The review is correct, and the contract was amended two years ago

Every finding in the report is correct. The limitation of liability is what the report says it is, the termination notice period is right, the assignment restriction is quoted accurately. And the whole thing is worthless, because the agreement was amended and restated 2 years ago and the file under review is the 2022 original that nobody unfiled.

This failure is worse than an inaccurate review, because inaccuracy shows up under scrutiny and this does not. A reviewer checking the report against the document will find perfect agreement. The error is one level up: not what the clause says, but whether that clause is still the clause. Operativeness has to be established before analysis begins, and it is almost never treated as a step.

Name every instrument that touches this agreement, then count the files

The check is a list, and it takes about 10 minutes for a contract with a normal history. It is worth running as a gate on every review rather than as an investigation after something goes wrong, because the cost of running it is fixed and the cost of not running it is not.

1. Start from the executed original and read its amendment provision. It states the form a variation must take — usually writing, signed by both parties — and that tells you what a valid instrument looks like before you go looking for one.
2. Ask the repository for every document naming the same counterparty and the same subject matter, not for documents linked to this one. Linkage is the thing you are testing; using it as the search key assumes the answer.
3. Read the recitals of the most recent amendment you find. Amendments almost always recite the chain that precedes them, so Amendment 4 will name Amendments 1 to 3 and tell you how many instruments should exist.
4. Count the amendments you hold against the highest number recited. A gap is a missing document, and a missing amendment is the single commonest cause of this failure.
5. Ask the deal team directly whether a side letter exists. Not the repository — the people. Side letters are frequently agreed outside the main filing route and never make it into the system at all.
6. Check for instruments that change a party rather than the text: novations, assignment consents, and change-of-control notices. These leave the wording untouched and make the review's conclusions about the counterparty wrong.
7. Record the answer as a version pointer with a date, not as a feeling. The review's conclusions are only valid against the stack you just enumerated, and in 6 months nobody will remember what that was.

6 instruments that change an agreement without replacing it

Each row alters the operative position while leaving the original file intact and plausible. The last column is what makes each one hard to find, and it is the part worth designing against.

The amendment that was filed, and filed nowhere near the agreement

In most repositories an amendment is a document with its own record, its own counterparty field and its own date, and nothing in the data model says it modifies anything. A person filing it does the right thing by every rule they were given. Six months later a review is run on the agreement, the agreement record contains 1 file, and that file is complete, executed and superseded.

The repair is structural rather than procedural. A contract record needs a parent-child relationship, so that opening the agreement produces the stack rather than the file, and the amendment cannot exist as an orphan. Where that relationship is missing, the recitals are the fallback: Amendment 3 naming Amendments 1 and 2 gives you a chain you can verify against, which is why reading the recitals is step 3 of the check and not an afterthought.

The side letter the deal team is still holding

Side letters are the hardest instrument in the list because they are the one most often deliberately kept out of the main flow. They get agreed late, they resolve a single sticking point, they are signed by people who were not thinking about the filing convention, and they are frequently not recited anywhere. A review can enumerate an entire amendment chain correctly and still be wrong because of 1 page nobody scanned.

There is no technical fix for a document that was never captured, so treat this as an intake question rather than a search problem: every review request should ask, explicitly, whether a side letter exists, and record the answer as either a document or a stated negative. A recorded "no side letter, confirmed by the deal lead" is evidence. Silence is not.

The report agrees with the document perfectly. That is the problem: it was checked against the file, and nobody checked the file against the deal.

The consent that changes a party without changing a word

A novation or an assignment consent leaves the agreement text untouched and changes who is on the other side of it. Every text-based check passes, every clause reads correctly, and the conclusions about the counterparty are wrong. This matters most where the analysis is directional — an obligation that is acceptable from a large customer may be unacceptable from the entity it was assigned to, and the same wording carries opposite risk depending on which side you are on, the branching argument in mutual and one-way review paths.

It also breaks the counterparty view the firm relies on downstream, because the entity in the agreement and the entity actually performing are now different names for a relationship the repository thinks is 1 party — a resolution problem in its own right, covered in the same counterparty appearing under nine names.

What operative has to mean before the record can be trusted

Every review should emit the version it was run against as an explicit, dated pointer to a stack of instruments, not as an implicit assumption about a file. That single field is what makes a stale review detectable: when Amendment 5 arrives, a query can find every review whose pointer predates it, and the alternative is discovering the problem one contract at a time. Binding that pointer once, at the point the review is created, rather than retyping it into 3 systems, is the same discipline as binding matter data to template fields once.

Two adjacent facts round it out. Where several documents govern at once, deciding which prevails is its own rule and not a version question, argued in which document wins when the MSA and the SOW disagree. And where the review standard is the firm's own past deals rather than a written playbook, operativeness becomes doubly load-bearing: a precedent set built from superseded text teaches the wrong positions, which is the quiet risk inside judging a clause against a playbook or against your own past deals.

None of this touches how the operative text is then read. Conformed and executed copies bring their own pathologies once you have identified the right file — initialled manuscript changes, schedules appended out of order, signature pages spliced from counterparts — set out in extraction that falls apart on the signed conformed copy. Wiring the version gate into the tool a legal team actually opens is ordinary internal tools and operations software, and it belongs with the rest of contract review, clause risk and redlining in our work with legal teams.

• amendments
• operative version
• contract review
• diagnosis