It was originally 1000Hz, but it turns out this could cause framing errors on D4 trunks because the framing recovery circuit could misidentify the bit pattern a

It was originally 1000Hz, but it turns out this could cause framing errors on D4 trunks because the framing recovery circuit could misidentify the bit pattern as valid framing bits. So they changed it to 1004Hz, enough to sidestep the problem, but still close enough that it would still represent "exactly" a milliwatt of energy, within the required precision of the instruments anyway. When ESF framing was invented, precise tones no longer posed a problem, but a lot of D4 trunks hung around for decades, so last I checked most milliwatt tones were still 1004Hz.