id,prompt,gold_json,reference_notes
EPRO-001,"Quad ePRO integrity analysis. Neurology trial. Daily symptom diary completion drops from 82% to 49% at two sites after app update. Reminder notifications disabled by default on new phones. Site staff began coaching participants to fill diaries during clinic visits. Governance: interim patient-reported endpoint readout in 2 weeks. Determine ePRO integrity risk and action. Return strict JSON only.","{""epro_integrity_risk"":true,""risk_type"":""compliance_drop_with_notification_and_coaching_bias"",""driver_nodes"":[""completion drop cluster"",""app update and notification default off"",""site coaching during visits"",""imminent interim PRO readout""],""recommended_action"":""restore_notifications_and_audit_entries"",""action_detail"":[""re-enable reminders by configuration"",""flag visit-day backfilled entries"",""retrain sites to avoid coaching bias"",""inform interim analysis team""],""rationale"":""Completion drop aligned with app update and disabled reminders suggests device-driven missingness. Coaching to fill during visits creates backfill bias. Interim timing requires restoring reminders and auditing entry patterns before analysis.","confidence"":0.9}","App update + coaching."
EPRO-002,"Quad ePRO integrity analysis. Oncology trial. Pain scores improve sharply at high-enrolling sites. Device swap to tablets occurred due to phone compatibility issues. Data sync delay averages 3 days. Governance: submission narrative relies on pain endpoint. Determine risk and action. JSON only.","{""epro_integrity_risk"":true,""risk_type"":""device_swap_with_sync_delay_and_endpoint_dependence"",""driver_nodes"":[""pain improvement cluster"",""device swap to tablets"",""sync delay"",""submission dependence on endpoint""],""recommended_action"":""device_and_sync_reconciliation"",""action_detail"":[""compare scores pre/post device swap"",""resolve sync pipeline delay"",""audit missing days and backfills"",""document impact for submission""],""rationale"":""Endpoint shift coinciding with device swap and sync delay suggests measurement and timing artifacts rather than true improvement. Submission reliance increases governance risk. Reconciliation and documentation required.","confidence"":0.88}","Device artifact risk."
EPRO-003,"Quad ePRO integrity analysis. Cardiology trial. Completion rates stable overall but missingness spikes on weekends at one region. Reminder schedule configured only Mon–Fri. Staffing reduced for weekend support. Governance: DSMB review next week includes PRO tolerability. Determine risk and action. JSON only.","{""epro_integrity_risk"":true,""risk_type"":""weekend_missingness_due_to_reminder_configuration"",""driver_nodes"":[""weekend missingness spike"",""weekday-only reminders"",""reduced weekend support"",""DSMB timing""],""recommended_action"":""weekend_reminder_and_support_fix"",""action_detail"":[""extend reminders to weekends"",""add weekend support coverage"",""audit regional missingness"",""brief DSMB on mitigation""],""rationale"":""Weekend missingness linked to reminder configuration and reduced support suggests systematic bias in tolerability reporting. DSMB timing requires correction and disclosure before review.","confidence"":0.87}","Weekend config."
EPRO-004,"Quad ePRO integrity analysis. Pulmonary trial. PRO entries surge the day before site visits. App allows offline entry with later upload. Site staff remind participants during scheduling calls. Governance: interim exacerbation analysis uses PRO triggers. Determine risk and action. JSON only.","{""epro_integrity_risk"":true,""risk_type"":""visit_proximity_backfill_bias"",""driver_nodes"":[""pre-visit entry surge"",""offline entry then upload"",""visit-call reminders"",""interim analysis tie to PRO triggers""],""recommended_action"":""backfill_detection_and_policy_reset"",""action_detail"":[""detect timestamp/upload mismatches"",""limit retroactive entry windows"",""standardize reminder scripts"",""inform interim analysis team""],""rationale"":""Entry surges near visits with offline upload suggests backfill patterns. Reminder timing may shape reporting and trigger-based analyses. Interim tie requires backfill detection and policy reset before analysis.","confidence"":0.89}","Backfill node."
EPRO-005,"Quad ePRO integrity analysis. Metabolic trial. Completion drops after OS update. Battery optimization kills background reminders. Sites report increased manual calls to participants. Governance: compliance audit scheduled. Determine risk and action. JSON only.","{""epro_integrity_risk"":true,""risk_type"":""os_update_with_battery_reminder_failure"",""driver_nodes"":[""completion drop post OS update"",""battery optimization disables reminders"",""manual calling by sites"",""audit scheduled""],""recommended_action"":""os_configuration_and_call_audit"",""action_detail"":[""publish device settings guide"",""verify reminders deliver"",""audit manual call influence"",""prepare audit documentation""],""rationale"":""OS-level reminder failure can drive missingness, while manual calls introduce site influence on reporting. Audit timing requires configuration fix, influence audit, and documentation.","confidence"":0.88}","OS reminder failure."
EPRO-006,"Quad ePRO integrity analysis. Vaccine trial. Symptom diary completion high but identical text appears across participants at one site. Copy-paste suspected. Staff workload high. Governance: regulator questions data integrity. Determine risk and action. JSON only.","{""epro_integrity_risk"":true,""risk_type"":""templated_entries_under_workload_pressure"",""driver_nodes"":[""duplicate diary text cluster"",""copy-paste suspicion"",""high staff workload"",""regulator scrutiny""],""recommended_action"":""duplicate_detection_and_site_intervention"",""action_detail"":[""run duplicate text detection"",""interview site staff and retrain"",""verify participant-level provenance"",""prepare regulator response""],""rationale"":""Duplicated entries suggest templating or staff entry, compromising participant-reported authenticity. Workload pressure increases risk. Regulator scrutiny requires detection, intervention, and provenance verification.","confidence"":0.9}","Duplicate text."