The current tests are:

 if ((overscanReadNoise - readNoise)/readNoise > 0.05)

# and

verify['NOISE'] = bool(np.abs(stats['NOISE'] - readNoise)/readNoise <= 0.05)}}

We debated if we should fail if the measured noise is smaller than the read noise, and decided we should, because that means something is seriously wrong (and is what DMTN-101 states). That justifies the second np.abs. If the overscan read noise is lower (the case that came up), then we get NOISE=False, CONSISTENT=True. If the overscan RN is high, then we get CONSISTENT=False and possibly NOISE=True (we measured the higher read noise).

It would be more understandable if the CONSISTENT check had the absolute value as well, so we note when the quoted read noise is lower than what is observed from the ovrescan.