Here are the tactical voting strategies. All of them are based on a combination of voter preferences, and a ranking of candidates that indicates that voter's belief about the most likely winners.

Plurality: Vote for your favorite among the top two most likely winners.

IRV: Vote honestly, except that if you prefer both of the expected 2nd and 3rd place winners to the expected 1st place winner, always vote for the expected 3rd place winner since you expect the 2nd place winner to lose head-to-head with the 1st place winner.

Borda Count: Among the top two likely winners, rank your preferred candidate 1st, and your less preferred candidate last. Otherwise, vote honestly.

Approval: Approve of your preferred candidate among the two most likely winners, and anyone else you like more than that.

Range: Rate your preferred candidate among the two most likely winners at the maximum possible score, along with anyone you like more than that. Rate your less preferred of the two at the lowest possible score, along with anyone you like less than that. Rescale any remaining candidates honestly between those extremes.

STAR: Same as range, except average in your honest preferences with a very low weight, just so that your ranking is the honest one for the final runoff round.

Condorcet: You should ignore tactical results here, as I made a mistake. I'll fix it when I have some time.

I did try a few things and validate that these strategies improve expected outcomes (or at least don't make them worse!) for specific subpopulations if they vote this way, and also that if everyone votes this way except a specific subpopulation, that subpopulation sees a worse expected outcome. That doesn't mean these are the *optimal* tactical voting strategies, but they are better.