How KNTRA works
From raw records to a verdict you can trace.
Not an estimate, not a sample. A proof of where your documents contradict each other, with the exact line that proves it.
The machine, end to end.
Five steps, from the documents you already have to a contradiction you can open and trace.
Sources
Contracts, policies, operating models, financial schedules and the data behind them, connected from the systems you already use. No migration and no re-keying: KNTRA reads the record where it already lives.
Read
A language model reads each document into discrete claims, every one tagged with its exact origin: the file, the page, the cell. This is the only place a model is used, and it only transcribes. It decides nothing, and it never enters the proof.
One graph
Every claim from every source becomes one connected graph of what your organization knows: a single object instead of a pile of files, where each statement can be weighed against every other, across documents.
Prove
Deterministic mathematics checks whether the whole model can hold at once. Not a sample, not a careful read of the parts that looked important: the entire body, tested for coherence in a single pass, the same way every time.
Located
Where the record cannot all be true, KNTRA returns the exact statements that conflict, each with its source and a counterexample. You get the full set of contradictions, not a confidence score and not a best guess.
A finding, as it comes back.
One record, the way the proof is returned: the two statements that collide, each traced to its source, and the state.
Liability, one contract
Statement A
“Liability is capped at the fees paid in the previous twelve months.”
Master agreement, clause 14.2
Statement B
“The supplier accepts unlimited liability for any data breach.”
Security addendum, section 6
Two caps, one breach: they cannot both govern. KNTRA returns the pair with its source and the state the moment the contracts are read in, not after a dispute.
Illustrative finding. Two statements, one state, the shape every proof comes back in.
Not a consultant.
Not a chatbot.
A proof.
KNTRA is a verification layer that sits under your stack, not another opinion on top of it. Think of a type-checker, but for everything your organization knows: it catches the contradiction before it becomes a finding, a write-down, or a failed audit.
It is not BI, not search, and not process-mining. It reads the whole record and returns one auditable answer, the same one every time.
What every proof carries.
Deterministic
The same records always return the same verdict. No sampling, no temperature, nothing that drifts between two runs.
Sourced
Every finding carries the exact clause it came from. A contradiction is something you can open and trace, never an opinion you take on faith.
Whole-record
The entire body is checked as one object, at once. A contradiction three documents deep is found in the same pass as one on the first page.
Resilient
A conflict is held as a precise state and located, not smoothed into a plausible answer. The record can disagree with itself and still be reported clearly.
Separated by construction.
An isolated EU instance
Every engagement runs on its own single-tenant instance on audited European infrastructure. Nothing is shared with another customer, and nothing leaves the EU.
Zero retention
Your records and their graph stay inside your own instance, yours to keep or delete. KNTRA holds no separate copy, the working artifacts from each proof are discarded, and nothing you bring ever trains a shared or a base model.
A person keeps the call
KNTRA surfaces the contradiction and its source. What to do about it, correct a clause, escalate it or accept it, stays a human decision.
Bring the questions.
KNTRA returns the proofs.
Focused, four-week pilots for regulated teams. You bring the records and the questions; KNTRA returns where they hold together and where they do not.