In SEPTEMBER, 17 out of 440 trains scheduled to call at Melksham were cancelled - that^s 3.86%
In OCTOBER, 10 out of 472 trains scheduled to call at Melksham were cancelled - that^s 2.11%
from 1st to 24th NOVEMBER, 9 out of 348 trains scheduled to call at Melksham were cancelled - that^s 2.58%

HOWEVER, the RecentTImes logging system doesn^t have a record point at Melksham and so considers that a train that calls at Trowbridge and at Melksham has also called at Melksham. To my knowledge, 4 trains in October and (so far) 3 trains in November have been diverted due to signalling problems via Batheaston, so are show as having run when in fact they were unable to make the Melksham call.  That^s good as cancelled for people wanting to leave / join there!.    Taking these extra cancellations into account, that^s 17, 14 1and 12 cancelled in the 4 periods, being 3.86%, 2.96% and 3.44% of services.

The vast majority of cancellations are described as being due to signalling problems, and it seems that once a signalling problem is reported it can take a considerable time to fix it.   The line was closed for nearly 2 days in September due to this, on our count days we had diversions (and late running trains with pilot men all day), and the first train didn^t run yesterday until mid afternoon.

Target reliability (not cancelled)  is 99.5% for our area, which comes under Bristol Suburban services - see https://www.firstgreatwestern.co.uk/About-Us/Our-business/Performance ^ and FGW^▸ (First Great Western) claim achieving 99.4% - however, for trains calling at Melksham only 96.58 % - nearly 7 times worse than the target.

As most of the problems seem to lie with Network Rail, I hope that First are
   a) Asking them to make lasting repairs
   b) To repair faults promptly.

It^s my belief, looking at the figures, that if Network Rail faults could be reduced such that they were only responsible for the cancellation of 1 train a month, First would be able to achieve their target. A cutting - even to zero - of faults attributed to First would make no significant difference as they^re not the ones causing the problems.