Reviewing for silence: how we detect missing mechanisms without generating noise
Identifying a poorly drafted clause is a matter of text comparison. Detecting a missing protection requires a structural map of the transaction.
Most review time is spent negotiating poorly drafted text. We argue over the definition of a material adverse change, or the precise formulation of an efforts standard. But the highest liability risk often comes from what is omitted entirely. An MSA without an audit right. A software licence without a transition-services mechanic upon termination. A DPA missing a sub-processor notification timeline.
Junior lawyers sometimes struggle with omissions because there is no text on the page to trigger their training. They must hold the ideal structure of the document in their head and notice the negative space.
Historically, automated systems shared this blind spot. Large language models are text prediction engines; they react to the context provided. If you ask a general-purpose model what is missing from a contract, it will suggest dozens of standard clauses, regardless of the deal’s size or context. This generates unacceptable noise.
Mapping the structural schema
We do not rely on open-ended prompts to find missing text. In Review, checking for silence is a deterministic process governed by the firm’s playbook.
When a draft is uploaded, ZAAN classifies the document type and maps its contents against a predefined structural schema. If the firm’s playbook dictates that every commercial lease must contain a casualty and condemnation mechanism, the schema expects to fill that node.
ZAAN-7B Counsel reads the entire agreement to see if the concept is addressed. It does not look for an exact heading. It looks for the legal mechanism. If the mechanism is absent, the node remains empty. A rule engine then checks the empty node against the severity tiers configured by the practice area. A missing waiver of jury trial might trigger a Note. A missing assignment restriction in an IP licence triggers a High severity flag.
Implicit coverage and false omissions
The difficulty lies in implicit coverage. A contract might not have a standalone survival section, but the confidentiality clause itself might state that it survives termination for three years.
Before Review flags a missing clause, it cross-references the empty schema node against the rest of the agreement. If the playbook requires a carve-out for gross negligence in the limitation of liability, and that carve-out is drafted into the indemnification section instead, the model registers the protection as present.
We classify this as a structural anomaly, but not a fatal omission. In this scenario, Review will surface a Note regarding the placement, rather than a High severity flag for a missing protection. We built ZAAN-7B Counsel specifically for this kind of clause-level reasoning. It evaluates the functional reality of the contract, not just its layout.
Restraint in the margin
When an omission is confirmed, the system must decide how to present it. Early in our development, we experimented with inserting missing text directly as an inline redline. This proved disruptive. Dropping three paragraphs of a firm’s standard force majeure boilerplate into the middle of a counterparty’s draft destroys the numbering and alters the reading flow.
Instead, Review handles omissions through the native Word comments API. We place a margin note at the most logical insertion point—usually the end of the relevant section, or just before the miscellaneous boilerplate.
The flag states the omission plainly: “Playbook requires a data-breach notification timeline (maximum 48 hours). Mechanism not detected.”
The reviewer can dismiss the flag if they have made a commercial decision to concede the point. If they wish to enforce it, they use the integration with Draft. A single click pulls the required phrasing from the firm’s precedent database, formats it to match the surrounding house style, and inserts it as a tracked change.
The cost of over-drafting
There is a temptation in legal engineering to build playbooks that demand every conceivable protection. We advise firms against this. A playbook that requires a thirty-page MSA structure for a low-five-figure software subscription will generate endless omission flags. This leads to alert fatigue, and eventually, associates will start ignoring the sidebar entirely.
To prevent this, playbook rules in ZAAN should be tied to deal variables. An escrow requirement should only trigger if the contract value exceeds a specific threshold. A mandatory governing-law check should adjust based on the jurisdiction of the counterparty.
Silence in a contract is not always an error. Sometimes, it is just a sense of proportion. A review system should only flag what is missing when the playbook demands it, and remain quiet when the omission is a commercial choice.