Deterministic action replay
Deterministic action replay is re-executing a recorded sequence of actions against a live or restored environment, as opposed to visual replay, which renders what was recorded without executing anything. The two answer different questions: visual replay shows what happened; action replay tests whether it happens again.
How it works
Visual replay reconstructs the recorded event stream in a sandboxed player. It aims to show the captured run without sending those actions to the original application. Action replay sends a higher-level action trace through automation against a live or restored system, where divergence can reveal a regression or flaky dependency. rrweb records low-level interactions and page state, which are useful reference evidence, but it does not by itself produce a reliable executable test plan. Correlate the recording with the automation framework's commands and stable element locators. Deterministic re-execution also depends on restoring backend data, time, network responses, feature flags, and randomness, so it works best in environments you control.
Where you encounter it
Agent debugging harnesses, flaky test investigation, and QA infrastructure built on recorded runs.