Finding aids drafted to DACS from every transcribed page, each fact cited and each line approved by an archivist
Backlogs described in weeks, not years, with every line approved by an archivist.
See one case, screen by screen ↓Archival Transcription & Finding Aids turns scanned letters, ledgers, catalog cards and oral-history audio into transcripts and DACS finding aids for university archives and special collections. Agents transcribe every line with a confidence, a second agent re-reads the uncertain ones, and every folder title, date and scope note cites the page it came from. An archivist approves every line, and every restriction is approved by a second person, before anything is published.
AI in education: a buyer's guide to agentic AI for colleges, universities and school districts →
Updated
Finding aids are how a researcher learns what an archive holds, and most university archives and special collections hold far more than they have described. At the most cited estimate of about 8 hours a linear foot, a processing backlog of 18,000 linear feet is decades of staff time, so letters, ledgers and oral histories sit on the shelf known to staff and invisible to everyone else. Minimal processing describes boxes and folders, not words, so even a described collection stays unsearchable line by line.
Transcription tools give raw text but no archival description, and every page still has to be read for student records under FERPA, medical details and Social Security numbers before anything goes online. Archival Transcription & Finding Aids puts seven agents in front of that work: they transcribe, re-read, draft DACS description with a citation for every fact, and hold anything sensitive, so the processing archivist spends the day approving and deciding, not typing.
Estimated split of the commonly cited 8 hours per linear foot, by hand and with the agents; the 'after' column totals 2 hours, the design target. Our arithmetic, kept modest: rehousing boxes is still done by a person.
Seven agents transcribe, re-read, describe and screen every box; archivists approve every line and every restriction.
Halden State University's Office of the Dean of Students records, 1946–1979 (UA-0214): 64 boxes, 32 linear feet, transferred in August 2024. Here is what happens to Box 18, the correspondence of 1962, screen by screen in the working solution, on Wednesday, October 21.
The agents transcribed 141 pages, 3,512 lines, and accepted 3,301 lines at confidence 0.90 or more. The Reader agent re-read the other 211 on its own and agreed on 188; 23 lines went to a person. The Describer drafted 12 folder descriptions with 64 citations, and the Sensitivity agent held one folder.
Needs you: Box 18, folder 4 on hold; 11 folder descriptions to approve; 23 lines the two readers disagree on.
UA-0214 is drawn as its 64 archival boxes in six series. Boxes 1–16 are full and public, Box 17 is half full, Box 18 is full with a hold mark, and Series 4, the conduct files, is closed by the transfer agreement. Under the shelf, Box 18's finding aid sits in its DACS hierarchy, every agent-written title in italic with a draft tag.
2,412 pages transcribed · 186 folders described · 8 folders restricted · 1 on hold.
A January 8, 1962 letter from the Dean of Men sits beside its transcript, line by line, with confidences from 0.91 to 0.99. A handwritten note on line 11, "Agreed - lunch Thurs. 18th? M.A.T.", read at 0.71, was re-read independently and shows "Second read agreed 0.93". Line 12 shows [illegible?]: the readers disagree, so a person decides.
Each line to check shows the scan crop and both agents' readings with their confidences. Ines reads "Eastgate" once in Dean H. V. Whitcomb's hand, and the same reading is proposed on 6 more lines in folders 2, 5, 9 and 10, accepted in one step, and the Reader agent re-reads 10 more of his lines with the corrected hand and now agrees. The 6 lines left are all in folder 4, which is on hold.
Nothing is guessed: a line the two readers cannot agree on stays [illegible?] until a person settles it.
For each folder the Describer drafted a DACS title, a one-sentence scope note, dates, pages and its sources; folder 1 is "Correspondence, January–February 1962", 14 pages, 6 sources. Tomas approves the 11 open folders in one step. Folder 4 cannot be approved while it is held.
The citation on folder 2's scope note opens Box 18, folder 2, page 3, lines 1–6: a March 1962 memo with the lines behind the note highlighted, such as "Council asks 12:30 a.m. on Fri. and Sat." The archivist checks the source before approving, and can ask about it.
