Plan a pilot box
Menu

Example collection: one box, five stages

The Aster County Historical Society and its Box 3 are invented. The work the box goes through is how a real pilot box runs.

Search the example box

Ten records from a fictional box: letters, ledger pages, cabinet cards, a clipping and a telegram, with their recognized text and catalog fields. Try a word that appears in the text, or one that only matches by meaning.

3 of 10 records match "smelter".

  1. Letter: Ida Okerlund to her sister Marit, ACHS-3-019

    The smelter men walked out on Monday and Walter says the store will carry them on credit until the matter is settled.

  2. Ledger page: Wren Gulch Mercantile day book, page 77, ACHS-3-L077

    Feb. 9. Smelter boarding house, flour 4 sacks, coffee 10 lbs., lamp oil 2 gal., on account.

  3. Clipping: Wren Gulch Courier, notice of the settlement, ACHS-3-N03

    The trouble at the smelter ended Thursday when the company agreed to the old scale. The men return Monday.

Example (fictional). The Aster County Historical Society, the Okerlund family, the town of Wren Gulch, the Wren Gulch Mercantile, the Halvorsen Studio, the Wren Gulch Courier and every record on this page are invented. No real project produced these numbers. They show how our method works and what the files and records look like at the end.

The search above runs over ten records from this box. This page follows the whole box through the five stages: Assess, Capture, Read, Catalog and Return.

The box

Box 3 holds the Okerlund family papers and the Wren Gulch Mercantile day book, 1893 to 1896: letters from Ida Okerlund (35) and Walter Okerlund (6) to Marit Okerlund, cabinet cards, the store’s day book, and a few clippings and a telegram.

Series Items Images
Letters 41 items, 96 pages, 24 envelopes 144 (pages, and envelopes front and back)
Cabinet cards 18 items 36 (front and back)
Store ledger (day book) 1 volume, 212 pages 216 (pages, both covers, spine and label)
Clippings and a telegram 9 items 14 (both sides where both carry text)
Total 69 items 410 images

1. Assess

The society wanted two things: an exhibit on the town in the 1890s, and a way for families to find the Okerlunds without opening every folder. The assessment found the letters folded in their envelopes, seven pages written in pencil, the cabinet cards in good condition with pencil notes on several backs, brittle edges on the newsprint, and a day book with a split spine that opens safely to about 110 degrees. There was no film, no glass and no mold. The rights question was noted: the letters are unpublished and the writers’ death dates are not yet known.

Because the box is small, the society chose to make all of Box 3 its pilot box. It approved the capture specification, a CSV import into its own collections database, and a preservation package.

2. Capture

The box traveled in its own carton inside a padded carrier, with a folder-level list signed at handover. Capture ran in six sessions at the studio, each starting with a color and scale target. The agreed specification:

Series FADGI 3rd ed. category Target Sampling Handling notes
Letters and envelopes 3.3 Unbound manuscripts, rare and special Three-star 300 ppi; 400 ppi for the 7 pencil pages Folds relaxed by hand only where the paper allowed
Cabinet cards 3.8 Prints and photographs Three-star 400 ppi Front and back; whole mount in frame
Day book 3.1 Bound volumes, rare and special Three-star 300 ppi Supported, opened no more than about 110 degrees
Clippings 3.7 Newspapers Three-star 300 ppi Carried on a board; brittle edges not handled
Telegram 3.3 Unbound manuscripts, rare and special Three-star 300 ppi Front and back

Masters are TIFF files in color, with bit depth and color space as the FADGI table lists for each category at three-star. Access copies are JPEGs 3,000 pixels on the long edge. Each file name uses the item identifier plus a four-digit sequence, so the first page of the April 12, 1893 letter is ACHS-3-012_0001.tif; day book images are named by page, so page 77 is ACHS-3-L077.tif.

Every image was looked at. Nine were recaptured (2.2 percent): four cabinet cards with glare on gilt mount edges, three day book pages soft in the gutter, and two envelopes with the flap cropped. All six target images measured within the agreed tolerances. The society reviewed twenty images from the first session before capture continued.

3. Read

A person transcribed sample pages exactly, and recognition was measured against them for each kind of page. Character error rate (CER) is substituted, missed and added characters divided by the characters in the reference. For Ida’s letters: (151 + 38 + 25) divided by 5,400 is 214 divided by 5,400, or 3.96 percent.

Material Reference sample Character errors CER WER
Ida Okerlund’s letters 6 pages 214 in 5,400 3.96% 11.57% (118 of 1,020 words)
Walter Okerlund’s letters 3 pages 192 in 2,450 7.84% 17.91% (77 of 430 words)
Day book 3 pages 264 in 2,640 10.00% 27.50% (143 of 520 words)
Clippings (print) 2 clippings 19 in 3,100 0.61% 2.59% (14 of 540 words)

Both sets of letters fall inside the 2 to 8 percent range that Transkribus, in its own guidance, calls good for handwriting; the day book, with its cramped columns and abbreviations, does not. A published range is not a result, though: the numbers that matter are the ones measured on your own pages. In Ida’s uncorrected text the model read “Wren Gulch” as “Wren Gutch” eleven times, which would have hidden those letters from a search for the town.

