Performance snapshot (sim)

Clinic model: P50 ≈ 818 ms, P95 ≈ 937 ms (n=1,680)

Inpatient model: P50 ≈ 614 ms, P95 ≈ 750 ms (n=1,680)

Daily SLO: Clinic ≥ 95% under 920 ms across days; Inpatient ~83–91% under 700 ms (Wilson 95% lower bounds shown in app)

That’s compelling for live dashboards—supports logic @ the speed of thought without overclaiming.

Live demo (sim data): streamlit run scripts/dashboard_app.py

Docs

Examples

Verified metrics (strict M = πF(D))

From sql/verify_theorems_public_auto.sql on 2025-08-18 (Asia/Riyadh).

ModelSoundnessCoverageDuplicate keysδ determinismMasks (bit→cat)Pruned
Clinic Visit 0 0 0 — 7 → 6 −1 (14.29%)
Inpatient Admission 0 0 0 — 20 → 10 −10 (50.00%)

Dead masks (bit-valid ∧ category-impossible) are diagnostics only: Clinic {8}; Inpatient {4,5,16,17,20,21,32,33,36,37}.