Model Registry

Model Status

Research triage ready

Final governance complete; clinical use prohibited

F1 @ 0.38

93.3%

Holdout threshold 0.38

Recall @ 0.38

87.5%

True positive rate

Specificity @ 0.38

100.0%

True negative rate

Holdout Rows

set11

set41 F1 @ 0.71

99.3%

Disjoint set41 surveillance gate

Final Governance & Use Boundary

Pass99 signs off v24 for research triage only. The final status remains approved_for_research_triage_use.

Research-triage sign-offapproved for research triage use

Clinical boundaryprohibited

Governed scorer.venv-training/bin/python scripts/python/score_v24_research_genes_v1.py --threshold 0.71 GENE_SYMBOL

Model IDv24_source_feature_integration

Candidate Threshold0.71

Endpoint Readinessready for research triage endpoint

Manifest Replaypassed

Set41 Surveillanceset41 surveillance passed

Production Useapproved research triage blocked clinical

Allowed use

Research triage for prioritizing genes for expert review.
Internal experiment planning and residual monitoring.
Batch candidate scoring through the governed v24 research scorer.

Prohibited use

Clinical diagnosis or treatment decisions.
Standalone pathogenicity, penetrance, or disease-causality claims.
Patient-specific interpretation or medical decision support.

Pass99 Final Governance Sign-off Final Reproducibility Manifest Pass99 Manifest Replay v24 Endpoint Readiness Contract Model Readiness Endpoint

Clinical Translation Readiness

The model is trained successfully and research-triage ready, but it is not ready for clinical disease detection until patient-level validation and prospective clinical gates are passed.

Research usageready for governed research triage

Disease detectionnot ready for clinical disease detection

Clinical useprohibited

Model Trainingtrained successfully

Clinical Readinessblocked pending clinical validation

Blocking Gates7

Can Claim Disease Detectionno

Patient-Level Validationready pending real adjudicated cohort

Clinical Endpointlocked for retrospective validation

Clinical readiness summary

v24 is trained successfully and ready for governed research triage, but patient-level validation missing or incomplete; prospective validation readiness exists through Pass111 but real prospective evidence is not complete. Therefore it is not ready for clinical disease detection, diagnosis, treatment decisions, or patient-specific interpretation.

/api/research/clinical-readiness pass100_clinical_translation_readiness_v1.json

Patient-Level Validation Protocol

Pass101 converts the missing clinical evidence gate into an executable validation package: locked intended use, cohort schema, cohort CSV template, and a validator that rejects PHI-bearing or undersized cohorts before scoring.

Validation packageready pending real adjudicated cohort

Clinical useprohibited

Disease detectionnot ready for clinical disease detection

Prepared Gates4

Remaining Clinical Gates6

Frozen Threshold0.71

Real Cohort Evidencepending

Cohort Unitpatient gene candidate

Clinical Claimblocked

Pass101 readiness summary

Pass101 locks the intended-use endpoint, schema, cohort CSV template, and patient-level validation runner; v24 remains trained and research-triage ready, but clinical use is still prohibited until a real adjudicated patient-level cohort passes validation.

/api/research/patient-validation pass101_patient_level_validation_protocol_v1.json pass101_patient_level_validation_schema_v1.json cohort CSV template

Retrospective Validation Runner

Pass102 dry-runs the locked patient-level validation analysis: score join, confusion matrix, AUROC, Brier score, subgroup suppression, and residual extraction. This is a dry-run sample, not clinical evidence.

Runner statusdry run sample not clinical evidence

Clinical useprohibited

Clinical evidencereal cohort pending

Rows Scored4

Score Joincomplete

TP / FP1 / 1

TN / FN1 / 1

Threshold0.71

Disease Claimblocked

Dry-run boundaryDry-run sample/template output is not clinical evidence and cannot support disease-detection claims.

/api/research/retrospective-validation pass102_retrospective_validation_dry_run_v1.json pass102_patient_validation_score_template_v1.csv

External Validation Intake

Pass103 creates the frozen intake package for a governed patient-level cohort: cohort hashes, model hash, score artifacts, and candidate score input. The current package is a dry-run sample, not clinical evidence.

