In the old days it was pretty vague. Nowadays we have definite limits. -43 dB IIRC. K9STH has written extensively on the Collins audio-injection-SSB-to-make-CW problem. From what I have learned from him, it came down to this: The Collins rigs which used that system (KWM-2, 32S-1) could be carefully aligned to produce a single clean CW signal. But they wouldn't stay in alignment long! Within a relatively short time (weeks?) the rig would be making at least three, and often more, CW signals where there should only be one. And the spurs weren't that far down! It's important to consider the spirit and intent of the regs, not just the letter. The FCC did not, and does not, expect that every amateur signal will be absolutely perfect, but they DO expect signal quality consistent with good engineering practice. The Collins 32S-1 and KWM-2 were mid-to-late-1950s technology. In those days, CW transmitters with good signal quality were all over the place; there was NO EXCUSE for the Collins problems. Not when transmitters costing less than 1/10th what the Collins did could and did produce clean signals. Collins eventually fixed the issue in the 32S-3, by the simple method of using a separate CW crystal for transmitting.