Documents are not facts.
A case file holds hundreds of assertions - dates, amounts, accounts of who said what and when. What a case runs on is smaller and harder-won: the facts the evidence actually supports. MatterOS is built to get you from one to the other, and to show its work on every step.
The expensive surprises live between the documents.
A folder of well-organized documents will not tell you that the delivery date in the contract disagrees with the one in the email thread. It will not tell you that the limitation date nobody calendared is four months out, or that the account your client gave is contradicted by an exhibit on page 40 of something you filed yourself. Those discoveries usually arrive late, and late is what makes them expensive.
So MatterOS treats the fact, not the document, as the unit of work. Files are read once, into structured facts with provenance - and from then on the matter runs on the facts: the chronology is built from them, contradictions are detected across them, gaps are measured against them, and every answer the AI gives quotes them back to you with citations.
Four commitments, kept in the product.
Every fact carries its source
Extraction doesn't summarize - it pins. Each event, party, and assertion records the document, the page, and where the text allows it, the verbatim line it came from, resolved to a highlight you can open with one click. A fact that can't show its source says so, in so many words.
Contradictions are the headline
When a new file lands, its facts are checked against everything already on record. Two dates for one delivery, two accounts of one meeting - the conflict becomes a first-class item citing both documents, visible the moment the matter opens. A contradicted fact is never presented as reliable.
Absence is a finding
Each practice area carries the facts a matter of its kind should have - the incident date, the notice, the limitation date. When the record doesn't contain one, that gap is shown at the top of the ledger, not discovered by opposing counsel first.
Confidence you can interrogate
Every fact wears a confidence band derived from signals already on record: source, quote, corroborating documents, your own confirmation, contradictions. Tap it and see the arithmetic. High confidence is reserved for facts a human confirmed or multiple documents agree on - provenance alone never gets there.
One view of what the record establishes.
Every matter carries a Fact Ledger: each fact with its status - corroborated, conflicting, uncorroborated, or awaiting your confirmation - its confidence and the reasons for it, its source a click away. Filter to the conflicts before a difficult call. Export the ledger or the cited chronology to Word, PDF, or a spreadsheet when the deliverable needs to leave the building. Confirm or correct facts inline; your confirmations are what the record's confidence is built from.
Asked and answered, without hedging.
Can I trust what it extracts?
You are never asked to. Every extracted fact arrives unverified and stays visibly unverified until you confirm it - and every one carries its citation: the document, the page, and where the text allows it, the verbatim line, one click from the highlighted source. Confidence is derived from stored signals you can inspect (is there a quote, does a second document corroborate it, did a human confirm it) - never from a second AI opinion grading the first.
Will it miss things?
Sometimes - and the product is built to say so rather than hide it. Every document gets a source ledger entry showing whether it was read in full, partially, or not at all. A fact with no quote or page to check against is flagged as thinly sourced. And the Fact Ledger shows the facts your discipline expects that the record doesn't contain yet - a gap is presented as a finding, not passed over in silence.
Is this just about speed?
No. Speed is a side effect. The point is that the factual record exists early and stays testable: by the time you decide strategy, you already know what the evidence supports, what it contradicts, and what is still missing - instead of discovering all three the week before a deadline.
What happens when documents disagree?
Both sides surface. A date two documents state differently ships with every competing date and its citation. An account one document contradicts is marked conflicting and its confidence is forced down, however well-sourced it otherwise is. MatterOS never picks a quiet winner.
AI-assisted features support the professional's own workflow and do not constitute legal advice.