Intake statusdry run sample not clinical evidence

Frozen intake packagereal cohort pending

Clinical useprohibited

Candidate Rows4

Cohort Checkschema passed sample not clinical evidence

Locked Validationdry run sample not clinical evidence

Score Joincomplete

Rows Scored4

Disease Claimblocked

/api/research/external-validation-intake pass103_external_validation_intake_package_v1.json candidate score input

Retrospective Acceptance Gates

Pass104 locks metric floors, calibration, subgroup reportability, residual-review, and prospective validation gates. The current dry-run is blocked sample not clinical evidence.

Acceptance statusblocked sample not clinical evidence

Retrospective gateblocked

Clinical useprohibited

Blocking Gates5

Rows Evaluated4

Sensitivity / Specificity50.0% / 50.0%

AUROC75.0%

Metric floorslocked

Disease Claimblocked

/api/research/retrospective-acceptance pass104_retrospective_acceptance_gate_policy_v1.json pass104_retrospective_acceptance_gate_dry_run_v1.json

Source Manifest Readiness

Pass106 turns the missing source-manifest blocker into a schema, fillable template, validator, and template check. The current template is blocked template source manifest until real governance metadata replaces placeholders.

Manifest statusblocked template source manifest

Cohort admissibilityblocked

Clinical useprohibited

Blocking Gates1

Deidentificationhipaa safe harbor

Provenance Sites2

Template Placeholdersblocked

Disease Claimblocked

Manifest Schemalocked

/api/research/source-manifest pass106_source_manifest_schema_v1.json source manifest template

Source Manifest Completion

Pass107 translates the Pass106 template blocker into a cohort-owner completion packet. The current handoff is blocked by missing owner evidence until governance, provenance, adjudication, and training-overlap fields are replaced with real source metadata.

Completion statusblocked missing owner evidence

Pass105 readinessblocked

Clinical useprohibited

Blocking Items9

Owner Evidencemissing owner evidence

Upstream Gates4

Disease Claimblocked

First Fieldcohort id

Owner Roledata steward

/api/research/source-manifest-completion pass107_source_manifest_completion_checklist_v1.json pass107_source_manifest_completion_packet_v1.json

Cohort Intake Preflight

Pass108 runs the source-manifest and cohort-admissibility gates as one preflight before Pass103 or Pass102 execution. The default dry run is blocked template inputs pending a real deidentified cohort.

Preflight statusblocked template inputs pending real cohort

Pass103 intakeblocked

Clinical useprohibited

Blocking Gates4

Pass102 validationblocked

Source Manifestblocked template source manifest

Cohort Statusblocked sample not clinical evidence

Command Plan6

Disease Claimblocked

/api/research/cohort-intake-preflight pass108_cohort_intake_preflight_dry_run_v1.json

Frozen v24 Scoring Handoff

Pass109 turns the governed v24 scorer into the locked score CSV contract required by Pass102. The default state is blocked preflight until Pass108 clears with real deidentified cohort files.

Handoff statusblocked preflight not ready for locked scoring

Locked scoringblocked

Clinical useprohibited

Threshold0.71

Candidate Rows4

Unique Genes4

Score Columns5

Pass102 validationblocked

Upstream Preflightblocked template inputs pending real cohort

Locked Rows0

Disease Claimblocked

/api/research/frozen-scoring-handoff pass109_frozen_v24_scoring_handoff_v1.json pass109_frozen_v24_score_schema_v1.json pass109_frozen_v24_score_template_v1.csv

Retrospective Validation Launch

Pass110 runs the Pass108, Pass103, Pass109, Pass102, and Pass104 chain as one launch sequence. The default state is blocked preflight until real cohort files and governed v24 scores are supplied.

Launch statusblocked preflight not ready for retrospective execution

Prospective readinessblocked

Clinical useprohibited

Threshold0.71

Candidate Rows0

Locked Scores0

Rows Scored0

Retrospective Gateblocked

Blocking Gates4

Pass104not run

Disease Claimblocked

/api/research/retrospective-validation-launch pass110_retrospective_validation_launch_dry_run_v1.json pass110_retrospective_validation_launch_suite_summary_v1.json

Prospective Validation Readiness

Pass111 defines the prospective multisite protocol, event schema, and event template that follow a successful Pass110 launch. The current state is blocked retrospective launch, so enrollment and clinical disease-detection claims remain prohibited.

