While TK. coded the SEQ v4+, i encountered a similar problem with “Undoing” recordings on the LoopA.
It’s nearly “impossible” for the sequencer to know when a recording (e.g. in Jam mode) started and when it stopped. I.e. consider recording a few notes, waiting for a few seconds, then recording another few notes. Was this one recording take or were there two takes? What happens if you change the time cursor during recording?
To cover for that, you’d basically need multiple Undo levels or Undo snapshots, but there is not enough RAM on the microcontrollers for that.
So: while undoing sounds easy from a human perspective (“just revert those few bad notes i just played”), it’s a quite hard job for the sequencer! 🙂
But: not all is lost! What works well (also works on the SEQ v4+, just checked) is that you “Copy” a track before you’re recording manually to it. If you don’t like the the recent recording, just push “Paste” and try again. This is the equivalent to manually capturing an Undo snapshot at a defined point in time - and “undoing” when necessary. After a while you’ll do it automatically, basically whenever you’re going for the next potentially risky recording operation.