You've basically come up with four criteria that describe the use of the word "signal" in a highly specific context - traits that exist for pure signalling purposes in evolution or game theory - and then decided, arbitrarily, that this is the one true meaning of "signal." I do not think you have provided adequate evidence or argument to back this claim up.

If everyone around me is a Republican and I am not, it might make sense that I would do things that would signal that I am a Republican, even if these are very cheap and have obvious positive returns. Your definition would not allow this - if it is cheap and has obvious positive returns, it is not "signaling" to you. What you're saying is that if I send a birthday card to a coworker I hate, then I am not "signaling" that I like that person because it's too cheap to send the card.

It may make sense to speak of weak or strong signals, or reliable or unreliable or misleading signals. But you've arbitrarily said that the word applies only when a certain arbitrary threshold is crossed (your 2 and 4).

Incidentally, your theory might actually work if 4 were eliminated and 2 read "the behaviour is more likely to occur if you possess a certain characteristic than if you do not." This would cover my birthday card example - it's cheap, but I'm more likely to do it if I like the person, so it does signal liking the person. But this change would also fix the counter-productive manager. She's doing things that she is more likely to do if she is decisive and in charge. Since she's being evaluated on those criteria, and not "good manager-ness" - which is not generally observable - it would make sense that she would choose to give those signals rather than not. But revising the theory appropriately seems to nullify most or all of your objections.