Because of a power outage, instance failure has occurred. From what point in the redo log does recovery begin and where does it end?

Describing recovery as spanning from the current redo log to an inactive redo log confuses redo log file activity states with the actual recovery range, which is defined by the checkpoint SCN position.

During crash recovery, Oracle uses the checkpoint position - the SCN at which all prior committed changes are guaranteed to be written to the datafiles - as the starting point for redo application. Recovery then rolls forward through all redo log entries up to the last record written before the failure, restoring all in-memory changes that had not yet been written to the datafiles at the time of the outage.

Beginning recovery from the very start of the redo log is incorrect because all changes before the checkpoint position are already safely persisted to the datafiles and do not need to be reapplied.

Recovery does not stop at the last committed transaction - Oracle rolls forward all redo including uncommitted changes, then performs a separate rollback phase to undo uncommitted transactions using undo segments.

This reverses the correct direction: recovery proceeds from the checkpoint position forward to the end of the redo log, not from the beginning of the redo log backward to the checkpoint position.