Folder 4, "Disciplinary correspondence, 1962", names three students in a disciplinary action on page 1, lines 6–14, with a medical detail on line 12 and student numbers on page 4. The agent proposed RS-26-0417 at confidence 0.97: closed, description only, under FERPA, access policy 4.2 and clause 6 of transfer agreement 2024-031, opening January 1, 2038.
The Sensitivity agent can only hold. It never publishes, lifts a hold or decides.
Tomas confirms the restriction at 10:14 a.m. The approval then opens for Lorraine, with the folder, the cited lines, FERPA and access policy 4.2 and the opening date in front of her; she approves at 10:22 a.m. The person who confirms a restriction can never approve it.
Four checks pass: 11 of 11 open folders approved, the restriction approved, all 17 lines in the open folders settled, and 14 of 14 WCAG 2.1 AA checks with both alt texts approved. The archival management system receives 12 archival objects in EAD3, the digital asset system 135 page images and 11 transcripts, and discovery 12 records.
Researchers see "Disciplinary correspondence, 1962 (restricted: student records)", opens 2038.
The audit trail lists who did what, when, why and on which evidence, with people and agents labeled. The export pack for Box 18 holds the EAD3 finding aid with its restriction, the per-line transcript trail of 3,512 lines with both readings, the 64 citations, the restriction's maker and checker, the accessibility report and the model and prompt versions used.
The same box, seen by the four people who carry it: what their day looked like, and what it looks like now.
Seven agents transcribe, re-read, describe and screen every box; archivists approve every line and every restriction.
Checks each capture batch by box and folder: image counts against the folder list, checksums, barcodes, blank or rotated pages, and the release form for audio.
Transcribes handwriting, typescript, carbons, ledgers, catalog cards, item labels and oral-history audio, with a confidence on every line and speakers and timestamps for audio.
Re-reads every uncertain line on its own, with a different model and without seeing the first reading. Agreement accepts the line; disagreement sends it to a person as [illegible?].
Drafts DACS description at folder, series and collection level, titles, dates, extent and scope, citing the page and lines behind every fact.
Screens every page for student records, medical details and Social Security and student numbers, applies deeds of gift, transfer agreements and the records schedule, and proposes a restriction with its reason, lines, rule, opening date and a public title.
Checks transcripts and finding-aid pages against WCAG 2.1 AA, headings, reading order, alt text and captions, and drafts the fixes.
Exports EAD3 to the archival management system, images and transcripts to the digital asset system and records to discovery, and assembles the audit pack.
Approve every line and restriction. The agents propose; a named person decides.
Staff can ask the assistant about anything their role may see. It answers with each page cited, and opens the screen where the work is done.
Why is Box 18, folder 4 on hold?
The Sensitivity agent found three students named in a disciplinary action on page 1, lines 6–14, a medical detail on line 12 and student numbers on page 4. FERPA protects education records, and access policy 4.2 closes them for 75 years. It proposes closing the folder until January 1, 2038, with a public title. You confirm; the head of special collections approves.
Which Box 18 lines do the two readers disagree on?
23 lines. 17 are in Dean Whitcomb's hand, and 7 of those differ on one word, "Eastgate" or "Eastgale". Settle that word once and the same reading is proposed on the other 6. The 6 lines not in his hand are all in folder 4, which is on hold.
Which collections have a researcher waiting?
Three: the Halden Normal School ledgers (due Oct 28; 8 of 14 ledgers transcribed), the Board of Trustees minutes (due Nov 4) and the Hollis Grange family letters (due Nov 9). The backlog is filtered to them.
Draft the scope and content note for Series 2
A draft for your approval, every fact cited: "Letters and memoranda of the Dean of Men and the Dean of Women with the President, the Dean of Women, student government and residence halls. Correspondence naming students in disciplinary matters is restricted folder by folder." It stays a draft until you approve it.
13 screens from the working solution, on its sample data. Pick one to see it large.
