Review 4.0: learning from rejections inside the matter

A reviewer on the commercial team of a Nordic full-service firm sent us a screenshot in March. It showed the same proposed redline on a limitation-of-liability clause, declined four times across four turns of the same MSA. Her note read: “I know. It knows I know. Why is it still asking?”

She was right. Until now, Review treated each pass over a document as the first. The playbook said the cap should be twelve months’ fees; the draft said six; so Review flagged it, every time, regardless of what the reviewer had already decided on this deal. That is the correct behaviour for a playbook. It is the wrong behaviour for a matter.

Review 4.0, available to all accounts from today, changes this.

What changed

Rejections are now matter-scoped state. When you decline a proposed redline, Review records the decision against that clause, in that matter. On the next pass over the same document, or a counterparty’s revised draft of it, Review will not propose the same change again. The underlying flag remains (the draft still departs from the playbook), but it moves to a collapsed “Settled on this matter” list in the side panel rather than appearing inline.

A reason, if you want to give one. The decline control now offers four one-click reasons: commercially agreed, not our position on this deal, wrong clause, and too aggressive. A free-text field remains. Reasons are optional. In the beta, 61% of declines carried one, which was higher than we expected.

Reasons change what happens next. Wrong clause tells Review the anchor was off, and it re-runs detection on the surrounding text. Too aggressive tells Review to offer the playbook’s next fallback position rather than the primary one. Commercially agreed and not our position on this deal simply mute. The distinction matters: a reviewer who declines because the proposal was clumsy should not lose the flag altogether.

Nothing escapes the matter. A rejection does not edit the playbook. It does not inform another matter at the same firm. It is not used to train anything, consistent with our no-train guarantee. If a pattern of rejections suggests the playbook itself is wrong, that is shown to the playbook owner as a weekly digest (“this fallback was declined 14 times across 9 matters”), and the owner decides.

Why we built it this way

We considered the obvious alternative: let rejections quietly tune the firm-wide playbook. We rejected it for two reasons.

First, a matter is not a sample. One associate declining a cap position on one deal tells you about that deal. Eight associates declining it on eight deals might tell you about the playbook, but it might also tell you that a particular counterparty has unusual leverage this quarter. Neither should silently change the firm’s stated position.

Second, playbooks are owned. At most firms we work with, a named partner or knowledge lawyer signs off on the fallback ladder. Software that edits it without them undermines the thing that makes the playbook worth encoding in the first place.

So the matter remembers. The playbook is told. A person decides.

What to watch for

Muting is visible, not silent. The “Settled on this matter” list is one click away and shows every suppressed flag with its reason and the date. We would rather you see what Review is not telling you than wonder.

Severity is preserved. A High flag that you decline stays High in the settled list. The export to the closing memo includes settled items under a separate heading, so the final record shows what was raised and what was conceded.

Counterparty redrafts are matched by clause, not position. If the other side moves the cap clause from 14.2 to 15.1 and rewords it, Review will usually still recognise it as settled. Usually. Where the rewrite is substantial, the flag will reappear, and that is deliberate: a materially different clause deserves a fresh look.

Shared matters. When several reviewers work the same matter, a rejection by one applies to all. We debated per-user state and decided against it; a matter has one position, not one per reviewer.

Known limitations

• Settled state applies within a single document and its revisions. It does not yet carry across documents in the same deal bundle (an MSA and its SOWs, for instance). We are testing this.
• The Word add-in needs build 3.8.2 or later. Older add-ins still work, but rejections made there are not recorded as matter state until the add-in is updated.
• Documents reviewed through the DMS connector before 4.0 have no rejection history. There is no retroactive import.
• The weekly playbook digest is email only for now.

Upgrade notes

There is nothing to migrate. Existing matters start with an empty settled list. Playbooks are untouched. Severity tiers configured in 3.6 and the Word integration from 3.8 carry forward without change.

If you want the old behaviour back, a matter-level toggle restores it. Fewer than one in twenty beta accounts used it, and most of those turned it off again within a fortnight.

A good reviewer does not raise the same point four times in one negotiation. Review now does not either.