Multi-document reviews and supplementary context

A real-world contract rarely arrives as one document. It comes with schedules, side letters, an executed precedent for reference, an email chain explaining the commercial intent, a policy document that has to be cross-checked. Clment’s review model handles this directly — one review, multiple documents, with each document classified by how it should be treated.

The two attachment modes

When you launch a review on a contract that has attachments, Step 1: Documents lets you tick each attachment and pick one of two modes:

• Review — findings are raised against this document. The AI evaluates it against the playbook the same way it evaluates the primary contract. Multiple Review-mode documents in one session means findings can span all of them.
• Supporting — the AI sees it for context but won’t raise findings against it. Use this for documents that inform the review (the precedent contract, an executive summary, a side-letter you need to respect) but aren’t themselves what you’re reviewing.

The picker shows the actual mode under each attachment — [Review] or [Supporting] — and the primary contract is implicitly in Review mode.

Tick an attachment to include it. Review — findings raised against this document. Supporting — the AI sees it for context but won’t raise findings against it.

When to use which

Mark Review when:

• The attachment is part of the contractual relationship (Schedule A pricing, the DPA, the SLA exhibit). Findings against the SLA’s uptime commitment matter just as much as findings against the main MSA’s liability cap.
• You want a single redline output that covers everything.

Mark Supporting when:

• The attachment is precedent or context — last year’s executed version, a competitor’s standard wording, a counterparty’s prior position.
• The attachment is policy — your internal information-security policy that the contract must comply with. You want the AI to flag conflicts but not raise findings against the policy itself.
• The attachment is the cover email explaining the deal. It frames the review without being part of it.

A common pattern: an MSA upload with its DPA and SLA as Review-mode attachments, plus the counterparty’s prior-year executed MSA as Supporting context. The review produces findings on the MSA, DPA, and SLA, with the AI implicitly informed by “what they accepted last time”.

Additional context (ad-hoc materials)

Sometimes the relevant reference material isn’t already an attachment on the contract. Maybe it’s:

• An email thread you want the AI to read before reviewing.
• A clause snippet from a precedent contract you don’t want to upload as a separate contract record.
• An internal policy passage.
• Negotiation notes from a call.

For this, the dialog has a separate Additional context field on Step 1:

Ad-hoc reference material for this review only — emails, cover sheets, policies, precedent snippets. The model sees them alongside the contract but raises findings against the contract, not these materials.

You can either:

• Upload files (PDF, DOCX, TXT) — the text is extracted and used as context.
• Paste text — for a quick clause snippet or email chunk where uploading a file is overkill.

Context materials are scoped to this review only — they don’t become attachments on the contract, they don’t appear elsewhere in the workspace, and they’re not visible to teammates outside the review. They’re stored on the review record for reproducibility, so re-opening the review later shows you exactly what was in the model’s context window.

The distinction from Supporting attachments is important enough that the dialog calls it out: “Not for attachments on the contract itself — set an attachment’s mode on the previous step.” Use Supporting mode for things that live on the contract; use Additional context for one-off material that’s only relevant to this review.

How findings work across documents

When a review covers multiple Review-mode documents, each finding is anchored to the specific document and clause it’s about. The finding list shows which document each finding belongs to, and the redline output produces one tracked-changes file per Review-mode document — so a vendor MSA + DPA + SLA review gives you three redlined .docx files.

The AI is told about ALL the documents in scope, so it can raise findings that span them (“the DPA’s data-retention period conflicts with the MSA’s record-keeping obligation”) — those are tagged against the document where the finding most naturally lives, with a cross-reference in the explanation.

Credit consumption

Multi-document reviews don’t multiply the per-review credit cost (1 credit) — they’re one review session, even when 4 documents are in scope. The AI runs longer (a few minutes for a typical 4-document review vs ~90 seconds for a single contract) and consumes more tokens internally, but billing is per-session.

See also

• Review scopes — whole document, tracked changes, version comparison
• Reviewing contracts
• Generating redlines
• Understanding finding verdicts