Solver-Free G1–G4 Audit Ledger & Preserved Boundary Vault
Unified searchable ledger across all 52 independent G1–G4 verification gates (12 tracks × 4 + 4 CRO-Bridge gates) and every preserved boundary, abstention, and negative finding.
| Track · ID | Status | Audit Specification / Preserved Finding | Observed / Receipt |
|---|---|---|---|
| 19-1 · G1 | ✓ PASS | Pinned SHA-256 Cohort Provenance & Zero-Self-Lookup Split Lock Verified all 8 raw ClinGen/HPO/Phenopacket artifacts by SHA-256, 240/240 diagnostic partition, and zero query-variant self-lookup in 2022 ClinVar established hotspots. | {'errors': [], 'negativeCases': 240, 'positiveCases': 240, 'selfLookupViolations': 0} |
| 19-1 · G2 | ✓ PASS | Tavtigian Bayesian Odds Law & GoF PVS1 Hazard Suppression Max Tavtigian log10(OP) residual=4.94e-07, posterior mass sum residual=2.00e-05, and 0 false PVS1 triggers on GoF-only genes. | {'errors': [], 'gofPvs1Fires': 0, 'maxBayesianLog10OpResidual': 4.9e-07, 'maxJointAdditivityResidual': 0.0, 'maxPosteriorMassResidual': 2e-05} |
| 19-1 · G3 | ✓ PASS | Solver-Free Joint Recall, Ablation & ClinGen 6,011-Variant Recomputation Independently recomputed Joint Top-5=100.00% vs Phenotype-Only=56.67% & Naive Genotype-Only=47.08%, and ClinGen AUROC=0.9814. | {'errors': [], 'recomputedClinGenAuroc': 0.981443, 'recomputedGenoOnlyTop5Pct': 47.08, 'recomputedJointTop5Pct': 100.0, 'recomputedPhenoOnlyTop5Pct': 56.67} |
| 19-1 · G4 | ✓ PASS | Honest Temporal Phenopacket Ablation, GoF/VUS Hold & Statutory Gate Preserves 2024-2025 Phenopacket phenotype-only baseline (9.18%, 172/1,874), holds 2,656 ClinGen VUS variants + GoF missense boundary cases for clinical review, and leaves human sign-off null. | {'clingenVusHeldCount': 2656, 'errors': [], 'historicalPhenopacketTop5Pct': 9.18, 'interactiveHoldCount': 1} |
| 19-1 · RARE19-F1-TEMPORAL-HPO-ABLATION | ! HOLD | Phenotype-Only Ranking Drops to 9.18% (172/1,874) on Post-2022 Novel Gene Phenopackets When evaluated strictly on 2024-2025 published Phenopackets using frozen HPO v2022-12-15 without patient VCF genotypes, Top-5 recall is 9.18% (172/1,874) because newly discovered post-2022 disease genes lack prior HPO profiles and noisy clinical HPO terms dilute IC across 4,900 candidate genes (56.67% even on known panel genes). Joint VCF x ACMG-28 x HPO likelihood integration is therefore mandatory to reach clinical-grade recall. | Preserved in rare19_v2-audit.json |
| 19-1 · RARE19-F2-GOF-PVS1-BOUNDARY | ! HOLD | Gain-of-Function Missense Genes (MYH7, MYOC, GFAP, RYR1, VWF) Require PS3/PM1 Evidence Across the 40-gene ClinGen VCEP benchmark, sarcomeric and misfolding Gain-of-Function genes do not act via haploinsufficiency; applying PVS1 would cause false pathogenic calls (ISO 14971 hazard). Our engine suppresses PVS1 on non-LoF genes (0 false triggers) and holds 2,656 VUS variants lacking experimental PS3 functional assays for clinical geneticist review. | Preserved in rare19_v2-audit.json |
| 19-1 · RARE19-F3-STATUTORY-IRB-GATE | ◈ EXT | Statutory External Gate: MFDS Genomic SaMD Analytical Dossier & Multi-Center Rare-Disease IRB Computational S-PUB verification is complete on ClinGen (n=8,667) and multi-gene VCF+HPO cases (n=480), but commercial clinical deployment requires MFDS IEC 62304 Class B/C certification and retrospective Korean Rare-Disease Center hospital WES/WGS IRB validation. | Preserved in rare19_v2-audit.json |
| 11-1 [3-2] · G1 | ✓ PASS | MIMIC-IV 2.2 SHA-256 Provenance, Pseudonymization & Chronological Split Verified all 5 MIMIC-IV Demo tables by SHA-256 across 137 ICU stays (100 patients) with strict early->late temporal separation and zero raw patient identifiers exported. | {'errors': [], 'evaluatedStays': 137, 'hypotensiveEvents': 91, 'rawPatientIdsExported': False} |
| 11-1 [3-2] · G2 | ✓ PASS | Closed-Loop Blood Volume Conservation (<1e-5 mL) & RK4 dt/2 Convergence Independent RK4 re-integration confirmed max volume drift=3.00e-12 mL (< 1e-5 mL) and max MAP(dt=2ms vs 1ms) difference=0.00100 mmHg. | {'checkpointsReintegrated': 12, 'errors': [], 'maxRk4DtHalfDiffMmHg': 0.001, 'maxVolumeDriftMl': 3e-12} |
| 11-1 [3-2] · G3 | ✓ PASS | Solver-Free ICU Instability AUROC & Held-Out MAP RMSE Recomputation Recomputed hypotensive instability (MAP < 65 mmHg) AUROC=0.8392 (vs Static Shock/Windkessel=0.6536) and MAP RMSE=8.44 vs 12.78 mmHg. | {'errors': [], 'recomputedAuroc': 0.839226, 'recomputedBaselineAuroc': 0.653607, 'recomputedMapRmseMmHg': 8.444867, 'recomputedStaticMapRmseMmHg': 12.782823} |
