The layer that operationalises the knowledge journey.

Not a product bought on its own, but the layer that assembles the capabilities each case needs, and preserves what must not be lost between them.

Capabilities

Called on as the case requires, never imposed as a single bundle.

Expert transcription of a sample, exact to the original, on which recognition models are built and against which accuracy is measured.

Models trained on the hand itself rather than on print, then used to generate the text of everything not transcribed by hand.

Material type, date, position, language and reading confidence.

People, places, dates, terms and items, each bound to where it appears in the original.

The same entity unified across sources, so each gains a file gathering every mention.

Across languages, with a terminology memory that accumulates instead of restarting.

Across text and entities together, with a permanent route back to the source.

Low-confidence passages are flagged and located on the image, so review is targeted, and each correction returns as new ground truth.

Into the institution's own systems, or into thematic files and finished catalogues.

Every value is bound to its position in the original and can be verified at any time.

Relationships between units survive export and any change of interface.

Who reviewed, what changed, and when. A full record, not scattered notes.

It means nothing outside a project with an asset, a problem and a measured outcome.

Sensitive outputs pass through review before becoming a source anyone relies on.

A project calls only on what it needs; what the journey does not require is not switched on.

That is not answered from a list. It is answered by looking at the material itself and the problem you want to solve.