Skip to main content

Reading a document

The Extract tab lists everything on the case that can be read, bought and uploaded together, because what matters is whether a document has a type with a schema behind it, not where it came from.

The Extract tab listing documents, each with a badge saying whether it came from a register or was uploaded, and its type
Each card says where the document came from, what type it is, and what can be done with it.

When a document cannot be read

The reason takes the place of the button, so the fix is one click away rather than something to go hunting for:

It saysDo this
Not classifiedGive it a type on the Documents tab
No extraction schema for this typeWrite one. The link is on the card

Running it

Read this document is a paid act: the document is parsed and read, and the pages billed are shown afterwards.

It reports where it has got to as it goes (Queued, Sending the file, Mapping the pages, Reading the document) and settles on its own if you leave the page.

A failed extraction says why, on the card, and offers Try again.

The evidence viewer

The part worth your time. Every value the schema pulled out, with the page it was read from.

The evidence viewer with extracted values listed beside the rendered source page, the cited passage highlighted
Click a value to see the page it came from, with the passage marked.

Three things you can do here:

Correct a value. The reading is a model's reading of the paper, and the paper is stronger than the reading. A correction made here outranks every automatic source on that field, and it lands next to the passage it disagrees with, which is why correcting from here beats correcting from the Fields tab when the document is the thing you are looking at.

Point a value at the right passage. When a citation landed on the wrong line, the source editor moves it.

The source editor dialog for choosing which passage on the page a value was read from
What the value rests on, corrected, not just what the value says.

Remove a citation that supports nothing.

Every one of those is recorded in the case's activity with the value before the change.

Why the citation matters beyond this tab

Because it is what the report prints. Every cited value in the report's evidence appendix carries a band of the source page, full width, with the passage marked, so a citation pointing at the wrong line becomes a picture of the wrong line in a document attached to a decision.

See Issuing a report.

Discarding an extraction

Discard removes the reading and everything it contributed to the dossier. The document stays; read it again with a corrected schema and you get a new reading. Discarding does not refund what the reading cost.