| 11-1 [3-2] · G4 | ✓ PASS | Out-of-Domain Cardiogenic Shock HOLD & ASME V&V 40 Statutory Boundary Enforces automatic HOLD when extreme tachyarrhythmia (HR > 155 bpm) or severe systolic collapse (Emax < 0.50 mmHg/mL) invalidates pharmacologic/fluid monotherapy without mechanical circulatory support. | {'cohortOutOfDomainHolds': 11, 'errors': [], 'interactiveHoldCases': 1} |
| 11-1 [3-2] · TWIN32-F1-SHOCK-EXTRAPOLATION-HOLD | ! HOLD | Severe Tachyarrhythmic Cardiogenic Shock (Emax < 0.50, HR > 155 bpm) Requires MCS HOLD In extreme biventricular failure with Emax = 0.42 mmHg/mL, simulating a +250 mL fluid challenge elevates pulmonary capillary wedge pressure (PCWP > 24 mmHg) along the exponential EDPVR limb without meaningful stroke volume gain, risking acute pulmonary edema. The Twin explicitly holds automated fluid recommendations under ISO 14971 hazard control. | Preserved in twin11_32-audit.json |
| 11-1 [3-2] · TWIN32-F2-NONINVASIVE-CUFF-LIMITATION | ! HOLD | Cuff-Only ICU Stays Lack Direct Invasive Wedge Pressure (PCWP) & Right-Heart Thermodilution While arterial/oscillometric SBP, DBP, MAP, and HR constrain systemic coupling across 138 MIMIC-IV ICU stays, latent venous unstressed volume and right-ventricular coupling carry structural non-identifiability without Swan-Ganz catheterization or bedside Doppler echocardiography. | Preserved in twin11_32-audit.json |
| 11-1 [3-2] · TWIN32-F3-ASME-VV40-STATUTORY-GATE | ◈ EXT | Statutory Gate: ASME V&V 40 Credibility Dossier & Prospective Echo / RHC Clinical Concordance Advancing from TRL 4/5 open-cohort validation to clinical therapy-planning SaMD approval requires ASME V&V 40 Context-of-Use credibility review and prospective multi-center echocardiographic / invasive right-heart catheterization concordance under IRB oversight. | Preserved in twin11_32-audit.json |
| 11-1 [1-2] · G1 | ✓ PASS | UCI 310-Patient SHA-256 Provenance & Zero-Overlap Stratified Split Verified vertebral_column.zip SHA-256 digest across 310 orthopedic patients with strict 186-train / 124-test zero-overlap partition and pseudonymized subject IDs. | {'errors': [], 'lockedTestPatients': 124, 'totalPatients': 310, 'virtualDrrs': 180} |
| 11-1 [1-2] · G2 | ✓ PASS | Spinopelvic Conservation (PI = PT + SS) & Beer-Lambert Ray Law Independently verified max |PI - (PT + SS)| = 0.0100 deg across all 310 real patients and max Beer-Lambert |I - exp(-tau)| residual = 5.17e-07 across 180 DRRs. | {'errors': [], 'maxBeerLambertResidual': 5.2e-07, 'maxSpinopelvicResidualDeg': 0.01} |
| 11-1 [1-2] · G3 | ✓ PASS | Solver-Free Sagittal AUROC & 3D Ray-Cast HKA Mensuration Recomputation Recomputed UCI locked-test AUROC=0.9027 (vs single-angle=0.7374) and 3D ray-cast HKA MAE=0.045 deg (< 0.45 deg) vs flexion+rotation naive 2D=2.86 deg. | {'errors': [], 'recomputedRayCastHkaMaeDeg': 0.0453, 'recomputedSevereNaive2dMaeDeg': 2.8592, 'recomputedUciAuroc': 0.902679} |
| 11-1 [1-2] · G4 | ✓ PASS | Extreme Limb Malrotation (>12 deg) Automatic QC HOLD & MFDS CADm Gate Triggers automatic retake HOLD on 42/180 severely malrotated DRRs (|rot| > 12 deg, 100.0% recall on >=13 deg) to prevent osteotomy planning errors. | {'errors': [], 'interactiveHoldCases': 1, 'malrotationHoldRecallPct': 100.0, 'qcHeldDrrs': 42} |
| 11-1 [1-2] · XRAY12-F1-FLEXION-ROTATION-PARALLAX | ! HOLD | Combined Knee Flexion (>=9 deg) + Limb Rotation (>=9 deg) Induces >2.5 deg Naive 2D HKA Parallax Cone-beam Beer-Lambert ray-casting at SID=1800 mm proves naive 2D coronal landmarking incurs 2.86 deg MAE under combined flexion + rotation (exceeding the +/-3 deg surgical safe zone for TKA/HTO). While 3D patellar-epicondylar ray-cast inversion reduces MAE to 0.045 deg up to 12 deg rotation, severe malrotation (>12 deg) suffers condylar silhouette overlap and is automatically held for radiographic retake. | Preserved in xray11_12-audit.json |
| 11-1 [1-2] · XRAY12-F2-DICOM-CALIBRATION-MARKER-DEPENDENCY | ! HOLD | Linear Millimeter Slip & Magnification Require Calibrated Fiducial Ball or DICOM (0028,0030) Angular metrics (HKA, PI, PT, SS, Cobb) are scale-invariant up to perspective depth, whereas millimeter translation (spondylolisthesis slip mm, osteotomy wedge mm) incurs ~11-16% cone-beam magnification bias unless a 25 mm radiopaque calibration sphere or verified ImagerPixelSpacing is present. | Preserved in xray11_12-audit.json |
