In the end what got me to thinking in n=3 terms was thinking of it in terms of musical meter – bars in triple time (’1,2,3,1,2,3′ – think of it as a slow waltz), with an imaginary heavy accent on the first beat of the bar (on 1). That helped serve as a marker so I wouldn’t lose my place.

Whilst holding the pattern of 3 in my head, when the new set came along I would actively mentally ‘predict’ the next move – eg if the last set had a C on the first beat in the top left hand corner, my mind would actively go to that spot. If it matched the thought in my mind I hit the A or L key as appropriate.

If it was different, I would mentally substitute that new letter/square for the last one in that slot in the pattern. In that way I would keep a pattern of 3 in my head at all times, and then as it changed, I would change it gradually one by one, holding the other parts of the pattern in my head until they weren’t needed and could be replaced by something else.

It sounds more complicated to describe than to do. But it was a lot easier than trying to mentally count back 3 spaces.

The only drawback I find with this is that sometimes the voices seem to have a heavier stress on some letters, and that can throw you a bit if the stressed letter doesn’t naturally occur on the dominant beat, especially when you get to n=5, because groupings of 5 are not common in Western music and a natural stress on a letter by the reader can tend to confound it more. It’s much easier when all the letters are read in the same tone (without undue stress on one letter), because I can mentally add my own.

I’ve had a couple of tries at n=6, not terribly successfully – I got bounced back to n=5 each time. But the beauty of using the musical meter technique is that theoretically n=3 should serve as a good preparation for n=6 and n=9, and the same technique for n=4 should serve for n=8, the difference being that instead of holding and manipulating one pattern of 3 or 4 in your head, you have to juggle 2. It should have a similar effect to ‘chaining’ techniques in digit span exercises (where you group the numbers into groups of 2, 3, or 4 instead of thinking of a single long string).

I haven’t had much chance to test that theory fully yet at those levels, but I think it should work.