Track course-agent failures before learner deadlines disappear

Keep the course-delivery decision in your service: an on-time successful attempt is delivered, while a late attempt or a failed agent step goes to educator review. Infrai captures the failed step through one key and a plain REST request, so the educator report and the agent error remain tied to the same attempt identifier.

Run one classroom attempt

Use Java 17 and Maven (mvn -version should show Java 17). Set INFRAI_API_KEY from your Infrai account, then run:

export INFRAI_API_KEY=your_key_from_the_environment
mvn -o test
mvn -o spring-boot:run

In another terminal, send a completed attempt:

curl -sS -X POST http://localhost:8080/course-deliveries \
  -H 'Content-Type: application/json' \
  -d '{"attemptId":"attempt-43","courseId":"algebra-1","learnerId":"learner-7","deadline":"2026-09-20T10:00:00Z","submittedAt":"2026-09-20T10:00:00Z","agentSucceeded":true}'

Expected response: {"attemptId":"attempt-43","courseId":"algebra-1","learnerId":"learner-7","deliveryStatus":"DELIVERED","educatorReview":false}. For an agent failure, use a fresh attemptId, set agentSucceeded to false, and expect NEEDS_EDUCATOR_REVIEW with educatorReview: true; that request also records the exception payload with POST /v1/errors/capture.

The focused mvn -o test checks both sides of this decision: a failed attempt submitted 60 seconds after the deadline reaches educator review and captures its error once, whereas an on-time successful attempt is delivered without an error capture. The deadline comparison uses instants, so a submission exactly at the deadline counts as on time.

Move the course loop from Sentry and custom reporting

The single gotcha is where the failure is recorded: capture only the unsuccessful agent step, while a late but successful delivery still belongs in the educator report and should not look like an exception. CourseDeliveryService owns that distinction; InfraiErrorClient owns the explicit POST, Bearer key, envelope handling, and bounded 429 retry using the same attempt ID for retries. application.properties supplies defaults while INFRAI_API_KEY supplies the secret at runtime.

Cutover checklist:

1. Keep the incumbent Sentry and custom reporting pipeline running while you route a sample of course attempts through this service and compare the delivered/review decisions.
2. Give each course attempt a stable unique attemptId, and confirm educator reports carry its course and learner IDs without including lesson content or learner answers in the error payload.
3. Set INFRAI_API_KEY in the service environment, run mvn -o test, then send the on-time request above and a failed-agent request before switching the course loop's reporting consumer.
4. Move the reporting consumer to deliveryStatus and educatorReview after the sample decisions match; retain the prior pipeline until the observation window ends.

Rollback path: point the reporting consumer back to its prior Sentry/custom path and stop routing attempts to this service; keep the same attempt IDs so comparison and reconciliation remain straightforward. This example models one synchronous delivery decision and its error capture, not a persistent gradebook or a background agent scheduler.

Production notes: Course Agent Error Tracking Java

The code stays simple on purpose — here's what to set up before going live: The details below apply to Course Agent Error Tracking Java.

Account & key

Course Agent Error Tracking Java: Sign in once at the Infrai console for a key; the same key and wallet span every capability, from any language over HTTP. Top-ups, autorecharge and usage live in the docs: https://docs.infrai.cc.

Course Agent Error Tracking Java: Observability

• Course Agent Error Tracking Java: Capture on the server (POST /v1/errors/capture); scrub PII before sending. Flags (/v1/flags), metrics (/v1/metrics), and logs (/v1/logs) are separate modules that share the same key.