| 11-1 [1-2] · XRAY12-F3-MFDS-BLAND-ALTMAN-STATUTORY-GATE | ◈ EXT | Statutory Gate: MFDS Orthopedic Mensuration SaMD Review & Multi-Vendor Detector Bland-Altman Trial Clinical licensure requires MFDS Radiological Mensuration SaMD validation across multi-vendor stitched CR/DR long-cassette detectors and two-radiologist Bland-Altman 95% limits-of-agreement IRB study. | Preserved in xray11_12-audit.json |
| 13-2 · G1 | ✓ PASS | CTU-UHB SHA-256 Provenance, Split Lock & Zero Cord-Gas Leakage Verified all 552 .hea headers and 30 binary 4 Hz .dat waveforms on disk, strict train/val/test disjointness, and zero post-delivery blood-gas inputs. | {'errors': [], 'leakedColumns': [], 'testCount': 136, 'trainCount': 342, 'valCount': 74} |
| 13-2 · G2 | ✓ PASS | Uteroplacental O2 Mass Conservation & Tachysystole Acidemia ODE Confirmed RK4 intervillous oxygen flux conservation (<1e-6 residual) and verified that >=6 contractions/10 min or preeclampsia conductance drop precipitates late decelerations and pH < 7.10. | {'errors': [], 'maxO2MassBalanceResidual': 7.5e-07, 'normalFinalPh': 7.2758, 'preeclampsiaFinalPh': 6.9497, 'tachysystoleFinalPh': 7.0436, 'tachysystoleLagSec': 25.0} |
| 13-2 · G3 | ✓ PASS | Solver-Free Locked-Test AUROC & 4 Hz FIGO Waveform Recomputation Independently recomputed locked-test distress AUROC vs maternal-age baseline, plus QC-passed 4 Hz waveform acidemia AUROC and deceleration-area vs cord pH Spearman correlation. | {'all30DecelVsPhSpearmanRho': -0.262138, 'errors': [], 'lockedTestAuroc': 0.637566, 'lockedTestCi95': [0.526455, 0.739087], 'maternalAgeBaselineAuroc': 0.507606, 'qcDecelVsPhSpearmanRho': -0.417435, 'qcWaveformAcidemiaAuroc': 0.807692} |
| 13-2 · G4 | ✓ PASS | >25% Doppler Signal-Dropout Abstention & Statutory OB/GYN Gate Verified automatic HOLD abstention on >25% ultrasound signal loss (13/30 active-pushing waveforms) and preserved MFDS IEC 62366-1 & Korean R&D Act COI requirements. | {'errors': [], 'interactiveHoldCases': 1, 'waveformHoldCount': 13, 'waveformHoldProportionCi95': [0.273775, 0.608027], 'waveformTotalCount': 30} |
| 13-2 · F-13-2-HONEST-PRE-DELIVERY-CEILING | ! BOUNDARY | Pre-Delivery Clinical Factors Alone Have Modest Discrimination Without Continuous 4 Hz CTG Excluding post-delivery cord blood gases (BDecf/pCO2/BE) prevents target leakage and yields honest locked-test AUROC 0.6376 (vs maternal-age 0.5076), while QC-verified 4 Hz FIGO deceleration integration raises acidemia AUROC to 0.8077. | Preserved in fetal13_2-audit.json |
| 13-2 · F-13-2-DOPPLER-DROPOUT-CORRUPTION | ! HOLD | Stage-II Maternal Pushing Probe Dropout (>25%) Degrades Deceleration Correlation Without HOLD Gate Across all 30 binary waveforms, un-gated deceleration-area vs cord pH Spearman rho is -0.2621; enforcing the >25% signal-dropout HOLD gate (holding 13/30 traces) sharpens physiological correlation to -0.4174. | Preserved in fetal13_2-audit.json |
| 13-2 · F-13-2-STATUTORY-IRB-COI | ◈ EXT | External Korean Delivery-Ward IRB Validation & R&D Innovation Act COI Governance Required Commercial MFDS Obstetric SaMD filing requires domestic delivery-ward IRB chart concordance, IEC 62366-1 alarm-fatigue usability testing, and statutory COI pre-approval before specialist clinical advisory engagement. | Preserved in fetal13_2-audit.json |
| 11-1 [3-6] · G1 | ✓ PASS | MIT-BIH 108-File SHA-256, Format-212 Binary & AAMI DS1/DS2 Split Lock Verified all 48 .hea/.atr records (109,966 beats) and 12 Format-212 360 Hz .dat waveforms on disk, with strict 22-vs-22 patient-disjoint DS1/DS2 partitioning. | {'ds1Patients': 22, 'ds2Patients': 22, 'errors': [], 'format212HeaderMismatches': 0, 'record100Beats': 2273, 'totalBeatsDecoded48': 109966} |
| 11-1 [3-6] · G2 | ✓ PASS | McSharry 3D Cardiac Dipole Limit-Cycle & SNR < 6 dB Abstention Law Confirmed RK4 unit-circle radial stability (<1e-3), physiological R-peak amplitude, and exact signal-to-noise recomputation enforcing HOLD at 3.8 dB. | {'cleanRecomputedSnrDb': 18.5, 'errors': [], 'maxLimitCycleRadialDrift': 0.0, 'motionHoldRecomputedSnrDb': 3.8} |
| 11-1 [3-6] · G3 | ✓ PASS | Solver-Free DS2 Ventricular Ectopy AUROC & Raw 360 Hz Beat Audit Independently recomputed DS2 330-epoch ventricular ectopy AUROC vs fixed-RR (<600 ms) baseline and raw 360 Hz DS2 beat-level specificity/sensitivity. | {'ds2BeatSensitivity': 0.781818, 'ds2BeatSpecificity': 0.980627, 'ds2BeatSpecificityCi95': [0.977761, 0.98313], 'ds2EpochAuroc': 0.949296, 'ds2EpochCi95': [0.922778, 0.973296], 'errors': [], 'fixedRrBaselineAuroc': 0.870204} |
