Contract drafts, redlines and negotiation threads
The negotiation history behind commercial agreements: successive drafts with tracked changes, margin comments, the emails or tickets where terms were discussed, and the signed result. Labs use it to train contract-drafting and negotiation agents on how clauses actually move between parties.
What a record contains
One negotiation: draft versions in order with redlines, comments by each side, linked correspondence, internal approvals and the executed version.
| Field | What it holds |
|---|---|
matter_id, agreement_type | MSA, DPA, NDA, SOW, order form |
drafts[] | Version, party, changes and comments |
correspondence_refs[] | Emails or tickets about the terms |
approvals[] | Internal sign-offs by role |
executed_ref | The signed version |
{
"matter_id": "L-0915",
"agreement_type": "MSA",
"drafts": [
{"version": 2, "party": "counterparty",
"changes": [{"clause": "12.1 Limitation of liability",
"from": "12 months of fees", "to": "24 months of fees"}],
"comment": "Our standard for vendors with data access."},
{"version": 3, "party": "company",
"changes": [{"clause": "12.1", "to": "18 months of fees; data breach super-cap 2x"}]}
]
}How AI labs use it
- Redlining agents
- Real markups and counter-markups.
- Negotiation strategy
- Which positions were conceded and which held.
- Playbook evaluation
- Compare a model’s markup with the company’s actual positions.
Typical preparation requirements
Agreed with the supplier before any work begins. Typical requirements include:
- Counterparty names and identifying deal terms redacted
- Agreements whose confidentiality terms prohibit disclosure excluded or cleared with the counterparty
- Privileged legal advice excluded
Every dataset has a documented owner and confirmed licensing rights. See data governance on sourcex.si.
What makes a strong package
- Full draft chains, not just final versions
- Many agreements of the same type
- Documented fallback positions
Compared with public datasets
Public sets such as CUAD and LEDGAR are useful references, but limited as enterprise training data. The legal and compliance category page compares them with licensed data.
Who typically holds it
- B2B software companies
- Procurement teams
- Healthcare and financial services vendors
Know a company like this?
Introduce the company to SourceX. If its data deal closes, you can earn up to $100,000 in referral fees, paid after the buyer accepts the data and SourceX receives payment.
Refer a companyQuestions
Isn’t contract confidentiality a blocker?
It’s the main check. Many agreements restrict disclosure of their terms, so packages are redacted and, where needed, agreements are excluded or cleared with the other side.
Need this data for a model?
Describe what you need: domain, volume, history, format and licensing terms. SourceX looks for companies that hold it and can license it.