Quoting Dave Mayall <david.mayall@ukonline.co.uk>: > If anybody thinks they can produce better figures from the data, they are > welcome to try! Dave, I think you've clearly shown that it's very difficult to produce better figures from THAT data. My question is whether the build process could collect better data from which better figures could be produced. I think it would be possible to do a better job at the page level. For each event-quarter we know how may pages there should be in total. If the build process can record how many unique pages in an event-quarter have been transcribed at least once (or how many have not been touched at all), then the ratio would give a very good estimate of the coverage irrespective of the amount of double keying. I assume you must have considered this and that there's a gotcha somewhere. Martin Cope (Posted earlier today to Admins list in error)