Measure two-reader agreement on handwriting

The study has one rule that everything else serves: the two readers must not see each other, and the comparison happens only after both have finished. Each reader writes its own CSV of readings, and the analysis joins them with the labels afterwards. The decision that matters most is normalising before comparing. "29/09/1987" and "29 September 1987" are the same answer; "7/11/2019" and "11/7/2019" are not. Without per-kind normalisation, agreement measures formatting, not reading.