DocQore
Connect document intake, text extraction, metadata review and archive ingestion. Keep extraction quality and release approval separate.
DocQore is government document processing that turns paper, microfilm and microfiche archives into a searchable, redaction-ready record system, so a FOIA officer finds the document in minutes instead of digging through boxes for weeks.
Who is this for?
Records managers, FOIA and public-records officers and agency IT directors with a backlog measured in boxes and a statutory clock on every request. It is for agencies that already run a request-tracking system and need the archive behind it to be searchable; it is not a request manager itself.
What problem does it solve?
The request arrives with a deadline and the record is on 35 mm roll film in a basement. Someone has to find the reel, read it, decide what is exempt, black it out by hand and defend that decision later. Fifteen years of records can take fifteen minutes each to locate, and the backlog grows faster than the staff.
Redaction is where the legal risk sits. A missed name under exemption (b)(6) or a misapplied (b)(7)(C) is a finding at the Inspector General audit or a problem in litigation, and a hand process has no record of who decided what.
How does it work?
Bulk intake takes paper, 16 and 35 mm roll film, COM microfiche and existing scans, with deskew, despeckle, contrast and edge preprocessing before OCR. Text extraction runs with named-entity recognition, date extraction and document classification, so dates, names, case numbers and document types become metadata without a data-entry step. Low-confidence fields go to a reviewer with the original image beside the OCR text.
PII is identified by AI and confirmed by a person in a dual-review workflow (redactor, then approver), with exemption code validation for (b)(6), (b)(7)(C) and (b)(4) tagging and a highlight pass for missed PII before release. Every action is logged: who, what, when; the log exports for the FOIA response file, litigation holds and IG audits.
Output meets federal digitisation standards: FADGI 3-star image quality, PDF/A archival files accepted for transfer to the National Archives, Dublin Core and NARA-required metadata. Accuracy is honest about the source: roughly 90 to 95 percent on machine-typed documents and 85 to 90 percent on handwriting, gated per batch before anything enters the archive.
Capabilities
- Document intake and preprocessing for archive digitization
- OCR and metadata extraction with quality review
- Configurable redaction proposals and reviewer decisions
- Archive ingestion, full-text search and export
From an original record to a reviewed archive entry
DocQore covers document intake, OCR, metadata extraction, configurable redaction, quality review and archive ingestion. Start with samples of the actual formats, languages and document conditions. Preserve a way to compare extracted text and proposed redactions with the original record.
Inputs and outputs
- Inputs to agree
- Agree the source formats, record identifiers, required metadata and target archive. Include representative typed, handwritten and damaged material with reviewer-established transcriptions for evaluation.
- Outputs to review
- Define the searchable text, metadata and reviewed record that the archive will receive. Retain the distinction between an extraction, a reviewer correction and a record approved for release.
Integration and operating limits
Map the reviewed record and metadata to the receiving archive's schema, permissions and export requirements. Name the reviewers responsible for exceptions and release decisions. Agree access, deployment and retention requirements before ingesting a collection.
Completed OCR is not release approval. Redaction patterns assist review; they do not decide legal exemptions. Evaluate missed redaction candidates and unnecessary redactions separately, and establish authorization for the proposed processing environment.
What to measure
- Extraction errors by document type and stated measurement unit
- Redaction misses and unnecessary redactions against a reviewer-established reference
- Reviewed throughput including correction time, failed pages and reprocessing
An OCR check that keeps the errors visible
The following invented record illustrates exact field comparison. It is a proposed evaluation worksheet, not a DocQore OCR result.
Scroll the table sideways to read every column.
| Field | Reference transcription | Candidate extraction | Expected comparison |
|---|---|---|---|
| Record ID | ARCH-1042 | ARCH-104Z | Mismatch requiring review |
| Record date | 1987-06-03 | 1987-06-03 | Exact match |
| Route | Route 7 | Route 7 | Exact match |
Two of these three fields match exactly. That is an exact-field result for this worksheet, not character accuracy, page accuracy or a forecast for an archive.
This worksheet compares the strings as written. Record any normalization rules, such as date formatting, before scoring a real evaluation set, and define which fields belong in its denominator.
If a reviewer corrects the identifier, preserve the distinction between the original extraction and the reviewed record instead of reporting the correction as unassisted OCR accuracy.
Review extraction errors separately for typed, handwritten and damaged material. Completed OCR and searchable text do not mean that a record is approved for public release.
Download the synthetic worksheet (JSON)What does it connect to?
- Archive portal with full-text and fielded search; CSV, XML and JSON metadata export and bulk file export
- Request-tracking systems such as FOIAXpress sit in front; DocQore clears the archive behind them
- Priority queues that move urgent requests to the front, with estimated completion on the dashboard
- Role-based access for redactor, reviewer, admin and auditor
How is it deployed, and how long does it take?
On-prem or air-gapped for sensitive agencies, cloud for non-classified records, with TLS 1.3 in transit and AES-256 at rest. The architecture is built to the FedRAMP High baseline with NIST 800-53 controls; classified deployments need an additional security review, and we say that before the contract, not after.
Throughput is over 1,000 pages an hour automated and 500 or more with AI assist, in jobs of up to 100,000 documents; a million pages in 90 days is the scale it is built for.
A searchable archive in 60 to 90 days, with FOIA backlogs typically cleared 60 to 80 percent faster once the archive is live.
Industry context
Explore the related workflows and other solutions for your industry.