Review 4.1: governing-law and jurisdiction mismatch detection

Across our active accounts, about one draft in forty reaches its third negotiation turn with a governing-law clause that no longer agrees with its jurisdiction clause, its arbitration clause, or one of its schedules. The mismatch is rarely anyone’s intention. It is what happens when two sides each edit the boilerplate they care about and nobody re-reads clause 24 against clause 25 and Schedule 4.

Review 4.1, available to all accounts from today, reads them together.

What changed

Governing-law and jurisdiction mismatch detection. Review now extracts, from the main agreement and every attached schedule, SOW, order form and DPA: the governing law; any exclusive or non-exclusive jurisdiction clause and its forum; any arbitration clause, its seat, rules and language; the law governing the arbitration agreement where stated separately; notice addresses and the parties’ stated domiciles; and any defined term such as “Applicable Law” that points at a jurisdiction. It then checks the set for consistency and flags disagreements.

Severity follows consequence:

• High. The main agreement and a schedule name different governing laws with no precedence clause. An exclusive jurisdiction clause and an arbitration clause both claim the same disputes. A governing law that is not a recognised system of law (a surprisingly common artefact after a jurisdiction swap: “the laws of Geneva”).
• Medium. Arbitration seat and governing law differ without an express choice of law for the arbitration agreement. A non-exclusive jurisdiction clause paired with an asymmetric right that only one party may exercise, where the playbook has no position on asymmetry. A DPA whose governing law differs from the main agreement where the DPA does not say it prevails.
• Note. Governing law and forum differ in a combination the playbook has whitelisted as intentional. Notice addresses in a third jurisdiction. A defined term “Applicable Law” that is narrower than the governing-law clause.

Whitelisted pairs. Intentional splits are common and lawful. English law with a Singapore seat; Swiss law with a Paris seat; New York law with exclusive English jurisdiction in a financing. The playbook can now hold a list of acceptable governing-law and forum pairs per practice group. A matched pair produces a Note at most. Unlisted pairs produce Medium.

Precedence awareness. Where the agreement has an order-of-precedence clause, Review reads it and suppresses schedule-level mismatches that the clause resolves, noting the resolution in the flag rather than raising a conflict.

ZAAN-7B Counsel in the background pass. Since October the in-house model has handled the inline pass. From 4.1 it also handles most of the document-level background pass, with routing to a larger model retained for long-clause summarisation and thinly covered jurisdictions. The trace view shows which model answered every flag.

Quieter Note flags. Note-severity flags no longer appear inline by default. They collect in the side panel and surface as a count. Reviewers at eleven beta accounts asked for this almost unanimously; the inline margin was becoming a place to ignore things. High and Medium are unchanged. The default is per user and can be reverted.

Why

Mismatch detection is a document-level check and could not be done inline within the 500-millisecond budget. It needed the background pass to mature, and it needed a model that could hold a forty-page agreement and its schedules in view at once without routing cross-region. Both arrived this autumn.

The whitelist exists because the first beta build flagged every English-law, Singapore-seat agreement at Medium, and a Swiss M&A boutique told us, correctly, that we were flagging their house style. The playbook is where house style lives. The whitelist puts it there.

What to watch for

The first run on an existing matter will raise flags that have been there for months. That is the point, but it can be an uncomfortable morning. The settled-on-this-matter list from 4.0 applies; decline what is intentional and it stays declined.

Arbitration clauses that name institutional rules are checked against those rules’ default seat where the clause itself is silent. If the rules have changed their default, Review uses the version in force at the agreement’s date, where known, and says so.

US state law. Review distinguishes state-level governing law and forum for the ten states most represented in our index. Other state combinations are checked at country level only, and the flag says so.

“Laws of England and Wales” and “English law” are treated as the same system. “Laws of the United Kingdom” is flagged as Note, because it is not one.

Known limitations

• Governing-law extraction from scanned schedules relies on OCR quality. Low-confidence extractions are shown with their confidence and are not used for High flags.
• Mixed-language agreements with a prevailing-language clause are checked in the prevailing language only.
• Mismatch detection does not yet run across separate documents in a deal bundle that are not attached as schedules. This is the same gap noted in 4.0, and we are still testing a fix.
• Whitelisted pairs are per practice group, not per matter. A per-matter override is on the list.

Upgrade notes

Nothing to migrate. Existing whitelists are empty; the playbook owner should populate them before switching on mismatch detection, or expect a week of Medium flags on intentional splits. The Word add-in needs build 4.0.1 or later for the side-panel Note count.

Governing law is the clause everyone assumes someone else checked. Review now checks it.