Display filters
Scoring presence = 2 is excluded globally from tiles, charts, and tables.
What was scored
Evaluation contextiHigh-level context for the current cohort and scoring view.
This band tells the user what dataset is being evaluated before they interpret the metric-specific results.
PatientsiNumber of unique patients loaded from the uploaded JSON dataset.
0
Items (patient × ICD-10)iCount of unique patient-condition pairs summarized across runs.
0
RunsiMaximum run count detected in the uploaded cohort.
—
Scored itemsiRows that had usable truth labels and were included in validation scoring after excluding clinically uncertain presence labels.
—
Shown when truth rows are available
Clinical uncertainty excludediNumber of conditions excluded from validation scoring because the clinician marked clinical presence as uncertain.
—
Shown when truth rows are available
Metric workspace
Clinical PresenceiAccuracy and stability view for clinical presence scoring.
Review how clinical presence performs against truth labels and how stable it remains across repeated runs.
Clinical Presence validation metrics unavailable
The current reference file appears to be supplying ICD-10 metadata but not the patient-level truth rows needed to calculate confusion-matrix metrics.
Threshold disagreementiCount of patient-condition items with at least one emitted run above the Presence threshold and at least one emitted run below it.
0
Across 0 patient-condition pairs
Vote agreementiAgreement rate across emitted run-level threshold votes for clinical presence.
0%
Disagree votes: 0
Swing (Med + High)iPresence range excludes omissions from min/max. Highlights patient-condition pairs with meaningful run-to-run variability.
0
Med: 0 · High: 0 · Max: 0.000
Omission instabilityiPatient-condition items missing in one or more runs. Presence omission is tracked separately from emitted-only swing.
0
Missing in 3+ runs: 0
Presence threshold = 0.92. Presence range excludes omissions from min/max and tracks missing runs separately.
Interpretation
Clinical analysisiOpen narrative interpretation of model behavior for Clinical Presence and MEAT in a tabbed modal.
These launchers open clinician-facing analysis content styled to match the dashboard.
Priority review
Highest-priority patient-condition pairsiTop-ranked patient-condition items ordered by disagreement, low consistency, omission instability, swing severity, and opportunity-type instability.
Presence range excludes omissions; missing runs are shown separately.
Top 12
PatientiPatient label and internal patient ID for the ranked condition row.ICD-10iDiagnosis code associated with the ranked patient-condition pair.OppiOpportunity-type composition across runs, with omitted runs shown separately.MEAT rangeiDifference between the highest and lowest max MEAT score across emitted runs.MEAT minority countiSmaller side of the threshold vote split across runs. Lower is better.Presence emitted-only rangeiRange across emitted clinical Presence scores only. Missing runs are excluded from min/max.Missing runsiNumber of runs where the Presence row was not emitted for this patient-condition item.Presence minority countiSmaller side of the threshold vote split across emitted runs. Lower is better.
No rows match the current filters.
Distribution
Swing distributioniDistribution of score ranges across runs for each patient-condition pair. Presence uses emitted-only range; omission counts are shown separately.
Range across runs per patient × ICD-10. Y-axis shows count.
001110–.020.02–.040.04–.060.06–.080.08–.100>.100Count
Buckets: 0–.02, .02–.04, .04–.06, .06–.08, .08–.10, >.10