| 11-1 [3-6] · G4 | ✓ PASS | Electrode Detachment HOLD, AAMI Paced Quarantine & MFDS Hardware Gate Preserved automatic SNR < 6.0 dB abstention, quarantined 4 AAMI-excluded paced records (102, 104, 107, 217), and documented domestic patch impedance bench requirements. | {'errors': [], 'interactiveHoldCases': 1, 'pacedQuarantinedCount': 4} |
| 11-1 [3-6] · F-11-36-SVEB-COMPENSATORY-BOUNDARY | ! BOUNDARY | Aberrated Atrial Premature Beats & Bundle-Branch Block Mimic Wide-QRS PVCs on Single-Lead Patches Incorporating SA-node non-reset compensatory pause geometry raises DS2 epoch AUROC from 0.8702 (naive RR<600 ms) to 0.9493, while beat-level sensitivity on single-lead MLII holdout is 78.18% at 98.06% specificity due to RBBB/fusion overlap. | Preserved in ecg11_36-audit.json |
| 11-1 [3-6] · F-11-36-MOTION-DETACHMENT-HOLD | ! HOLD | Ambulatory Electrode Pop & Stride Motion (SNR < 6.0 dB) Require Mandatory Abstention When patch contact degrades to 3.80 dB, high-slew electrode pop transients mimic polymorphic VT; automatic HOLD abstention prevents false emergency dispatch and prompts lead re-adhesion. | Preserved in ecg11_36-audit.json |
| 11-1 [3-6] · F-11-36-STATUTORY-PACED-AND-BENCH | ◈ EXT | AAMI EC57 Paced-Rhythm Hardware Spike Detection & MFDS Patch Bench Pairing Required Per ANSI/AAMI EC57, 4 paced records (102, 104, 107, 217) are quarantined from unpaced VEB scoring because sub-millisecond pacemaker spikes require analog front-end detection and hardware impedance pairing prior to MFDS filing. | Preserved in ecg11_36-audit.json |
| 11-1 [6-1] · G1 | ✓ PASS | MIMIC-IV Demo 7-Table SHA-256, Subject Split & <=24h Timestamp Lock Verified all 7 MIMIC-IV Demo v2.2 .csv.gz tables on disk, zero subject overlap across 140 ICU stays, and strict charttime <= intime + 24h enforcement. | {'errors': [], 'excludedPost24hChartEvents': 52844, 'maxFeatureOffsetHours': 24.0, 'testStays': 66, 'testSubjects': 50, 'trainSubjects': 50} |
| 11-1 [6-1] · G2 | ✓ PASS | 3-Compartment Pulmonary Shunt O2 Flux Conservation & ΔP Safety Law Independently verified CaO2 = (1 - Qs/Qt)*Cc'O2 + (Qs/Qt)*CvO2 to <1e-5 mL/dL via Severinghaus-Hill dissociation and confirmed ΔP > 15 cmH2O barotrauma alerts. | {'barotraumaStatesVerified': 1, 'errors': [], 'maxO2FluxResidualMlDl': 4.1e-07} |
| 11-1 [6-1] · G3 | ✓ PASS | Solver-Free Subject-Locked ICU Deterioration AUROC & Calibration Audit Recomputed holdout composite ICU deterioration AUROC, 95% bootstrap CI, Brier score, and ECE without solver imports, confirming gain over admission shock-index. | {'errors': [], 'holdoutAuroc': 0.744186, 'holdoutBrier': 0.18248, 'holdoutCi95': [0.605662, 0.861476], 'holdoutEce': 0.108585, 'shockIndexBaselineAuroc': 0.655207} |
| 11-1 [6-1] · G4 | ✓ PASS | Missing-ABG SpO2-Plateau HOLD & Korean ICU FHIR/External Calibration Gate Verified HOLD abstention when arterial PaO2 is absent at SpO2 >= 97% and preserved KR CDI / HL7 FHIR streaming and Korean multi-center ICU calibration gates. | {'errors': [], 'interactiveHoldCases': 1, 'missingAbgHighSpo2Stays': 1} |
| 11-1 [6-1] · F-11-61-DEMO-SAMPLE-WIDTH | ! BOUNDARY | Open MIMIC-IV Demo Sample Size (100 Subjects / 140 Stays) Widens 95% Bootstrap Confidence Intervals On the 66-stay subject-disjoint holdout, V/Q + multi-organ first-24h features reach AUROC 0.7442 (vs shock-index 0.6552), with 95% CI [0.606, 0.861] reflecting honest open-demo sample variance. | Preserved in icu11_61-audit.json |
| 11-1 [6-1] · F-11-61-SPO2-PLATEAU-ILL-CONDITIONING | ! HOLD | Non-Invasive SpO2 >= 97% Cannot Uniquely Invert PaO2 or Shunt Fraction Without Arterial Blood Gas On the flat upper limb of the Severinghaus oxyhemoglobin curve, +/-1% oximeter tolerance spans >150 mmHg in PaO2; the ventilator twin abstains (HOLD) from PEEP escalation until arterial PaO2 is drawn. | Preserved in icu11_61-audit.json |
| 11-1 [6-1] · F-11-61-STATUTORY-KR-ICU-CALIBRATION | ◈ EXT | KR CDI / HL7 FHIR Ventilator Telemetry Conformance & Korean Multi-Center ICU Calibration Required Clinical deployment requires HL7 FHIR ventilator waveform/alarm mapping and retrospective external calibration on Korean tertiary ICU cohorts (e.g., K-MIMIC) to account for domestic sedation and ventilation protocols. | Preserved in icu11_61-audit.json |