The thresholds and restriction rules every box is checked against are settings, versioned, and every change needs a reason.
| Setting | Default | Choose from |
|---|---|---|
| Accept a line at | Confidence 0.90 | Versioned; every change needs a reason |
| Second read | 0.60 – 0.89 | Below 0.60, straight to a person |
| Sensitivity hold at | 0.50 | Low on purpose: a hold costs a minute of a person's time |
| Every description fact cites a page | On | Locked |
| Student records (FERPA) | Closed 75 years from creation | Access policy 4.2; medical details closed 75 years |
| Social Security and student numbers | Redacted in the access copy | The folder stays open |
| Oral history release forms | No transcription without a signed form | Seals applied to audio and transcript |
| Clocks | Hold decision in 5 business days | Researcher request 10 business days; processing plan 30 days from accession |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
Sources: Greene and Meissner, "More Product, Less Process" (2005), about 8 hours per linear foot · ALCTS Exchange lightning round (UNT Digital Library), $581.98 per processed linear foot · 28 CFR Part 35, subpart H, compliance dates extended April 20, 2026 · FERPA, 34 CFR Part 99 · Society of American Archivists, DACS; EAD3 (2015) · "target" = design goal, measured in the live solution · "demo" = seen in the working solution, on its sample data · "typical" = published figures · "estimated" = our arithmetic from published inputs. Halden State University, its collections and the people named on this page are fictional: characters and sample data in the working solution.
Archival Transcription & Finding Aids turns scanned letters, ledgers, catalog cards and oral-history audio into transcripts and DACS finding aids for university archives and special collections. Agents transcribe every line with a confidence, a second agent re-reads the uncertain ones, and every folder title, date and scope note cites the page it came from. An archivist approves every line, and every restriction is approved by a second person, before anything is published.
A finding aid describes an archival collection so a researcher can find what is in it: the collection, its series, boxes and folders, with titles, dates, extent, scope and content, and the conditions governing access. In the United States it is written to DACS, the Society of American Archivists' content standard, and exchanged as EAD. Here the Describer agent drafts it at each level, and the archivist approves it.
It does not guess. Every line gets a confidence. Lines from 0.60 to 0.89 are re-read by a second agent with a different model that never sees the first reading, and lines below 0.60 go straight to a person. If the two agree, the line is accepted with both readings on record; if not, it is marked [illegible?] and goes to a person with the scan crop and both readings, and one correction is proposed on every other line in the same writer's hand.
Yes. Audio is transcribed with speakers, timestamps and a confidence on every line, with a segment index and captions exported as WebVTT. No audio is transcribed without a signed release form, and a narrator's seal is applied to both the audio and the transcript until its date.
The Sensitivity agent reads every page for student records under FERPA, medical details and Social Security and student numbers, and applies deeds of gift, transfer agreements and the records schedule. It can only hold a folder, with the lines and the rule cited; the processing archivist confirms and the head of special collections approves. Identifiers can instead be redacted in the access copy so the folder stays open.
Yes. The Accessibility agent checks transcripts and finding-aid pages against WCAG 2.1 AA, including headings, reading order, alt text and captions, and drafts fixes for a person to approve; a box cannot be released until the checks pass. The ADA Title II web rule sets April 26, 2027 for public entities serving 50,000 or more.
Yes. Every agent-written line stays a draft until an archivist approves it, and a rejection or release of a hold needs a reason code. No agent publishes, approves or lifts a hold, and the person who confirms a restriction can never approve it.
The Agentic Solution Engine builds and deploys it from your requirements and documents, such as your access policy, records schedule and a sample of scanned boxes, and it goes live once every quality gate has passed. We will walk you through it on your own boxes first.
Two blind votes on every record, a person on each conflict, every number tied to its page.
1-min demoResearch Integrity ScreeningDuplicated figures fixed before submission, not retracted after publication.
1-min demoResearch Conflict of Interest & Research Security ReviewUndisclosed ties found before just-in-time, and explained by the researcher first.
1-min demoWe'll run Archival Transcription & Finding Aids on a sample of your own scans and oral-history audio.