Current Combined Findings

Latest v24 source-feature integration challenger consumes set37 labels as training data, keeps the Set11 comparator window at F1 93.3%, recall 87.5%, specificity 100.0%, and repairs the set37 v23 boundary failure on training fit (delta F1 +48.1 pp, delta recall +56.2 pp, delta specificity +25.0 pp).

V24 preserves the current checks at set36 F1 90.9% (recall 83.3%, specificity 100.0%) and set29 F1 73.5%, but remains research-only because set37 is now training evidence and set38 fresh replication is required.

Disjoint external holdout set41 (n=144) shows v24_sf at F1 99.3%, recall 100.0%, specificity 98.6%.

Pass98 live-scores set41 as the first post-governance surveillance gate for v24, using a 144-row fresh cohort of adjudicated positives and hard boundary controls.

On set41 at the 0.38 anchor, v24_sf reaches F1 99.3%, recall 100.0%, specificity 98.6%; at candidate threshold 0.71, v24_sf reaches F1 99.3%, recall 100.0%, specificity 98.6%.

Pass98 set41 surveillance decision: set41 surveillance passed.

Pass95 calibration/stability analysis across set38+set39 identifies candidate threshold 0.71 with F1 99.0%, recall 100.0%, specificity 97.9%, and 0 false negatives; set40 is now frozen as a 192-row candidate gate.

Pass97 packages v24 governance and reproducibility artifacts, and Pass98 clears the required set41 surveillance gate; production use remains blocked pending final governance sign-off at candidate threshold 0.71 (F1 99.5%, recall 100.0%, specificity 99.0%).

Operational recommendation: v24 can be used for governed research triage through the approved scorer and readiness contract; clinical use remains prohibited.

Pass99 completes final governance sign-off for v24 research triage: approved for research triage use, endpoint readiness ready for research triage endpoint, with clinical diagnosis and treatment use still prohibited.

Pass100 clinical translation readiness: v24 is trained and ready for governed research triage, but patient-level validation missing means it is not ready for clinical disease detection.

Pass101 patient-level validation readiness: the patient-level validation runner, locked endpoint, schema, and cohort CSV template are ready pending a real adjudicated cohort; clinical use remains prohibited.

Pass102 retrospective validation runner dry-run: score join and locked threshold metric computation are executable on the sample template (TP 1, FP 1, TN 1, FN 1), but the dry-run sample is not clinical evidence.

Pass103 external validation intake: the frozen intake package and candidate score input are now executable for a DUA/IRB-appropriate cohort, but the current dry-run remains sample/template evidence only.

Pass104 retrospective acceptance gate: locked metric floors, calibration, subgroup, residual-review, and prospective-validation gates are now executable; the current dry-run is blocked as sample evidence only.

Pass105 cohort admissibility audit: template fingerprints, source-manifest governance, PHI-like cell values, duplicate candidate rows, site/label diversity, and Pass101 carry-forward gates are now executable before any patient cohort can become retrospective evidence.

Pass106 source manifest readiness: the external cohort source-manifest schema, template, validator, and template check are now available so cohort owners can satisfy Pass105's source-manifest gate without weakening the clinical-use boundary.

Pass107 source-manifest completion packet: the Pass106 template blocker is now translated into a cohort-owner checklist with required evidence roles and gate mappings; current status remains blocked by missing owner evidence.

Pass108 cohort-intake preflight: the source-manifest and patient-cohort gates are now orchestrated as one runnable preflight before Pass103/Pass102 execution; the default run remains blocked template inputs pending a real deidentified cohort.

Pass109 frozen v24 scoring handoff: the governed v24 scorer now has a locked Pass102 score CSV contract and assembly path; default status remains blocked preflight until real cohort files clear Pass108.

Pass110 retrospective validation launch: the full Pass108/103/109/102/104 execution chain now has a one-command launcher; default status remains blocked preflight until real cohort files and governed v24 scores are supplied.

Pass111 prospective validation readiness: the multisite prospective protocol, event schema, and event template now exist; default status remains blocked retrospective launch, and clinical use remains prohibited.

Pass112 retrospective cohort evidence dossier: the real cohort-owner evidence contract now exists for governance, provenance, adjudication, score artifacts, and sign-off; default status remains blocked missing real cohort evidence, and clinical use remains prohibited.

Pass113 retrospective execution bundle: the governed launch point now requires Pass112 real-cohort evidence before running Pass110; default status remains blocked Pass112 dossier, and clinical use remains prohibited.

Pass114 real cohort intake packet: the owner upload manifest, schema, and checklist now make the remaining real-cohort evidence handoff explicit; default status remains blocked owner upload manifest, and clinical use remains prohibited.

Pass115 Pass112 dossier bridge: the validated owner upload manifest can now be mapped into a Pass112-compatible generated dossier for owner review; default status remains blocked Pass114 owner manifest, and clinical use remains prohibited.