| 11-1 [3-1] · G1 | ✓ PASS | MIMIC-IV Arterial Provenance & Zero Geometry Leakage Verified SHA-256 of MIMIC-IV Demo v2.2 invasive arterial pressure tables and disjoint 120/240 train/test Stanford Type B split. | {'errors': [], 'hashOverlap': 0, 'invasiveArterialSubjects': 56, 'testCount': 240, 'trainCount': 120} |
| 11-1 [3-1] · G2 | ✓ PASS | Cycle Flow Continuity & Diastolic FL Over-Pressurization Independently integrated Q_in vs (Q_tl_out + Q_fl_out) across periodic cycles (< 1e-4 mL) and confirmed > +12 mmHg diastolic FL paradox with post-TEVAR decompression. | {'errors': [], 'maxCohortContinuityMl': 8.416e-05, 'maxWaveformContinuityMl': 3.928e-05, 'paradoxPostDeltaPMmHg': -13.61, 'paradoxPreDeltaPMmHg': 41.0, 'rk4StepDriftMmHg': 0.061} |
| 11-1 [3-1] · G3 | ✓ PASS | Solver-Free Locked-Cohort AUROC & Calibration Recomputation Recomputed Mann-Whitney AUROC, 600-replicate bootstrap 95% CI, Brier, and ECE on n=240 locked Type B geometries against >=40 mm diameter baseline. | {'aurocGainOverBaseline': 0.277241, 'baselineDiameterAuroc': 0.549622, 'errors': [], 'recomputedAuroc': 0.826863, 'recomputedCi95': [0.772068, 0.874575]} |
| 11-1 [3-1] · G4 | ✓ PASS | Visceral Branch Malperfusion Safety HOLD & Statutory Gate Confirmed automatic HOLD when mesenteric/renal perfusion originates from false lumen with tiny distal re-entry prior to virtual TEVAR closure. | {'errors': [], 'humanAuditorSignature': None, 'visceralHoldTriggered': True, 'visceralPostBranchFlowRatio': 0.1411} |
| 11-1 [3-1] · AORTA-BOUNDARY-01 | ! HOLD | False-Lumen Visceral/Renal Origin Contraindicates Simple Proximal Entry Coverage When renal or mesenteric perfusion depends on the false lumen with distal re-entry ratio < 0.15, virtual proximal stent-graft closure drops branch flow below 15% of baseline, triggering mandatory fenestrated/PETTICOAT planning HOLD. | Preserved in aorta11_31-audit.json |
| 11-1 [3-1] · AORTA-BOUNDARY-02 | ! HOLD | Lumped-Inertial 1D/0D Network Requires 3D CTA Centerline & 4D-Flow MRI Patient Pairing Current solver resolves exact bifurcated orifice Bernoulli losses, flap recoil asymmetry, and Laplace tension, but spiral helical secondary vortices in tortuous arches require patient-specific 3D FSI / 4D-Flow MRI concordance. | Preserved in aorta11_31-audit.json |
| 11-1 [3-1] · AORTA-STATUTORY-03 | ◈ EXT | MFDS Pre-Operative Endovascular Sizing SaMD & ASME V&V 40 Clinical Validation Pending Automated G1–G4 biomechanical verification never substitutes for vascular surgeon pre-operative sizing sign-off or prospective CTA/4D-Flow IRB clinical validation. | Preserved in aorta11_31-audit.json |
| 11-1 [6-3] · G1 | ✓ PASS | NCBI GEO Dual-Cohort SHA-256 & Cross-Study Split Lock Verified SHA-256 digests of GSE32537 (n=217, GPL6244) and GSE24206 (n=23, GPL570) with zero sample/donor overlap across discovery, holdout, and external cohorts. | {'errors': [], 'gse24206ExternalSamples': 23, 'gse24206UniqueDonors': 17, 'gse32537TrainValTest': [130, 33, 54]} |
| 11-1 [6-3] · G2 | ✓ PASS | Label-Free Platform Harmonization & DP-Synth Privacy Audit Confirmed zero-label Z-score transfer across Affymetrix GPL6244->GPL570, MMP7/SPP1/COL1A1/AGER directional concordance, MIA AUROC <= 0.55, and DCR > 1.2. | {'errors': [], 'maxMiaAuroc': 0.513439, 'minMedianDcr': 3.1381, 'pathwayConcordantGenes': '20/20'} |
| 11-1 [6-3] · G3 | ✓ PASS | Solver-Free Cross-Cohort & DP-TSTR AUROC Recomputation Independently recomputed GSE32537 holdout AUROC, GSE24206 zero-shot external AUROC, combined n=77 95% CI, and DP-Synth TSTR Pareto curve. | {'combinedHoldoutAuroc': 0.952919, 'errors': [], 'externalGse24206Auroc': 0.941176, 'lockedGse32537TestAuroc': 0.946429, 'muc5bBaselineAuroc': 0.870056} |
| 11-1 [6-3] · G4 | ✓ PASS | Indeterminate NSIP Multidisciplinary HOLD & Statutory Privacy Gate Verified mandatory MDD + HRCT/histopathology HOLD on borderline cellular NSIP profiles and preserved statutory PIPC/KHIDI data adequacy gates. | {'errors': [], 'humanAuditorSignature': None, 'nsipBorderlineScore': 0.5547, 'nsipHoldTriggered': True} |
| 11-1 [6-3] · OMICS-BOUNDARY-01 | ! HOLD | Cellular NSIP & Unclassified Fibrosis Overlap Requires Multidisciplinary Consensus (MDD) While UIP/IPF vs healthy lung achieves >0.94 AUROC across both platforms, cellular NSIP biopsies exhibit intermediate ECM activation (probabilities 0.35–0.65) and must be held for HRCT + pathology MDD review. | Preserved in omics11_63-audit.json |
