Can we accept their uncapped confidentiality clause?

A health system has sent its own contract to a software company's lawyers. The associate has a redline ready for the partner. Everything needed to answer the question is already written down somewhere, in seven files that nobody has read side by side.

Scroll through to watch the files become a context graph, see where the draft breaks, and see why the answer differs from what a general AI assistant would give you.

and one thing the partner knows that no file says

Seven files, one negotiation

Saint Aldric wants uncapped liability for any breach of confidentiality. Brightline's lawyers hold a redline, a DPA nobody has read yet, an insurance summary, a client playbook, the firm's precedents, an old partner email and a deal workbook.

Each file says something specific

Every page is read and the clauses, numbers and precedents that bear on the question come out as typed entities, each one tied back to the passage it came from.

They connect into a context graph

Typed relationships show how the provisions depend on each other: what pulls data into a clause, what authorizes a cap, what the cap draws on, and which precedent already went through this. That connected structure is a context graph, and it is what Tenuto Labs builds for each client.

The carve-out swallows the super-cap

§8.1 puts all customer data inside Confidential Information. §11.2(b) makes any confidentiality breach uncapped. A data incident is both, so the new 2x super-cap never applies. The firm signed the same mistake for Tamsin in 2024, and the fix cost the client 4% of fees.

And the DPA overrides it anyway

DPA §9.1 removes every limit for data protection breaches, and §15.4 says the DPA wins. Even with the carve-out fixed, the cap would not hold. The partner's 2025 email said to check exactly this. The DPA is still marked not yet reviewed.

The number nobody wrote down

Each precedent landed its super-cap near 30% of that client's cyber limit. No file says that is a rule, and Ardent's $3M is the tempting number to copy against a tower half the size. No AI model, however capable, can read a rule that was never written down, which is why this question goes to the partner next.

The partner answers one question

The pattern becomes a single question for Margaret Okafor. Her answer takes two sentences and becomes a rule in the context graph, linked to the super-cap it limits and the precedent it rules out. The next associate who asks gets her answer without needing her time.

Question generated from the context graph

Every precedent's super-cap sits near 30% of that client's cyber limit, and the playbook doesn't mention it. Is that a rule, and what does it mean for Saint Aldric?

Answer, Margaret Okafor

It's a rule. One customer never gets more than about 30% of the client's cyber limit, because that limit is shared by every customer all year. Here that's $1.5M, and only after 11.2(b) and DPA 9.1 are fixed.

M.O.

Added to the context graph: 1 rule · 2 relationships · source: partner, Oct 2026

Same question, asked twice

Ask the hero question both ways. A general AI assistant searching the firm's files retrieves passages that point in different directions, and it has no way to know the partner's rule. The context graph follows the provisions and returns an answer with its reasons.

A general AI assistant retrieves

DOCXPlaybook, Fallback 1: add breach of confidentiality (mutual).reads as yes

DOCXArdent, 2025: greater of 2x fees or $3,000,000.reads as a number to copy

EMLOkafor, 2025: the carve-out must expressly exclude data incidents.reads as no

finds the passages, not the reasons

It can summarize these. It cannot know the 30% rule, because no file contains it.

The context graph answers

No, not as drafted.

§8.1 puts all customer data inside Confidential Information. MSA
§11.2(b) leaves any Article 8 breach uncapped, so §11.3 never applies to a data incident. Same failure as Tamsin, 2024. MSADOCX
DPA §9.1 removes the cap anyway, and §15.4 lets the DPA win. DPAMSA
Counter: exclude Data Incidents from 11.2(b), conform DPA 9.1, and set §11.3 at the greater of 2x fees or $1.5M, aggregate. Okafor rule

Your turn: change the rule

Swap the partner's rule for another one. The super-cap, Brightline's exposure against its cyber limit and the answer all move with it, because each is derived from the rule rather than typed in.

About 30% of the cyber limit (Okafor's rule): Super-cap for Saint Aldric $1.5M; Signed super-caps if this signs $4.65M of $5M (93%); Signed super-caps against the $5M cyber limit 93%. Counter at $1.5M once the drafting is fixed, and tell Brightline's broker that signed super-caps reach 93% of the limit.
Greater of 2x fees or $3M (Copy the Ardent precedent): Super-cap for Saint Aldric $3M; Signed super-caps if this signs $6.15M of $5M (123%); Signed super-caps against the $5M cyber limit 123%. On paper Brightline would promise more than its whole cyber limit. One large claim here would use most of the cover every other customer relies on.
2x annual fees (Playbook Fallback 1 only): Super-cap for Saint Aldric $960K; Signed super-caps if this signs $4.11M of $5M (82%); Signed super-caps against the $5M cyber limit 82%. Fits inside the limit, but the number was never the problem. The context graph still flags §11.2(b) and DPA §9.1, and a 2x offer leaves nothing to trade against an uncapped ask.

this is what Tenuto Labs builds

A context graph built on your documents and your experts.

This is a worked example on a synthetic matter. The documents, parties and figures are invented; the method is the one we use. In an engagement we model your domain, connect your own documents into a context graph, and put a few short questions to your experts each day, so their judgment lives in the system and not only in their heads.

General AI assistantContext graph, built by Tenuto Labs

Where the knowledge livesWhatever it retrieves or you paste in, for one conversationEvery provision, precedent and rule, connected and kept across matters

Expert judgmentOnly what someone wrote downCaptured from your experts as rules, like the 30% limit

Showing its workA fluent answer you have to checkAn answer with its path, each step tied to a page

When a rule changesRe-explain it in every new conversationChange it once, and every answer that depends on it updates

Start a conversation

The seven files behind this example

PDFSaint Aldric MSA, Brightline redline v2
PDFSaint Aldric DPA, customer form
PDFCyber and Tech E&O coverage summary
DOCXBrightline contracting playbook v3.1
DOCXHalverson Reid precedent bank
EMLOkafor email, March 2025
XLSXDeal intake workbook

Worked example on a synthetic matter. The context graph was drawn from the seven invented files listed. All parties, people, policies and figures are fictional.

See the consulting example