Five stages from the box to the search bar
Assess, Capture, Read, Catalog, Return. At every stage you can see what happened to the originals, what the records gained and what is waiting for your decision.
Every project we take on, from one pilot box to a whole collection, moves through the same five stages: Assess, Capture, Read, Catalog and Return. The names stay the same in our proposals, our updates and our reports, so you always know where a box is and what it is waiting for. A pilot box is simply these five stages run on a small batch of your real material first, so you can see the images, the text, the records and the measured error rates before you commit anything larger.
The five stages at a glance
| Stage | What happens to the originals | What the records gain | What you review or approve |
|---|---|---|---|
| 1 Assess | Counted, measured and condition noted, usually where they live | Inventory, priorities, a capture specification and a dated plan | The plan, the specification and the pilot box selection |
| 2 Capture | Imaged by hand, kept in original order, at the studio or on site | Preservation masters, access copies, a capture log and checksums | Sample images from the first batch |
| 3 Read | Back in their boxes; no further handling needed | Page text with an error rate measured per batch | The correction level for each batch |
| 4 Catalog | No handling | Records with names, places, dates and subjects, each field labeled | Vocabularies, sample records and any fields you review yourself |
| 5 Return | Returned in original order against a signed receipt | Files delivered, a search site or an import, and a preservation package | The handover checklist and the return receipt |
1. Assess
Before anything moves, we look at the collection where it is kept. We count it by series and format, measure anything oversize, and note condition with the same points as our condition assessment checklist: brittle paper, tape, mold, bindings that will not open far, glass plates, film that smells of vinegar. We ask what people come looking for, and we read whatever inventory or finding aid exists, however rough.
The assessment produces an inventory, priorities and a capture specification, which for each kind of material names the category in the FADGI Technical Guidelines (third edition, 2023), the target quality level, the sampling resolution and the file formats. It also sets the file naming scheme and maps our record fields to the fields in your collection system. You approve the plan and the specification in writing, and together we choose the pilot box, which should include your hardest material, not your cleanest. See collection assessment.
2. Capture
Originals are imaged by hand, one folder open at a time, in the order they arrived. Bound volumes are opened only as far as they open without strain and are supported while they are imaged; nothing is forced flat. Fronts and backs are both captured wherever the back carries writing, a stamp or a studio imprint. Each session starts with a color and scale target, so images are checked against the specification rather than judged by eye.
The records gain preservation masters (usually TIFF), access copies for everyday use, and a capture log that ties every file to its session and settings. Checksums are generated as each master is written.
Quality checks happen while the originals are still here. Every image is looked at for completeness, focus, glare, cropping and order. Target images are measured against the tolerances for the agreed level, and a random sample of each batch is checked in more detail (the QC sample size calculator shows how large that sample needs to be). Defects are logged and recaptured. Before capture continues past the first batch, you review sample images and tell us if anything should change.
3. Read
Text recognition needs only the images, so originals can go home early if agreed. We run optical character recognition (OCR) on print and typescript and handwritten text recognition (HTR) on manuscripts, then measure the result against pages a person has transcribed exactly. That gives a character error rate and a word error rate for each kind of page: print, typescript and each distinct hand.
With those numbers you choose the correction level for each batch: full correction, targeted correction of names, places, dates and numbers, or uncorrected text clearly labeled as such. After correction, a fresh sample is measured again. The records gain plain text and ALTO XML for every page, with word positions so a viewer can highlight search hits on the image. The text recognition page and the OCR accuracy calculator explain the arithmetic.
4. Catalog
Models read the text and images and suggest names, places, dates, subjects and a short summary for each item. A cataloger then accepts, edits or rejects each suggestion against the item itself and the vocabularies you have chosen. Every field carries a label saying whether it is machine-suggested, reviewed, edited or entered by a person, and those labels stay in the records we deliver. Our AI policy sets out the rules.
You approve the vocabularies and sample records before the batch is cataloged, and your own staff or volunteers can review some fields if you prefer. We count how many suggestions were accepted, edited and rejected for each kind of field, so you can see where machine help is saving time and where it is not. Records are mapped to your system’s fields with the metadata mapping worksheet and delivered as CSV, Dublin Core, MARC 21 or EAD, whichever your system imports. See AI cataloging.
5. Return
Return covers four things. The originals come back in their original order, checked against the list that went out and the photographs taken before they left, with a receipt signed at handover. The files are delivered: masters, access copies, text and records. If publishing is in scope, the records go into a search site or are imported into your own system. And the whole set is packaged for preservation as BagIt bags with checksum manifests and an inventory, as described on the digital preservation page.
Before you sign off, checksums are verified again on your storage, and a handover checklist records what was delivered and where each copy now lives.
Where the work happens
Most collections can come to our studio in Colorado Springs. Some should not travel: material that is too fragile, too large, under donor restrictions, or simply too much to move. For those we can work on site when originals cannot travel, to the same specification and with the same checks. What we need from you on site is a room that can be closed, a clear table and power. How originals are packed, moved and receipted is set out on our collection care page.
How we keep you informed
At the start we agree one contact on each side and how often you want to hear from us: usually a short written note when a batch changes stage, and a conversation at each decision point. When we need a decision, we send what you need to make it: the sample images, the error rates or the sample records, not just a question.
Nothing passes a decision point without your approval. Each batch ends with a short report: items and images, defects found and recaptured, error rates, how many fields were reviewed, and anything unexpected.
When something unexpected turns up
Boxes hold surprises: a strip of cellulose nitrate film, a moldy album, letters that are not on the inventory, a folder marked restricted. When that happens we stop work on the item, tell you, and agree in writing what to do before going on. The plan can change; the order of your originals and the record of what happened to them do not.
To see all five stages applied to one box, read the fictional example collection. To compare the ways a project can be scoped, see engagements.
Send one box first.
A pilot box shows you the images, the recognized text and the catalog records before you commit a whole collection. Tell us what is in the box and what you want people to find.
Studio and contacts
24 South Weber Street, Suite 225Colorado Springs, CO 80903(719) 244-8450sales@dastok.com