Monitoring

Heimdall observability spans the frontend, auth, and predict services.

In-app surfaces

Surface	Path	Purpose
Data	/data	Bronze, silver, and gold catalog — where governed datasets live before training
Usage	/usage	Billing summary and data platform usage
Health	/health	Polls each service root endpoint
Dashboard	/dashboard	Unified projects hub (models, forecasts, forge, loops)

PostHog product funnels

Client events (heimdall-frontend-next):

• forge_started — user begins a forge upload
• forge_completed — forge job reaches completed
• model_built — ML build wizard completes
• first_prediction — first successful test prediction on a model page

Predict service (I/O frozen)

External inference routes are unchanged. Internal observability is optional:

• Set OTEL_ENABLED=true and OTEL_EXPORTER_OTLP_ENDPOINT in heimdall-predict/app/.env
• Spans: predict_single, run_forecast (no PII in attributes)
• GET /health for load balancers

Before any predict deploy, run:

cd heimdall-predict && pytest tests/ -q

See heimdall-predict README for the full I/O freeze policy.

Deploy pinning

Production compose pins joelreji/heimdall:predictservice with pull_policy: if_not_present. Bump the image only after contract tests pass.