The error value that is supplied does not presuppose any particular frequency of operation. A 1E-6 value (plus or minus) indicates that your error can be up to one part per million or 1ppm. A crystal oscillator alone would typically have, as Tim says, something like 0.5ppm (5E-7) to 5ppm (5E-6) stability. An ovenized oscillator (OCXO) might be 0.5ppm down to 0.01ppm (1E-8), also written as 10ppb or parts per billion. On a bad day, you might see 100 parts per TRILLION out of this GPSDO or 1E-10.
Now what does this mean? If you grab your scientific calculator and punch in the value from the GPSDO and multiply it by your frequency in Hz, it will tell you how far off you could be in Hz at that frequency. So let's say David is operating on 10m at 28.2MHz, taking the two different numbers provided at different times provided from his GPSDO, we would multiply and get:
28.2 E 6 x 1.25 E -12 = .000035Hz
28.2 E 6 x 9.51 E -12 = .000268Hz
(The E6 is one million for MHz to Hz conversion) So the first number is better -- it's a smaller error. But heck, the second number says that on 10m you are only off by 268 MICROHERTZ ... That's pretty good so from a frequency precision perspective you're in great shape either way.
The really great thing is that with a GPSDO, you never drift over time. You will always be on frequency, even 10 years from now, with no adjustment necessary.