What a document record holds

Draft for Reid to review. Written by Claude from the Oct 13 and 14 calls. Plain language first; the column names are only suggestions.

The idea

Every document that arrives gets one record. Each fact pulled from it gets its own row, pointing back to the page and the exact words. The match to a deal is a separate row too, so a person can confirm or correct it without touching anything else.

1. The document

What it holds	Example	Suggested name
A unique id	doc 1042	id
What kind of document it is	Division order	doc_type
Where it came from	Scanner / email / mail / upload	source
When it arrived	Oct 14, 10:12	received_at
Where the file is	The SharePoint inbox file	file_location
A fingerprint of the file, so the same document never lands twice	(automatic)	file_hash
Pages	3	page_count
Where it stands	Arrived → read → matched → confirmed, or needs a person	status

2. Each fact pulled from it

What it holds	Example	Suggested name
Which document	doc 1042	document_id
What the fact is	Decimal interest	fact_type
The value	0.00215000	value
Page and exact words	Page 2, "0.00215000 RI"	page, quote
What it links to, if known	Tract 4, Smith lease	linked_to
Who or what pulled it out	The reader, version 3	extracted_by
Checked by a person?	Not yet / Kelsey, Oct 15	checked_by, checked_at

3. The match to a deal

What it holds	Example	Suggested name
Which document, which deal	doc 1042 → Smith deal	document_id, deal_id
The proof: which facts match the deal	Same tract, same lease, same owner name	evidence
Other deals it could have been	Jones deal (same county)	other_candidates
Matched alone, or sent to a person?	Sent to Reid by email	decided_by
Confirmed or corrected by a person	Confirmed by Reid	confirmed_by

Rules that come from the calls

Certain means proof. It matches alone only when the evidence column names a matching tract, lease or owner. Otherwise it emails a person.
Nothing is deleted. A correction adds a new row and marks the old one replaced, so we can always see what changed and why.
Account numbers are masked before anything is saved, as the deal-data reader already does.
Every confirm request shows the page image, the quoted words, and which deal and why.

Questions for Reid

Does the inventory schema already have tables for deals, tracts and leases we should link to directly?
Should facts from the test go into the same tables the real system will use, or a separate test area?
Anything a division order or check stub needs that doesn't fit here?