Model Management

Lifecycle pipeline

Training Tuned ShadowMode Active Retired
Training v4.9-train · 62% Tuned v4.8-tuned · ready ShadowMode v4.8-shadow · 12d Active v4.7-prod · 31d Retired v4.6 · 31d
Shadow agreement
97.4%
▲ 0.6 pts · ready to promote
Target ≥97% (CR-001)
Precision · shadow
94.1%
▲ 1.1 pts vs prod
7-day rolling
Recall · shadow
89.2%
▲ 2.4 pts vs prod
vs analyst-confirmed labels
FPR drift · High-LTV
+0.08 pts
Acceptable, monitor
Stability check (CR-007)

Model versions

All transitions captured in immutable audit (CR-005)
Version State (WS-004) Training data window Precision Recall FPR (High-LTV) Activated Approver Actions
scoring-v4.8-shadow ShadowMode 2024-11 → 2025-04 94.1% 89.2% 0.36% R. Patel
scoring-v4.7-prod Active 2024-09 → 2025-03 93.0% 86.8% 0.44% 2025-04-22 R. Patel + S. Liu
scoring-v4.8-tuned Tuned 2024-11 → 2025-04 R. Patel
scoring-v4.9-train Training 2024-12 → 2025-05
scoring-v4.6-archive Retired 2024-07 → 2025-01 91.3% 83.6% 0.51% 2025-03-22 R. Patel + M. Cohen

Precision & recall by LTV tier · shadow vs production

7-day rolling
100% 80% 60% High-LTV Mid-LTV New/Low Guest
v4.7 Production v4.8 Shadow

Shadow agreement over 14 days

target ≥97%
99% 97% 95%
Crossed 97% threshold on day 9 · sustained above target for 5 consecutive days · eligible for promotion review.

Recent lifecycle events (audit)

View full log →
WhenFrom → ToModelActorCo-approverRef
2025-05-11 09:12TunedShadowModev4.8R. PatelS. Liuaud_8721
2025-05-08 17:30TrainingTunedv4.8System (auto)aud_8588
2025-04-22 14:02ShadowModeActivev4.7R. PatelS. Liuaud_8401
2025-04-22 14:02ActiveRetiredv4.6R. PatelS. Liuaud_8402