Measurement of reconstructed carrier from the 31.5 hour message,
In approximately 8 hour blocks (34.7uHz):

 From       To         Signal    S/N      Phase  Noise
 16th 21:00 17th 05:00 0.077 fT  11.0 dB  112.7  0.022  fT
 17th 05:00 17th 13:00 0.063 fT  17.4 dB   75.2  0.0085 fT
 17th 13:00 17th 21:00 0.041 fT   9.2 dB   83.5  0.0143 fT
 17th 21:00 18th 04:30 0.05  fT   3.8 dB  125.0  0.032  fT

Day/night phase shift up to 50 degrees which is about what was
expected, although this is a long average so it could be varying
more on shorter timescales to make the signal apparently weaker.

The 18th from 00:00 onwards was noisy.

Overall, would have needed to average about 14dB in this
bandwidth (34.7 uHz) to have a reasonable chance (~50%)
of decode.

Daytime is doing better than nighttime. In 17.36 uHz we have

 17th 04:00 to 20:00 0.057 fT  18.2 dB

 (eg 11 chars, 72 seconds, Eb/N0 = +0.23 dB)

Or, in 34.7 uHz

 17th 05:00 to 13:00 0.063 fT  17.4 dB

 (eg 9 chars, 40 seconds, Eb/N0 = +0.2 dB)

Conclusion is that it's not the phase that's blocking these
long messages, it's the poor nighttime S/N.

--
Paul Nicholson
--