Pipeline Observability: Data Outputs¶
Every time PressureAI runs, it produces two structured data outputs: a Clinical Audit Record and an Operational Telemetry log. These outputs support clinical governance, model performance monitoring, and outcome validation, giving your organisation a comprehensive, continuously updated picture of how the model is performing in your environment.
| Log stream | What it records | Contains patient data |
|---|---|---|
| Clinical Audit Record | A complete record of every risk prediction made for every patient | Yes — handled as sensitive clinical data |
| Operational Telemetry | Pipeline performance, timing, and patient volume metrics | No |
Both outputs are written automatically at runtime. No manual steps are required to generate or maintain them.
1. Clinical Audit Record¶
2. Operational Telemetry¶
PressureAI writes a separate operational telemetry log that records pipeline performance and system health. This log is distinct from the Clinical Audit Record and contains no patient-identifiable information — no PHI or PII of any kind.
3. Data Governance¶
Storage and access¶
Log file locations are configured during deployment. Both outputs are stored as newline-delimited JSON files (one record per line), making them straightforward to ingest into existing log management, SIEM, or analytics platforms.
PII handling¶
- The Clinical Audit Record contains patient-identifiable data (patient ID, age, gender, ethnicity) and must be subject to your organisation's standard information governance controls for clinical systems.
- The Operational Telemetry log contains no patient-identifiable information and may be handled under standard IT operational data policies.
Completeness guarantee¶
The audit record is written for every patient in every run, regardless of whether a prediction was successfully produced. If an error occurs during processing, a record is still written capturing the patient ID, prediction ID, run timestamp, and the nature of the error.
Model versioning¶
Every record in the audit log includes the version of the model that produced it. This enables full traceability of predictions to specific model releases and supports impact assessment when models are updated.