Module 09 Activity

Scenario

The provider will rate limit you, time out on you, and go down. Decide now what happens.

What you build

A retry policy that classifies failures, backoff with jitter and a ceiling, a degradation ladder, and distinguishable error messages.

Steps

1. Classify every failure mode as retryable or not, and say why.
2. Implement exponential backoff with jitter and a hard retry ceiling.
3. Handle a timeout as ambiguous: retry with an idempotency key rather than assuming failure.
4. Build at least three rungs of the degradation ladder, including returning source text without a generated answer.
5. Write five distinguishable error messages containing no internals.
6. Give each one a request id the user can quote.

Evidence to hand in

• The classification with reasons.
• The backoff schedule with jitter and ceiling.
• The timeout handling with its key.
• Three degradation rungs.
• Five error messages.

Review checklist

• A 400 is not retried.
• Jitter is present and the ceiling is bounded.
• The timeout path retries with an idempotency key.
• The retrieval-only fallback exists.
• No message contains an exception, model name or prompt version.

Mark complete