I've analysed historic frequency data on
the National Grid website, which gives a frequency reading every second, to work out the deviation in 'grid time' error over the last two years.
I found a blog where some horologists visited the NG Control Room and there's a photo of a screen showing the time error as being -4 seconds at 17:20 on 11/6/16.
Using Excel and the NG frequency data, I've compared how many seconds a clock
actually 'counted' since then and the number of seconds it
should have counted.
I've checked up to Jan '18 so far and the grid time error was as follows at these points:
mid-Sep '16: -67 seconds
mid-Dec '16: -76s
mid-Mar '17: -133s
mid-Jun '17: -205s
mid-Sep '17: -256s
mid-Dec '17: -224s
I was surprised at how 'out' it was and doubted my calculations, but the average frequency over the entire period is slightly below 50, implying the slowdown shown is correct. I note it starts to correct itself at the end of 2017.
Also, in Jun '17, I tweeted the NG control room and asked what their current grid time error was and they said -7
minutes! They said it was unusually high, but that doesn't tally with my calculations above. I wonder if they gave me the wrong reading.