Not quite every time you'd bullet vote for A in an approval vote! You don't need a belief that the contest is currently A = B = C or A > B = C in order to motivate a bullet vote.

There are basically thirteen possibilities for expected results if we classify them by ">" and "=" expectations (expecting a lead vs not certain at all). We only expect second-Borda-rank-O ballots from those two specific types of expectations.

Strategic approval bullet-voting is clearly motivated if the expectation is:

A > B > C

B > A > C

A > C > B

A = B > C

B > A = C

A > B = C

The general bullet voting heuristic for approval voting is to approve of the candidate polling in 1st or 2nd that you like the best of those two, and then any candidates you prefer to that candidate.

On the merits, if the expectation is A = C > B, A = B = C, C > A > B, and C > A = B, strategic bullet voting has a real chance of backfiring. It follows the basic bullet-voting heuristic, but those are the cases in which you *should* see sincere voting on the merits of uncertainty.

Then in the three cases where A is expected to place third, you should expect strategic double-voting even if the voter does not really like B that much.

Out of those thirteen classes of expectations, you should expect bullet voting in at least six, and behaviorally perhaps as many as ten classes.