All I’ve read so far is the BBC article about how the pilot at 5 councils of electronic counting has failed. That and some of the more tedious leglislation on the matter.
So having not found out much yet I predict the following will turn out to be key factors in the failure of the electronic counting pilot:
Final hardware not specified (specification of hardware changed throughout project life-cycle).
No full-scale tests. That is, no test using a similar number of votes as would be expected on election night.
No test performed using the final hardware and software at the site where counting was to be done using real counting staff (until the real thing on election night, naturally).
Nobody opened an image-processing textbook (see secret software).
Results of pre-election night tests suppressed.
Requirements for error rates not specified numerically (in quantifiable terms).
At least one class of errors (in categorising votes) not considered.
Requirements on speed of processing not specified numerically.
Requirements on number of votes presented for adjudication (compared to votes adjudicated in a hand count) not specified.
Changing the design of the ballot paper so that electronic voting would be “easier” meant that electronic counting systems could not be tested with the actual voting slips from previous elections.
Inadequate staff training (see power cycling).
Concurrency problems (for example, presenting the same image to two different adjudicators, then screwing up the count database).