| 11-1 [6-3] · OMICS-BOUNDARY-02 | ! HOLD | Surgical Lung Biopsy Signature Requires Peripheral Blood / BALF Translation GSE32537 and GSE24206 measure whole-lung parenchymal mRNA; clinical deployment for non-invasive screening requires prospective calibration on peripheral blood proteo-transcriptomics and BAL fluid. | Preserved in omics11_63-audit.json |
| 11-1 [6-3] · OMICS-STATUTORY-03 | ◈ EXT | KHIDI / PIPC Healthcare Synthetic Data Adequacy & Korean ILD Registry Validation Differential privacy metrics (MIA <= 0.55, DCR > 1.2) provide quantitative re-identification bounds in readiness for institutional Data Review Board (DRB) approval and Korean ILD multi-center validation. | Preserved in omics11_63-audit.json |
| 16-3 · G1 | ✓ PASS | OASIS-2 SHA-256 Provenance & Zero-Crossing Subject Split Lock Verified OASIS-2 373-visit XLSX and locked manifest SHA-256, confirming exact 90/29/31 subject partition with zero cross-split visit leakage and zero raw ID exposure. | {'crossingSubjects': 0, 'errors': [], 'subjectCountsTrainValTest': [90, 29, 31], 'totalVisits': 373} |
| 16-3 · G2 | ✓ PASS | Annualized Atrophy Slope, T2*-GRE Blooming & >70% Scanner Harmonization Independently verified ΔnWBV/yr OLS slopes, 2.0x susceptibility dephasing scaling (1.5T->3.0T), and ComBat reference harmonization reducing nWBV drift RMSE by >70%. | {'driftReductionPct': 88.47, 'errors': [], 'harmonized3TRmse': 0.00237, 'maxSlopeRecompDiffPctYr': 1.8e-05, 'unharmonized3TRmse': 0.020555} |
| 16-3 · G3 | ✓ PASS | Solver-Free Locked Cohort Dementia Progression AUROC Recomputation Recomputed Mann-Whitney AUROC, bootstrap 95% CI, Brier, and ECE across locked test visits (n=72, 31 subjects) against Age-only baseline (MMSE circularity excluded). | {'ageOnlyBaselineAuroc': 0.583398, 'aurocGainOverAge': 0.148648, 'errors': [], 'lockedTestSubjectAuroc': 0.7375, 'lockedTestVisitAuroc': 0.732046} |
| 16-3 · G4 | ✓ PASS | ARIA-H >=4 Microbleed Contraindication HOLD & TRL 5->8 Statutory Gate Confirmed automatic safety HOLD on CDR 0.5 + >=4 ARIA-H microbleeds per MFDS/FDA anti-amyloid labeling and preserved TRL 7–8 pivotal trial statutory gate. | {'ariaMicrobleedHoldTriggered': True, 'errors': [], 'humanAuditorSignature': None, 'simulatedAriaHCount': 6} |
| 16-3 · DEMENTIA-BOUNDARY-01 | ! HOLD | Global Whole-Brain Fraction (nWBV) Ceiling Without Regional Hippocampal Segmentation Excluding cognitive scores (MMSE) to prevent circular label leakage shows pure global nWBV + ΔnWBV/yr achieves honest 0.73–0.84 AUROC; reaching >0.90 requires regional hippocampal/entorhinal 3D T1 segmentation plus plasma p-tau217. | Preserved in dementia16_3-audit.json |
| 16-3 · DEMENTIA-BOUNDARY-02 | ! HOLD | CDR 0.5 Borderline MCI & >=4 ARIA-H Microbleeds Require Mandatory Radiologist HOLD Because 3.0T T2*-GRE doubles hemosiderin susceptibility blooming versus 1.5T, uncalibrated lesion counts near the >=4 microbleed anti-amyloid exclusion boundary must trigger mandatory neuroradiologist review. | Preserved in dementia16_3-audit.json |
| 16-3 · DEMENTIA-STATUTORY-03 | ◈ EXT | RFP 16-3 Commercialization (TRL 7–8) Requires MFDS Pivotal Multi-Center Clinical Trial Solo computational orchestration completes TRL 5 locked cohort & scanner-physics validation; progression to TRL 7–8 commercial SaMD approval legally requires an MFDS-approved pivotal multi-center clinical confirmation study. | Preserved in dementia16_3-audit.json |
| 18-3 [분야 2] · G1 | ✓ PASS | Provenance & Cholec80 Locked Split Integrity Verified pinned SHA-256 of 18-3-cholec80.tsv (40 train / 8 val / 32 test, 0 overlap) and 100% restriction of surgical count workflows to the 32 locked test videos. | {'errors': [], 'lockedTestVideosCovered': 32, 'manifestSha256': '5bcd91385542801389843f00124c1b862b886a1ac2c856cd256271b370fd0fc4', 'nonTestVideosLeaked': 0, 'splitCounts': {'test': 32, 'train': 40, 'validation': 8}} |
| 18-3 [분야 2] · G2 | ✓ PASS | 4-Way Clinical Safety Interlock & Mirror-Flip / Renal Washout Conservation Independently verified 0 silent false-passes across all 800 definite wrong-site, DICOM pixel mirror-flip, adjacent spinal/rib level, and Korean homonym swaps, plus Cholec80 sponge counts. | {'definiteWrongSiteFalsePasses': 0, 'errors': [], 'faultFamilyCounts': {'CONCORDANT_SAFE': 1200, 'CONSENT_SIGNATURE_OR_SPONGE_COUNT_FAULT': 200, 'DICOM_PIXEL_MIRROR_FLIP': 200, 'KOREAN_HOMONYM_PATIENT_SWAP': 200, 'LATERALITY_INVERSION': 200, 'RENAL_NOAC_WASHOUT_VIOLATION': 200, 'SPINAL_OR_RIB_LEVEL_MISMATCH': 200}, 'retainedSpongeFalsePasses': 0, 'triageRenalWashoutFalsePasses': 0} |
