Defined terms: no clause can be judged without its definition

A defined term is a word the contract has redefined for its own purposes, marked by a capital letter. Losses, Affiliate, Confidential Information and Services do not carry their dictionary meanings in a contract; they carry whatever that document assigns them. The capital letter is the signal that a reader should stop and look the word up.

The practical consequence for anyone reviewing at volume: a clause cannot be assessed from its own text. It has to be assessed with its definitions substituted in, because a definition somewhere else in the file can change the clause from routine to unacceptable without altering a word of it.

Five places the meaning is actually set

• The definitions clause. The obvious one, usually near the front, and the only one most tooling looks at.
• Inline parentheticals. The pattern "the Supplier's data centre (the Facility)" defines a term in the middle of an operative clause, far from any list.
• Schedules and annexes. Often with their own definitions, sometimes stated to apply only within that schedule.
• Amendments and side letters. A later instrument can replace a definition wholesale, which makes the operative meaning a function of the document family rather than the file — see which document wins when the MSA and the SOW disagree.
• Incorporated documents. A policy or service description brought in by reference can supply terms the main agreement never defines.

There is also the term that is capitalised and never defined at all, which is a drafting defect rather than a definition problem. On the generation side that failure has its own diagnosis in the generated draft that uses a term it never defines; on the review side it is a finding to raise, not a value to guess.

One word two schedules away, and the clause changes

Take a confidentiality clause permitting disclosure to the receiving party's Affiliates. Read alone, it is unremarkable — every mutual NDA has one.

The clause text is identical in both rows. Only the pointer moved. This is why a system that scores clause language without resolving definitions is measuring the wrong object — and why a playbook comparison run on unresolved text produces noise in both directions, part of what makes every clause come back flagged.

The same mechanism runs through indemnities, where the definition of Losses decides whether indirect and consequential loss sits inside the promise. An extraction that captures the indemnity clause and not its defined terms has captured half the obligation.

The resolution pass, before anything is scored

1. Harvest every definition in the family, recording for each one its term, its text, the document and page it came from, and its scope — whole agreement, one schedule, or one clause.
2. Detect collisions. The same term defined twice is normal, not an error: resolve it by scope first, then by precedence between instruments, then by date.
3. Bind occurrences to definitions. Each capitalised occurrence in the operative text gets a pointer to the definition that governs it at that position.
4. Flag the unresolved. A capitalised term with no definition, or with 2 candidates that scope cannot separate, is a finding for a person — never a guess.
5. Only then score. Any rule, rubric or playbook comparison runs on the resolved reading, and the resolution is stored so a reviewer can see which definition was applied.

The dangerous definition is never the one you cannot find. It is the one you found first, in the definitions clause, and never checked for a later replacement.

What skipping resolution actually costs

Two failures, and they point in opposite directions. A widened definition makes a clause look acceptable when the exposure behind it is not. A narrowed one makes a protection look present when it covers almost nothing — a warranty about the Services that means only the items listed in a schedule the reviewer never opened.

Both are silent. Neither produces a low-confidence extraction or an odd-looking clause, which is why resolution belongs in the pipeline rather than in a reviewer's discipline. It is also the step that decides whether the whole review was aimed at the operative text, alongside establishing that the version in front of you is the current one — the failure behind a review that is right about a superseded version.

Building the harvest-and-bind layer, with citations back to the page each definition came from, is a small and well-bounded piece of engineering, of the kind we describe under MVP and product builds. Neighbouring definitions sit in contract review and redlining, inside our work with legal teams.

• defined terms
• contract review
• clause analysis
• legal ops