Inkwell - Archive Digitisation Platform with Handwriting Recognition and Volunteer Correction

Dashboard, collections, collection detail with 6 tabs, the transcription queue with 4 tabs and the line-by-line page editor, plus panels for boxes, line history, closures and new capture batches.

Awaiting conservation, ready to capture, captured, QC passed, recognised, corrected, reviewed, described and published, each with a page count and amber when its median age passes the target.

Line boxes on the page image turn amber below 0.80 and red below 0.60 (client policy). On the sample baptism page, 8 of 13 lines fall below the threshold and 11 entities are tagged.

2,000,000 estimated pages across 2 institutions, 677,020 captured, 25 queue pages, 5 volunteer projects, 5 HTR models, 8 closures with their sources and 11 staff users.

Machines Read Old Hands Unevenly

Handwriting recognition does well on Victorian copperplate and badly on 16th-century court hand. Without a confidence score per line, staff could not tell which pages needed a person and which only needed a check.

Living People in Old Registers

The GDPR generally does not cover the deceased, but a 1930s register can name someone still alive, and parish registers reveal religious belief. Closures come from depositor conditions, diocesan rules and client policy, and closed pages must never reach the public site or a copy order.

Volunteer Work Nobody Reviewed

Volunteer corrections had no review, no record of who changed what and no licence attached. Every correction now needs a second pair of eyes, a version history and the contribution licence the volunteer agreed to.

Dates, Places and Description Standards

Ireland moved to the Gregorian calendar in 1752 and earlier years began on 25 March, so dates must be kept as written and normalised. Places must match real townlands, and descriptions have to follow ISAD(G) and export as EAD.