_gold_seed"]] |
| assert len(published_records) == total_published |
| assert len(gold_records) == total_gold |
| assert len(all_records) == total_published + 100 |
|
|
| print("Building immutable status events...") |
| status_events = [] |
| for idx, rec in enumerate(all_records, start=1): |
| supp_ids = [s["source_id"] for s in rec["source_support"]] |
| need_adj = rec["publication_class"] in ("DISPUTED-OPEN", "PARTIAL-PROGRESS") or not rec["published"] |
| adj_id = f"ADJ-MED-{idx:06d}" if need_adj else None |
|
|
| status_events.append({ |
| "event_id": f"STAT-MED-{idx:06d}", |
| "problem_id": rec["id"], |
| "previous_status": None, |
| "new_status": rec["status"], |
| "effective_date": RELEASE_DATE, |
| "recorded_at": RELEASE_TIMESTAMP, |
| "reason": rec["status_assessment"]["rationale"], |
| "supporting_source_ids": supp_ids, |
| "reviewer_ids": rec["status_assessment"]["reviewer_ids"], |
| "adjudication_id": adj_id, |
| "confidence": rec["status_assessment"]["confidence"], |
| }) |
|
|
| print("Writing non-duplicated JSONL tables (data/gold/problems.jsonl, data/archived/problems.jsonl, relations/relations.jsonl, history/status_events.jsonl)...") |
| write_jsonl(ROOT / "data" / "gold" / "problems.jsonl", gold_records) |
| write_jsonl(ROOT / "data" / "archived" / "problems.jsonl", archived_records) |
| write_jsonl(ROOT / "relations" / "relations.jsonl", relations_list) |
| write_jsonl(ROOT / "history" / "status_events.jsonl", status_events) |
|
|
| print("Writing compact Parquet release tables (compression=zstd, zero redundant JSON string columns)...") |
| flat_rows = [] |
| for r in published_records: |
| flat_rows.append({ |
| "id": r["id"], |
| "problem_number": r["problem_number"], |
| "title": r["title"], |
| "statement": r["statement"], |
| "clean_statement": r["clean_statement"], |
| "plain_language_statement": r["plain_language_statement"], |
| "background": r["background"], |
| "question_type": r["question_type"], |
| "category_id": r["category_id"], |
| "category_num": r["category_num"], |
| "primary_category": r["medical_categories"]["primary"], |
| "set_id": r["set_id"], |
| "is_gold_seed": r["is_gold_seed"], |
| "condition_name": r["condition_or_health_state"]["name"], |
| "population_description": r["population"]["description"], |
| "question_framework_type": r["question_framework"]["type"], |
| "intervention_or_focus": r["intervention_exposure_test_or_predictor"]["name"], |
| "comparator_or_reference": r["comparator_or_reference"], |
| "time_horizon": r["time_horizon"], |
| "translational_stage": r["translational_stage"], |
| "evidence_state": r["evidence_state"], |
| "consensus_state": r["consensus_state"], |
| "difficulty_level_id": r["difficulty_level_id"], |
| "research_difficulty": r["research_difficulty"], |
| "status": r["status"], |
| "publication_class": r["publication_class"], |
| "proposed_by": r["proposed_by"], |
| "proposed_year": r["proposed_year"], |
| "source_url": r["source_url"], |
| "source_citation": r["source_citation"], |
| "active_trial": r["status_assessment"]["active_research"][0] if r["status_assessment"]["active_research"] else "", |
| "literature_papers_count": 3, |
| "total_sources_per_problem": len(r["source_support"]), |
| "published": r["published"], |
| }) |
| table = pa.Table.from_pylist(flat_rows) |
| (ROOT / "releases").mkdir(parents=True, exist_ok=True) |
| pq.write_table(table, ROOT / "releases" / "unsolvedmed_current.parquet", compression="zstd") |
| if args.version == "1.5": |
| pq.write_table(table, ROOT / "releases" / "unsolvedmed_v1_5_current.parquet", compression="zstd") |
|
|
| cat_counts = Counter(r["category_id"] for r in published_records) |
| gold_cat_counts = Counter(r["category_id"] for r in gold_records) |
| qtype_counts = Counter(r["question_type"] for r in published_records) |
| pub_class_counts = Counter(r["publication_class"] for r in all_records) |
| current_pub_class_counts = Counter(r["publication_class"] for r in published_records) |
| status_counts = Counter(r["status"] for r in all_records) |
| diff_counts = Counter(r["research_difficulty"] for r in published_records) |
| stage_counts = Counter(r["translational_stage"] for r in published_records) |
| ev_state_counts = Counter(r["evidence_state"] for r in published_records) |
| src_tier_counts = Counter(s["source_tier"] for s in all_sources) |
| src_type_counts = Counter(s["source_type"] for s in all_sources) |
|
|
| statistics_obj = { |
| "dataset": "ulamai/UnsolvedMedical", |
| "version": semver, |
| "release_date": RELEASE_DATE, |
| "total_records": len(all_records), |
| "published_open_problems": len(published_records), |
| "gold_seed_problems": len(gold_records), |
| "archived_and_withheld_records": len(archived_records), |
| "total_registered_sources": len(all_sources), |
| "peer_reviewed_literature_sources": len(lit_sources), |
| "clinical_trials_gov_sources": len(nct_sources), |
| "institutional_status_anchor_sources": len(anchor_sources), |
| "literature_papers_per_published_problem": 3, |
| "total_sources_per_published_problem": 5 if nct_sources else 4, |
| "clinical_subdomains_count": 400, |
| "question_template_structures": 100, |