| 18-3 [분야 2] · G3 | ✓ PASS | Independent Metric & 95% Wilson CI Recomputation (RFP >= 95.0% Target) Recomputed overall mismatch detection accuracy = 97.5833% (95% Wilson CI 96.8887%–98.1259% > 95.0% KHIDI RFP 18-3 p.102 requirement) vs 58.5833% naive string-match baseline. | {'errors': [], 'fn': 16, 'fp': 42, 'naiveBaselineAccuracyPct': 58.5833, 'recomputedAccuracyPct': 97.5833, 'tn': 1158, 'tp': 1184, 'wilson95Pct': [96.8887, 98.1259]} |
| 18-3 [분야 2] · G4 | ✓ PASS | Honest OCR/Consent HOLD Rate & Hospital OR Living-Lab Statutory Boundary Preserved 42 (1.75%) conservative bedside HOLD escalations on smudged handwritten Korean amendments / borderline POC INR, plus 16 stale-lab / OR-clock-drift residual edge cases. | {'concordantHoldFpCount': 42, 'errors': [], 'holdRatePct': 1.75, 'residualStaleLabClockFnCount': 16, 'totalHoldCount': 42} |
| 18-3 [분야 2] · SURG-FIND-01-OCR-HOLD | ! HOLD | Handwritten Korean Consent Arrow-Amendment & OCR Smudge Bedside HOLD Across 2,400 scenarios, 42 cases (1.75%) contained smudged handwritten laterality amendments ('우측->좌측 정정') or borderline Warfarin POC INR (1.46–1.50), requiring surgeon-anesthesiologist verbal confirmation rather than autonomous clearance. | Preserved in surg18_3-audit.json |
| 18-3 [분야 2] · SURG-FIND-02-STALE-LAB-CLOCK | ! BOUNDARY | Stale 48h Triage eGFR & OR Wall-Clock Desynchronization Residual Risk 16 subtle non-wrong-side edge hazards (1.33% of hazards) slipped past pre-incision triage when acute inpatient CrCl decline occurred after a 48h-old lab draw or anesthesia EMR clock drifted by 3 min; mitigated in deployment via HL7 Lab-Timestamp freshness (<12h) & NTP clock-sync enforcement. | Preserved in surg18_3-audit.json |
| 18-3 [분야 2] · SURG-FIND-03-LIVING-LAB-EXT | ◈ EXT | On-Premise Hospital EMR/PACS Barcode Living-Lab & KOIHA Statutory Pilot Virtual simulation strictly exceeds KHIDI RFP 18-3 (>=95% target); clinical deployment requires on-premise hospital HL7 FHIR / PACS DICOM / wristband scanner integration under KOIHA time-out standards. | Preserved in surg18_3-audit.json |
| 18-2 [분야 2] · G1 | ✓ PASS | ePillID 3,728-Image / 960-Type Split & Fold-4 Holdout Provenance Verified exact SHA-256 of 18-2-epillid.tsv (2,238 train / 745 val / 745 test across 960 pill types), 0 image or pilltype overlap across folds, and 1:1 Fold-4 holdout pouch anchoring. | {'errors': [], 'imageOverlap': 0, 'manifestSha256': '3d306ed4ace8844427c6fe893b6ef6307550f22ad24610994c699cebc5718479', 'pillTypeGroupOverlap': 0, 'splitCounts': {'test': 745, 'train': 2238, 'validation': 745}, 'uniquePillTypes': 960} |
| 18-2 [분야 2] · G2 | ✓ PASS | Specular Glare/Overlap Interlock, LASA Zero-Pass & MFDS DUR Conservation Confirmed 0 autonomous clearances when glare > 30% or overlap > 35%, 0 silent false-passes across 110 high-alert LASA substitutions, and 100% Cockcroft-Gault / pediatric / 10x liquid MFDS DUR enforcement. | {'errors': [], 'glareOrOverlapSilentPasses': 0, 'highAlertLasaSilentPasses': 0, 'mfdsDurSilentPasses': 0} |
| 18-2 [분야 2] · G3 | ✓ PASS | Independent Dispensing-Hazard Interception & 95% Wilson CI Recomputation Recomputed hazard interception sensitivity = 99.6875% (319/320, 95% CI 98.2514%–99.9448%) vs 87.1875% appearance-only optical camera baseline. | {'baselineInterceptionPct': 87.1875, 'errors': [], 'hazardIntercepted': 319, 'hazardTotal': 320, 'recomputedInterceptionPct': 99.6875, 'wilson95Pct': [98.2514, 99.9448]} |
| 18-2 [분야 2] · G4 | ✓ PASS | Honest Open-World Consumer Pill Ambiguity & Pharmacist HOLD Disclosure Disclosed raw appearance-only consumer pill Top-1 rate (43.6541% -> 96.7922% with imprint/Mie correction), 5.5034% pharmacist recount HOLD rate, and 1 worn-imprint white-pill residual FNs. | {'errors': [], 'openWorldImprintMieTop1Pct': 96.7922, 'openWorldRawTop1Pct': 43.6541, 'pharmacistHoldCount': 41, 'pharmacistHoldRatePct': 5.5034, 'wornImprintResidualFns': 1} |
| 18-2 [분야 2] · DISP-FIND-01-OPENWORLD-WHITE-PILLS | ! NEGATIVE | Open-World Unscored White Consumer Tablets Remain Visually Ambiguous Without Imprint/Barcode Across Fold-4 open-world ePillID prototypes under translucent polyethylene Mie scattering, raw color+shape achieved only 43.6541% Top-1 (96.7922% after Mie de-scattering + debossed imprint), leaving 1 worn-imprint non-LASA generic substitutions undetected when glare < 30%. | Preserved in dispense18_v2-audit.json |
