This page shows how much of the work is done and how much is waiting. The published records are the part you can read. Behind them sits a much larger pile of scanned books at every stage: found, photographed, read by machine, mined for names, checked. The size of that pile is the honest measure of this project. It is also an invitation, because the slowest step is careful reading, and anyone careful can share it.
Published, and waiting
The two record counts below are counted fresh every time this page loads, by the same count the front page uses. The two pages cannot disagree.
- 605maker records published
- Every line on them names the document it rests on, and every one of those can be opened and read. This is the same count the front page states, taken the same way at the moment you loaded the page.
- 4,329records in progress
- Each is waiting for its lines to be read against the pages they cite. Nothing a draft says is shown here or anywhere else on the site until that has happened: where one has a page at all, it gives the name and which books print it and nothing more. A further 805 were retired in place or merged into another record. They are on file, not waiting, and not deleted.
Behind them, the pile
The stage figures below are a different kind of number. Some of them count thousands of files on disk, so they are read from one snapshot of the working files rather than recounted for every visitor. The snapshot says when it was taken, immediately below, and its last two stages give the record counts as they stood at that moment rather than now. If a figure cannot be read, the page shows a dash and says which one failed. It never shows a guess.
the progress snapshot has not been written on this site So every figure on this page shows a dash instead of a guess.
The pipeline, stage by stage
A scanned book becomes a published record in these stages, in this order. Material thins out at every step. These are the counts at each stage right now.
Books are found and weighed
— scanned books have been looked at in public online libraries. Most are city and trade directories, the yearly books that listed every tradesman in a town. Most do not make the cut. — were set aside as unlikely to name a single maker. — are blocked, usually because their copyright status is unclear. — are queued to be fetched, and — are located and next in line.
Their pages are fetched as photos
— books are held as photos of their pages. Only the pages likely to matter are taken. Nothing is ever typed up unless we hold the picture first, so any later reading can be checked against the page itself.
The photos are read by machine
— of those books have been read by software that turns a photo of a page into text. Each page is read more than once. Where the readings disagree, both are kept, and nothing picks a winner.
Names are mined out of the text
— books have been combed for the names of violin and bow makers, dealers and repairers. A name found this way is a lead, not a fact.
Each name waits in a queue to be judged
— names are waiting to be judged. So far — have been accepted, which means only that the name really is printed in that book on that page. — were rejected as misreadings or off topic. No machine ever decides that two mentions are the same person.
Accepted names become draft records
— maker records are in draft. Each says only what its sources say, often a single directory line. None of them is public. What a draft says stays hidden until every line on it has been read against the page it cites.
A record is published once every line has been checked
— records have passed the whole chain and are public, every line carrying the document it rests on. “Checked” means the line was found in the document it cites and the answer written down — and most of those answers were written by software matching the line against the machine text of the page, character for character, not by a person looking at the page image. A record publishes itself only where that match succeeds for every line on it and nothing about the person’s identity is in question; where either fails, it waits for a person. This is the narrow end of the funnel, and it is meant to be.
So the backlog is not just a number, here is a little of it by name. Each name below has a draft record: somewhere in a scanned book, a page names this person. Until someone checks that draft against the page, the site shows the name and nothing else. The list changes as the work moves. Once a record is checked, it can publish, and its name comes off this list.
Waiting in draft: a sample
the progress snapshot has not been written on this site No names are listed, rather than an old set.
How current the public data files are
The one figure here that is about the site itself: when the files you can download last caught up with the working copy.
The records are worked on here and copied out to the public data files in batches. There is no live feed yet. The files were last exported on —. Since then, — maker records have changed in the working copy, and those changes reach the public files at the next export.
How to help
Every draft above is waiting on one question: does the page really print what the record says it does? Answering it takes patience and a suspicious eye, nothing more. The first step is one email to info@bowandbridge.com saying you would like to check records. The contribute page has the details.
Nothing a draft says is shown here or anywhere else on the site until it has been checked. That rule does not bend for a progress report. And if you find a published line that its source does not support, that is the most useful thing you can send. It becomes a logged correction on the record itself.