Service 7

Coding Engine

Trajectory-justified ICD-10 / CPT

Version

2026.02.1

Retrained

2026-01-24

Status

healthy

Performance

Key metrics

Calibration & discrimination against the engine's published target. Under-target values are highlighted.

Code precision

94.0%target 90%

Code recall

88.0%target 85%

Specificity (vs unspecified)

81.0%target 75%

Clinician acceptance

79.0%target 75%

14-day inference volume

864 in the last 24h · avg latency 78 ms

Metrics are tracked on held-out evaluation cohorts; values here reflect the last retrain.

Model

Training basis & runtime

Transparent about what the engine was trained on and how it behaves in production.

Model family

Retriever + trajectory-conditioned classifier

Trained on

18M coded encounters · CPT + ICD-10

Suggested codes / consult

3.1

Specificity bias

0.81

versus baseline 0.64

Billing bridge hit-rate

72%

QOF register hit-rate

89%

History

Changelog

Each retrain is documented; clinicians can see what changed and when.

2026.02.12026-01-24
QOF CV003 and DM020 bridges tuned with trajectory-qualifying logic.
2025.09.02025-09-11
Switched from bag-of-claims retriever to syndrome-conditioned retriever.

Safety

Guardrails

Hard limits and runtime checks the engine enforces.

Codes are suggestions; the clinician is always the final arbiter.
Trajectory-qualifying codes are flagged distinctly from history-based codes.
Never codes off a single out-of-range lab result.

Observability

Recent incidents & retrains

Automatic monitoring — warn / critical thresholds on calibration, drift, and latency.

Specificity bias plateauwarnCoding Engine14/04/2026, 09:20
Specificity has plateaued at 0.81 for three weeks. Candidate retraining slot next scheduled 2026-04-28.

All engines