Is multi-layer ESCAPE expected?

Question

We design an application, containing a grid with an active filter search and inline-editable columns. Each row represents a record, of which multiple fields are editable. The application should be fully handled by keyboard (rapid data manipulation, mostly numbers).

[_Search___________[X]

________________________________________
| COLUMN 1    | COLUMN 2  | COLUMN 3   |
|_____________|___________|____________|
| Some value  |    10,000 |      20,000|
+-------------+-----------+------------+
| [editing__] |   [__5000]|    [__4000]|
+-------------+-----------+------------+
| Some other  |      3000 |        3000|
|_____________|___________|____________|

We have a problem deciding on behaviour of the ESC key. We have basically the following questions:

• if user hits ESC while editing the second field, does they expect to cancel only that field, or, if previous edits have been made, all other fields edited in the current record?
• which would create less mistakes due to unexpected behaviour: edited fields remaining there despite intended to be canceled or canceled fields which were supposed to retain value?
• shall we use double ESC for reseting row?
• shall we use triple ESC for getting out of the active filter of a search?
• would multi-esc conflict with natural flee behaviour ("ESC-ESC-ESC, I wanna get out of here!")

Are there any researches / widespread applications dealing with this multi-layer cancel problem?

Or in case not, how should a test be set up to acquire behavioural (not self-reporting) results?

Answer 1

Escape should cancel what can clearly be considered as 'the current active task'. It essentially means stop but also quit, exit, cancel or abort. It's used to stop an edit without change or close a transient dialog without altering data.

Escape should never be considered a sequential historical operation. It is not an alternative, replacement or substitute for an Undo operation (eg Ctrl+Z) which would be the correct way to revert multiple sequential actions without aborting the larger task at hand.

To answer the questions specifically as asked:

if user hits ESC while editing the second field, does they expect to cancel only that field, or, if previous edits have been made, all other fields edited in the current record?

Only that field, as that is the active task

which would create less mistakes due to unexpected behaviour: edited fields remaining there despite intended to be canceled or canceled fields which were supposed to retain value?

The first of those two options will create less mistakes, because users are expecting ESC to stop the edit, not undo previous edits.

shall we use double ESC for reseting row?

No, no

shall we use triple ESC for getting out of the active filter of a search?

No, no, no!

would multi-esc conflict with natural flee behaviour ("ESC-ESC-ESC, I wanna get out of here!")

It so totally would!

Is multi-layer ESCAPE expected?

That's a no.