Failure is a condition.
Retry is a decision.

What the policy means

The waiting is intentional.

Local backoff doubles after each retryable failure, up to its cap. Equal jitter draws from the upper half of that window; full jitter uses the whole window. Both have a 1 ms floor.

A server delay is a lower bound

The actual wait is the larger of local backoff and Retry-After. A local cap never shortens that server delay. If a deadline arrives first, the operation stops.

Bounded, inspectable, reproducible

The same policy, fixture and seed produce the same trace. No real sleeping, recursive retry calls, or background title threads are needed.

An honest continuation

This is the maintained successor to the historical tiktodv4 fork. The original automation is preserved as inert source with upstream credit. This lab does not send engagement, bypass challenges, or evaluate a live service.

Retry-After supports integer seconds and preferred IMF-fixdate syntax. This teaching simulator is not a complete HTTP client.