Prospective statusblocked retrospective launch not ready for prospective validation

Enrollmentblocked

Clinical useprohibited

Event Rows0

Positive / Negative0 / 0

Sites0

Schema / Minimumspassed / blocked

Prospective Gateblocked

QMS Reviewblocked

Sensitivity / Specificityn/a / n/a

Disease Claimblocked

/api/research/prospective-validation-readiness pass111_prospective_validation_protocol_v1.json pass111_prospective_event_schema_v1.json prospective event template readiness artifact

Retrospective Cohort Evidence

Pass112 turns the missing real-cohort blocker into an owner-fillable dossier for governance, provenance, adjudication, locked score artifacts, and sign-off. The current state is blocked missing real cohort evidence, so Pass108/Pass110 remain blocked as clinical evidence.

Dossier statusblocked missing real cohort evidence

Pass108 preflightblocked

Clinical useprohibited

Blocking Gates10

Cohort Rows0 / 300

Positive / Negative0 / 0

Sites0 / 3

Template Evidenceblocked

Disease Claimblocked

/api/research/retrospective-cohort-evidence evidence schema evidence template dossier artifact

Retrospective Execution Bundle

Pass113 is the governed real-cohort launch bundle. It requires a ready Pass112 dossier before running Pass110; the default state is blocked Pass112 dossier, so no retrospective clinical evidence is claimed.

Bundle statusblocked pass112 dossier not ready

Pass110 launchblocked

Clinical useprohibited

Pass112 Statusblocked missing real cohort evidence

Pass112 Gates10

Pass108 Preflightnot run

Rows Scored0

Retrospective Gateblocked

Disease Claimblocked

/api/research/retrospective-execution-bundle execution runbook execution bundle

Real Cohort Intake Packet

Pass114 is the owner upload manifest and checklist for the missing real-cohort evidence. It collects references, hashes, counts, and role sign-off readiness; the current state is blocked owner upload manifest, so Pass112 and Pass113 remain blocked.

Intake statusblocked owner upload manifest not ready

Pass112 completionblocked

Clinical useprohibited

Required Uploads11

Missing Uploads10

Template Evidencedetected

Pass112 Gates10

Target Rows0

Disease Claimblocked

/api/research/real-cohort-intake-packet intake packet owner upload manifest owner checklist

Pass112 Dossier Bridge

Pass115 maps a validated Pass114 owner upload manifest into a generated dossier for Pass112 owner review. The current generated dossier is blocked because the owner manifest is still a template.

Bridge statusblocked pass114 owner manifest not ready

Dossier replacementblocked

Clinical useprohibited

Pass114 Statusblocked owner upload manifest not ready

Missing Uploads10

Generated Dossierblocked missing real cohort evidence

Field Mappings15

Blocked Mappings13

Pass113 Executionblocked

/api/research/pass112-dossier-bridge bridge summary generated dossier field mapping

Cohort Admissibility Audit

Pass105 checks whether a patient cohort is admissible as retrospective evidence before score joins or acceptance gates: template fingerprints, source manifest, PHI-like values, duplicates, and diversity. The current dry-run is blocked sample not clinical evidence.

Admissibility statusblocked sample not clinical evidence

Retrospective evidenceblocked

Clinical useprohibited

Blocking Gates3

Sites / Labels2 / 2p 2n

PHI-like Values0

Cells Screened4

Template fingerprintblocked

Disease Claimblocked

/api/research/cohort-admissibility pass105_cohort_admissibility_dry_run_v1.json audited cohort CSV

Training Snapshot — Active Challenger Holdout (Set 11)

Detailed metrics from ndd_model_training_report_v24_source_feature_integration.json

Selected Modelsource_feature_integration_stacked_mlp

Holdout Setset11

Holdout Rows16

F1 @ 0.3893.3%

Recall @ 0.3887.5%

Specificity @ 0.38100.0%

External Validation Snapshot — set41

Latest disjoint-holdout readout used for promotion-gating decisions. Source: set41_suite_summary_v1.json

Cohort Rows144

Best Model @ 0.38v24_sf

v24_sf F1 @ 0.3899.3%

v24_sf Recall @ 0.38100.0%

v24_sf Specificity @ 0.3898.6%

Residual FN Genesnone

Related Evaluation Files

Auto-detected from artifacts/models. 25 artifacts found.