How it works
Six steps, and a named holder at each one.
The question that matters in this business is not what the pipeline does. It is who is holding the data while it does it, and under what agreement.
01. Rights
Before anything is exported we establish what the company actually owns, what is covered by somebody else's agreement, and what has to be excluded. This is the step that kills most bad deals, and it is the one an introducer has no incentive to do properly. It ends in a signed data licence and revenue share agreement with a written exclusion schedule.
Held by: the company. Under: its own systems.
02. Export
The company exports from its own admin consoles, in native format, to a controlled destination. We size from the console rather than from a sample, so the manifest is counted rather than estimated. Nothing is scraped and nothing is pulled through a credential we hold.
Held by: the company. Under: its own systems.
03. De-identify
The named standard is applied: direct identifiers suppressed or salted-tokenised, quasi-identifiers generalised, free text scanned, scanned documents redacted by pixel destruction. A second reviewer checks the output against the known identifier list. Both operators are recorded.
Held by: us, under licence. Staff under a signed data-handling agreement.
04. Structure
Our engineers normalise formats, preserve native structure where it carries information, label every item by document type, industry, format, page count and word count, and screen for duplicates. The result is a corpus rather than a folder.
Held by: us, under licence.
05. Quality assurance
Two people, an exclusion re-screen, a substring scan against the identifier list, a duplicate check, and a residual-risk note written for that corpus. A delivery certificate is signed before anything moves.
Held by: us, under licence.
06. Licence
Delivery in labelled tranches against an agreed acceptance window, with the provenance record available for audit. The source company is paid its share from receipts, quarterly, with a statement showing the allocation. When the corpus is licensed again, it is paid again.
Held by: the lab, under licence.
What we destroy, and when
Raw identifiable copies are held only for as long as steps three and four require, then destroyed on a fixed schedule with the destruction recorded in the chain-of-custody log. See retention and deletion.