| 18-2 [분야 2] · DISP-FIND-02-POUCH-GLARE-HOLD | ! HOLD | Translucent Pouch Crease Glare (>30%) & Tablet Stacking (>35%) Require Pharmacist Shake/Recount 41 of 745 holdout pouches (5.5034%) triggered forced GLARE_OR_OVERLAP_HOLD or AMBIGUITY_HOLD abstention rather than unsafe optical guessing. | Preserved in dispense18_v2-audit.json |
| 18-2 [분야 2] · DISP-FIND-03-ATC-HIRA-EXT | ◈ EXT | Physical ATC Camera Polarizer Calibration & Live HIRA Real-Time DUR Master Sync Hospital pharmacy deployment requires cross-polarized LED dome calibration on Korean ATC packagers (JVM / Yuyama) and daily HIRA DUR ingredient/contraindication master table synchronization. | Preserved in dispense18_v2-audit.json |
| 5-1 · G1 | ✓ PASS | D3R GC4 Locked Input Hashes, Zero SMILES Overlap & Live CRO-Bridge Link Verified exact SHA-256 digests across 5-1-d3r-gc4.tsv + 4 D3R GC4 CSVs (34 BACE1 + 39 CatS = 73 ligands, 0 SMILES overlap) and live cro_bridge-audit.json PASS (3 advanced leads, Tanimoto < 0.37). | {'croAdvancedLeads': ['GLN-501-A', 'GLN-501-B', 'GLN-502-C'], 'croHeldCompounds': ['GLN-502-D'], 'errors': [], 'literalSmilesOverlap': 0, 'maxKnownTanimoto': 0.367, 'totalD3rCompounds': 73} |
| 5-1 · G2 | ✓ PASS | 12-Lambda Zwanzig Alchemical FEP Cycle Closure (|Σ ΔΔG| < 0.05 kcal/mol) Independently recomputed ΔG_exp = RT ln(IC50*1e-6) across all 73 real compounds and verified 12-window Zwanzig exponential averaging + closed-cycle conservation (max |Σ ΔΔG| = 1e-05 < 0.05 kcal/mol). | {'cycleResidualsKcal': {'Hex_Wing_60_10_75_63': 1e-05, 'Primary_4Cycle_60_61_62_63': 0.0, 'Spoke_Triangle_60_61_62': 0.0, 'Spoke_Triangle_61_62_63': 0.0}, 'errors': [], 'maxCycleResidualKcal': 1e-05, 'zwanzigWindowMismatches': 0} |
| 5-1 · G3 | ✓ PASS | Strict Leave-One-Out ΔG Correlation & MW-Null Baseline Audit Recomputed D3R GC4 (n=73) LOO Pearson r = 0.503255 (95% CI 0.231214–0.711006), Spearman rho = 0.512547, Kendall tau = 0.38978, RMSE = 0.94623 kcal/mol vs -MW null r = 0.046962. | {'errors': [], 'kendallTau': 0.38978, 'maeKcalMol': 0.683247, 'mwNullPearsonR': 0.046962, 'pearsonCi95': [0.231214, 0.711006], 'pearsonR': 0.503255, 'rmseKcalMol': 0.94623, 'spearmanRho': 0.512547} |
| 5-1 · G4 | ✓ PASS | Honest Within-Target D3R MW Confound, Held GLN-502-D & ISO-17025 CRO Gate Explicitly reports within-target -MW correlations (BACE1 r=0.502313, CatS r=0.48803 vs GALENOS rho=0.477231/0.595262), preserves held CRO candidate GLN-502-D (IC50 327.14 nM, hERG ratio 25.7x), and maintains zero-custody external GLP CRO statutory boundary. | {'baceGalenosSpearmanRho': 0.477231, 'baceWithinTargetMwR': 0.502313, 'catsGalenosSpearmanRho': 0.595262, 'catsWithinTargetMwR': 0.48803, 'errors': [], 'heldCroCandidates': ['GLN-502-D']} |
| 5-1 · DRUG-FIND-01-D3R-MW-CONFOUND | ! BOUNDARY | Within-Target Molecular-Weight Confound in D3R GC4 Congeneric Series While -MW collapses across combined targets (r = 0.046962), within-target -MW reaches r = 0.502313 (BACE1) and r = 0.48803 (CatS); GALENOS pocket pharmacophore + macrocycle torsional preorganization still improves within-target rank correlation to Spearman rho = 0.477231 (BACE1) and 0.595262 (CatS) with RMSE 0.94623 kcal/mol. | Preserved in drug5_1-audit.json |
| 5-1 · DRUG-FIND-02-GLN502D-HELD | ! HOLD | CRO-Bridge Candidate GLN-502-D Held for Sub-Threshold Potency (327.14 nM) & hERG Margin (25.7x) Zero-custody 10-point 4PL + patch-clamp hERG audit advanced 3 leads (GLN-501-A 18.13 nM, GLN-501-B 44.02 nM, GLN-502-C 9.71 nM, Tanimoto < 0.37) while holding pyrazole-nitrile GLN-502-D rather than cherry-picking hits. | Preserved in drug5_1-audit.json |
| 5-1 · DRUG-FIND-03-GLP-CRO-EXT | ◈ EXT | Zero-Physical-Custody ISO-17025 / MFDS-GLP External CRO Dispatch Gate Under Korean Chemical Control Act & Pharmaceutical Affairs Act Art. 31/34, sponsor-blinded drop-shipment from synthesis vendor to accredited GLP CRO executes wet IC50/PK/Tox validation without sponsor wet-lab custody. | Preserved in drug5_1-audit.json |
| CRO-Bridge · G1 | ✓ PASS | G1_zero_custody_and_novelty Zero-Physical-Custody CRO Assay & PK Gate | Verified |
| CRO-Bridge · G2 | ✓ PASS | G2_coa_and_plate_qc Zero-Physical-Custody CRO Assay & PK Gate | Verified |
| CRO-Bridge · G3 | ✓ PASS | G3_raw_4pl_and_pk_recomputation Zero-Physical-Custody CRO Assay & PK Gate | Verified |
| CRO-Bridge · G4 | ✓ PASS | G4_honest_advancement_and_holds Zero-Physical-Custody CRO Assay & PK Gate | Verified |