| "total_relations": len(relations_list), |
| "total_status_events": len(status_events), |
| "by_primary_category": dict(cat_counts), |
| "gold_seed_by_primary_category": dict(gold_cat_counts), |
| "by_question_type": dict(qtype_counts), |
| "by_publication_class_all": dict(pub_class_counts), |
| "by_publication_class_current": dict(current_pub_class_counts), |
| "by_status_all": dict(status_counts), |
| "by_research_difficulty": dict(diff_counts), |
| "by_translational_stage": dict(stage_counts), |
| "by_evidence_state": dict(ev_state_counts), |
| "sources_by_tier": dict(src_tier_counts), |
| "sources_by_type": dict(src_type_counts), |
| } |
|
|
| print("Writing compact root UnsolvedMath-compatible JSON files...") |
| write_json_array(ROOT / "problems.json", published_records) |
| write_json_array(ROOT / "source_registry.json", all_sources) |
| (ROOT / "statistics.json").write_text(json.dumps(statistics_obj, indent=2, ensure_ascii=False) + "\n", encoding="utf-8") |
|
|
| |
| categories_obj = json.loads((ROOT / "categories.json").read_text(encoding="utf-8")) |
| difficulty_obj = json.loads((ROOT / "difficulty_levels.json").read_text(encoding="utf-8")) |
| sets_obj = json.loads((ROOT / "sets.json").read_text(encoding="utf-8")) |
| dataset_summary_problems = [ |
| { |
| "id": r["id"], |
| "problem_number": r["problem_number"], |
| "title": r["title"], |
| "statement": r["statement"], |
| "clean_statement": r["clean_statement"], |
| "question_type": r["question_type"], |
| "category_id": r["category_id"], |
| "set_id": r["set_id"], |
| "is_gold_seed": r["is_gold_seed"], |
| "difficulty_level_id": r["difficulty_level_id"], |
| "research_difficulty": r["research_difficulty"], |
| "status": r["status"], |
| "publication_class": r["publication_class"], |
| "proposed_by": r["proposed_by"], |
| "proposed_year": r["proposed_year"], |
| "source_url": r["source_url"], |
| "published": r["published"], |
| } |
| for r in all_records |
| ] |
| dataset_bundle = { |
| "metadata": statistics_obj, |
| "categories": categories_obj, |
| "difficulty_levels": difficulty_obj, |
| "sets": sets_obj, |
| "problems": dataset_summary_problems, |
| } |
| (ROOT / "dataset.json").write_text( |
| json.dumps(dataset_bundle, ensure_ascii=False, separators=(",", ":")) + "\n", encoding="utf-8" |
| ) |
|
|
| research_results = [ |
| { |
| "id": r["id"], |
| "problem_number": r["problem_number"], |
| "status": r["status"], |
| "publication_class": r["publication_class"], |
| "published": r["published"], |
| "confidence": r["status_assessment"]["confidence"], |
| "rationale": r["status_assessment"]["rationale"], |
| "literature_summary": r["literature_assessment"]["summary"], |
| "active_research": r["status_assessment"]["active_research"], |
| "source_ids": [s["source_id"] for s in r["source_support"]], |
| "solution_reference": r["solution_reference"], |
| } |
| for r in all_records |
| ] |
| write_json_array(ROOT / "research_results.json", research_results) |
|
|
| statement_audits_root = [ |
| { |
| "id": r["id"], |
| "problem_number": r["problem_number"], |
| "published": r["published"], |
| "statement_audit": r["statement_audit"], |
| "record_text_license": r["rights"]["record_text_license"], |
| "risk_level": r["responsible_release"]["risk_level"], |
| "rejection_code": r["rejection_code"], |
| } |
| for r in all_records |
| ] |
| write_json_array(ROOT / "statement_audit.json", statement_audits_root) |
|
|
| release_files = [ |
| "problems.json", |
| "dataset.json", |
| "research_results.json", |
| "source_registry.json", |
| "statement_audit.json", |
| "statistics.json", |
| "releases/unsolvedmed_current.parquet", |
| "data/gold/problems.jsonl", |
| "data/archived/problems.jsonl", |
| "relations/relations.jsonl", |
| "history/status_events.jsonl", |
| ] |
| if (ROOT / "releases" / "unsolvedmed_v1_5_current.parquet").exists(): |
| release_files.append("releases/unsolvedmed_v1_5_current.parquet") |
|
|
| release_manifest = { |
| "dataset": "ulamai/UnsolvedMedical", |
| "version": semver, |
| "release_date": RELEASE_DATE, |
| "counts": { |
| "published_current": len(published_records), |
| "gold_seed": len(gold_records), |
| "archived_and_withheld": len(archived_records), |
| "all_records": len(all_records), |
| "sources": len(all_sources), |
| "peer_reviewed_literature_sources": len(lit_sources), |
| "clinical_trials_gov_sources": len(nct_sources), |
| "status_anchor_sources": len(anchor_sources), |
| "literature_papers_per_problem": 3, |
| "total_sources_per_problem": 5 if nct_sources else 4, |
| "clinical_subdomains": 400, |
| "relations": len(relations_list), |
| "status_events": len(status_events), |
| }, |
| "files": release_files, |
| } |
| (ROOT / "releases" / "release_manifest.json").write_text( |
| json.dumps(release_manifest, indent=2, ensure_ascii=False) + "\n", encoding="utf-8" |
| ) |
| (ROOT / "releases" / f"release_manifest_v{args.version}.json").write_text( |
| json.dumps(release_manifest, indent=2, ensure_ascii=False) + "\n", encoding="utf-8" |
| ) |
|
|
| print(f"Successfully generated clean, compact UnsolvedMed v{semver} ({total_published:,} published problems)!") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
