The batteries were reporting 85% but replacing them fixed the missing samples and repeated log file headers.
Whilst there’s no schematic publicly available, I’ve had a close look at my Mooshimeter and I can’t find any bulk bypass capacitors. My guess is that the design assumes that batteries will always have a very low ESR, which isn’t true. The actual SD card writes are relatively power intensive and could easily cause brownouts if the batteries are feeling a bit poorly. I might consider adding a capacitor across the battery terminals and see if that helps.
I’m still seeing about some time drift – about 2ms/s with 1s logging. That’s well outside what I’d expect even with really cheap crystals and I wonder if there’s a bug in the timekeeping code.
And I’m still seeing a delay of several minutes between enabling logging and the first entry in the logfile.
That said, I’m now confident that I have a complete enough set of logs and can correlate them with logs from the data generation I was doing.