With the numbers in hand, the society chose a correction level for each batch: targeted correction (names, places, dates, numbers and low-confidence lines) for Ida’s letters, full correction for Walter’s letters and the day book, and uncorrected text for the clippings, labeled as uncorrected. The telegram’s eight words were simply typed in by a person. Fresh samples measured after correction came to 1.12 percent CER for Ida’s letters, with no errors left in names, places or dates, and 0.38 percent for the day book. The OCR accuracy calculator runs the same arithmetic on your own counts.

4. Catalog

A model suggested names, places, dates, subjects and a short summary for the 68 loose items. The society asked for names, places, dates and summaries to be reviewed in the pilot, and left subjects for its own volunteers later.

  • Names: 152 suggestions; 131 accepted, 17 edited, 4 rejected.
  • Places: 74 suggestions; 66 accepted, 6 edited, 2 left as suggestions for the society to decide.
  • Dates: 68 suggestions; 57 accepted, 11 edited, mostly cabinet cards dated from mount style.
  • Summaries: 68 drafts; 48 accepted, 18 edited, 2 rejected and rewritten by the cataloger.
  • Subjects: 196 suggestions, not reviewed in the pilot and still labeled.

One of the two undecided places is in the July 21, 1895 letter: “a band came over from the Springs”. The model suggested Colorado Springs; the cataloger could not rule out another town and left it labeled as a machine suggestion. The day book has a volume record and 212 page records, so each page can be found by its date; their names and dates come from fully corrected text. That makes 281 records in all.

The record for the April 12, 1893 letter:

Field Value Label
Identifier ACHS-3-012 Human-entered, from the inventory
Title Ida Okerlund to her sister Marit Edited (the model suggested “Letter about a flood”)
Date 1893-04-12 Reviewed, read from the dateline
Names Okerlund, Ida; Okerlund, Marit Reviewed
Places Wren Gulch (Colo.) Reviewed
Subjects Floods; Bridges Machine-suggested, not yet reviewed
Summary Ida writes that the creek rose in the night and took the footbridge below the schoolhouse, and that the county will not rebuild it before the snow. Edited (the draft said a dam had failed, which the letter does not say)
Extent 2 pages and an envelope, 4 images Human-entered at capture
Rights Copyright Not Evaluated Human-entered; unpublished, writer’s death date still to be researched

The rights row comes from the rights review worksheet, and it records a working assessment, not legal advice. In the CSV the society imports, every descriptive field has a label column beside it:

identifier,title,date,date_label,places,places_label,subjects,subjects_label
ACHS-3-012,Ida Okerlund to her sister Marit,1893-04-12,reviewed,Wren Gulch (Colo.),reviewed,Floods; Bridges,machine-suggested
ACHS-3-027,Ida Okerlund to her sister Marit,1895-07-21,reviewed,Colorado Springs (Colo.),machine-suggested,Fourth of July celebrations; Bands (Music),machine-suggested

5. Return

Everything was delivered as one BagIt bag, copied to two drives that the society keeps in different buildings. The checksums were verified on both drives before sign-off.

achs-box3/
  bagit.txt
  bag-info.txt                 who, what, when, file count
  manifest-sha256.txt          a checksum for every file in data/
  tagmanifest-sha256.txt
  data/
    masters/                   410 TIFF files
    access/                    410 JPEG files, same names
    text/plain/                page text, one file per page
    text/alto/                 ALTO XML with word positions
    records/achs-box3.csv      281 records with label columns
    records/achs-box3-dc.xml   the same records as Dublin Core
    reports/pilot-report.pdf
    reports/capture-log.csv
    reports/inventory.csv

Box 3 came back in folder order. Every folder was checked against the outbound list, condition was compared with the photographs taken before it left, and the society’s collections volunteer signed the return receipt. The day book came back wrapped as it left, with a note recommending that a conservator look at the split spine before it is handled again.

The pilot report recorded the time each stage took for Box 3, in hours rather than dollars:

Stage Hours
Preparation and ordering 6.2
Capture, including recapture 9.8
Image quality control 1.6
Reference transcription and measurement 3.0
Text correction 36.9
Catalog review 10.2
Total 67.7

Full correction of the day book took 31.1 of those hours. For the society’s two other day books, the report set out both options in hours: full correction, or correction of names and amounts only.

What this example does not show

Real error rates depend on the hand, the ink, the paper and the model, so your pilot box will measure its own. The search above uses a small hand-made list of related words to show how meaning-based search behaves; a production index uses a language model’s embeddings, as our AI policy explains.

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.

Plan a pilot box with your own box

Studio and contacts

24 South Weber Street, Suite 225
Colorado Springs, CO 80903
(719) 244-8450sales@dastok.com