Four kinds of operating record.

Suppliers are not named publicly. Each corpus is described by what it contains, how far back it runs and what has been excluded, which is what a licence is negotiated against.

Education operations

How a school or an education platform actually runs: the mail and chat between staff, the documents they revise, the tickets they close, the ledgers they keep. Business systems only. Never student records, never minors, never platform content.

Systems
Mail, chat, CRM, application database, accounting, document store.
History
Useful from about three years back. Strongest across the past five.
Formats
DOCX PPTX PDF XLSX structured exports
Excluded
Student and end-user data. Staff channels and private messages.
Method
Direct identifiers suppressed at source, free text scanned, then two-person review.

Hospitality and point of sale

The operating record of a venue that trades every day: supplier invoices and order confirmations, point-of-sale bills and sales mix, cash-out detail, back-office correspondence.

Systems
Point of sale, supplier mail, accounting.
History
Useful from two years back. A full trading cycle matters more than length.
Formats
PDF CSV structured exports
Excluded
Payroll and shift records, payment data, buyer identity, other locations.
Method
Buyer identity columns dropped, staff tokenised, payment fields never exported.

Trade and cross-border finance

Proforma invoices, shipping and customs correspondence, and the document trail of moving goods and money between jurisdictions. English language.

Systems
Mail, document store.
History
Useful from two years of continuous trading.
Formats
PDF
Excluded
Account numbers, counterparty identifiers, banking credentials.
Method
Account and routing fields masked by pixel destruction on scanned pages.

Clinical and laboratory administration

The administrative record around clinical work rather than the clinical content itself: reporting workflow, scheduling, correspondence and document handling.

Systems
Reporting system, document store.
History
Useful from two years back.
Formats
PDF
Excluded
All direct patient identifiers. Images carrying a residual text layer.
Method
Identifiers removed by pixel destruction, never an overlay, then re-scanned.

A category appearing here is not an offer of a specific corpus. Every licence is negotiated against a manifest for one named source, and we do not deliver a corpus whose written rights record is incomplete, whatever the commercial pressure. What we hold at any moment, and for whom, is confidential to the companies involved.

Not seeing your industry.

Tell us the segments and systems you need. Sourcing a new corpus to a signed manifest typically runs weeks, not quarters, and we will tell you honestly if we cannot get there.