Layer 01
Council review
Every critical analysis runs through evidence, logic, citation, and counterargument checks. Outputs that disagree are retained in the dissent log and routed to human review.
05 / Precision
Litigation-grade evidence, clearly marked for review.
Patent work tolerates uncertainty differently than other domains. A 90 percent answer is a losing answer if the missing 10 percent invalidates a claim. PatentOS is built around four precision layers, not one model call.
Four layers
Critical output passes through verification before external action.
Layer 02
Confidence intervals
Numerical outputs come with stated confidence intervals, not point estimates. Damages, infringement probability, claim breadth, validity strength. Every figure shows its uncertainty so attorneys can weight it correctly.
Layer 03
Source citations
Every claim element traces to an exhibit. Every damages figure traces to a comparable settlement. Every prior art reference includes the specific paragraph. No assertion without a paper trail.
Layer 04
Counsel review path
Selected deliverables can route to a licensed benrishi, bengoshi, or the customer's own counsel. Low-confidence outputs are flagged for senior review instead of being presented as final.
Why this matters
A bad claim chart costs more than a manual one.
Wrong element mapping triggers a counterclaim. Inflated damages signal weak preparation to opposing counsel. Missed prior art destroys a case in discovery. The downside of an unreliable analysis is not a missed deadline. It is a lost case.
PatentOS was designed by people who have lived this. The precision layers are not a marketing artifact. They are the difference between an artifact a partner will sign and one they will throw out.
"Confidence intervals on every number. Exhibits on every claim. No more reviewing twenty pages to find one assertion we can defend."
Iwase Kanji · EVP and Patent Attorney
Measured outcomes
What precision looks like in practice.
3+
Model families per output
100%
Outputs with source citations
Route
High-stakes outputs to counsel review
A / B / C
Conviction grades on every finding
What we do not do
Some honesty about the limits.
No
Autonomous filing.
PatentOS does not file applications, send licensing letters, or initiate litigation on its own. Every external action is gated by a human decision.
No
Legal advice.
The platform produces evidence and analysis. When legal advice is needed, it comes from licensed patent attorneys or the customer's own counsel, not from the agents.
No
Training on your data.
Customer documents are not used to train PatentOS models. Your portfolio analysis stays in your tenant. Full details in our data processing addendum.
No
Unmarked certainty.
When the data is thin, the platform says so. Confidence intervals widen, conviction grades drop, and review paths are surfaced. We do not paper over uncertainty.
Verify it yourself
Audit our precision on a case you already worked.
Send a patent and a product where you already know the answer. We will return the platform's analysis side by side with the ground truth. You judge.