After uploading: the processing and the details of a document

Where to find it

In your account menu, under Dri's knowledge, Documents tab. Click the title of a document (or anywhere on its row) to open its details. The Knowledge base button, at the top of the details, takes you back to the list.

The states of a document in the list

After Upload, the screen says "Document received. It is in the queue, and the list follows it." Processing runs in the background, and the State column updates by itself while any document is in progress; you do not need to reload the page.

A scanned manual can take a few minutes; a short text, seconds.

Why a document failed

Hover over Failed in the list, or open the details: the red strip says "Ingestion stopped:" followed by the reason. The most common ones:

• "the file has no text I can use": an empty file or one with only images without legible text;
• "I could not read the scanned pages": the vision transcription failed;
• "the file claims to be … but I could not open it": a corrupted file or one exported halfway; export it again and upload it;
• "content triage failed", "fact distillation failed" or "fact embedding failed": a temporary failure of the reading service; click reprocess later.

The Yield and Output columns

Yield says how much of the document passed triage, with the total beside it: "40% 12/30" means 12 passages kept out of 30. Green from 50% up, amber between 20% and 50%, red below. It shows — when the document has not been triaged yet. That is not zero.

Yield shows how many facts and how many passages the document generated. A fact is each piece of information Dri can cite; a passage is the searchable block that groups facts under the same heading. Having more facts than passages is normal.

Low yield is not always a defect: a catalog full of prices and images keeps little, and that is right. In a document written for Dri, it is a sign to open the details and see what was discarded.

The numbers at the top of a document's details

• Passed triage: the same rate as in the list, with "12 of 30" beside it. Right below, "% answers in search": how many facts of this document Dri actually reaches through this search. A fact can pass triage and stay out of search for three reasons: it repeats another document (the other one answers, nothing was lost), it was linked to a catalog product (it starts answering together with the product) or it is in conflict, and only this last one is a real loss.
• Extracted facts, Chunks in the index and Awaiting review: the last one is how many passages are in the Review drawer.

When some fact is out because of a conflict, the strip "N facts of this document are out of search due to a conflict with another document" appears with the link See the conflict queue. The decision is made there, with both sides in view.

The Triage, Distilled and Indexed passages tabs

Triage shows each passage of the document with the literal text (What the document says) next to the reason (Why it went to this drawer). The All, Review, Fact, Volatile and Noise buttons filter by drawer:

• Fact: entered the base;
• Volatile: product price, stock or a time-limited promotion; left out on purpose;
• Noise: footer, page numbering, repeated address;
• Review: triage could not decide; it stays parked.

Distilled is the text the reading produced from the facts. Indexed passages shows what Dri's search sees, each passage with the section heading it came from.

Moving a passage to another drawer

On each passage of the Triage tab, Move to: changes the drawer with one click: it works for any passage, not only those in Review, because triage also errs the other way.

⚠️ Moving to Fact does not put the passage in the base right away: the screen says "Moved to Fact. Reprocess the document for it to enter the base." Reprocess, and the passage enters. Reprocessing redoes the document's triage, but keeps the drawer a person chose for each passage with identical text. If the text of that passage changed in a new version of the file, it is read again and triage decides once more.

Reprocessing and deleting a document

The two icons in the Actions column (and the Reprocess button in the details):

• Reprocess redoes the whole reading from the stored original file. It is for a document that Failed because of a temporary problem, or that got stuck at a step. It is not for updating the content; for that, upload the new file.
• Delete asks for confirmation and removes the document, the facts extracted from it and what is in the index right away. Dri stops answering with that material. There is no trash: to bring it back, upload the file again.

The Team answers document

When you answer a question on the Open questions tab and the answer is stored, it goes into a document created automatically, called Team answers, with origin Answer. It appears in the list like any other.

⚠️ Deleting Team answers erases all the answers stored through the questions queue. And it has no original file, so Reprocess on it ends in Failed. Do not use it.

When it goes wrong

It has been "In queue" for a long time. The list checks again by itself every few seconds. If it is still stuck after a few minutes, click reprocess.

It is "In the base" and Dri does not answer. Test the question in Quality: the test shows what Dri read. See Knowledge base quality: test what Dri answers.

It stayed in "Review". The document already answers with what passed. Open the details, filter by Review and see whether something important is parked there.