I am not sure I follow that. If you have more than 1 version of the watermarked work (i.e. one with watermark A, another with watermark B, and a third with watermark C) you should be able to diff them to locate the watermark scheme, and take countermeasures, right? Or am I misunderstanding? Are you assuming a single version of the watermarked work?
Version A: 1234AA567890
Version B: 1234BB567890
Version C: 1234567CC890
If versions like A & B wouldn’t you simply zero or average the bits* at the common address?
If versions like A & C wouldn’t you simply look for discontinuities in otherwise identical strings, and once again zero or average the bits?
* Or something similar depending on encoding scheme of media and details about the pattern of watermarking.
