Only the preferred position is required, and even that becomes optional if you link a vetted clause instead.
Everything else is depth you add where the negotiation actually happens.
Language
What the AI writes.Linked clause from your library
Linked clause from your library
Instead of describing the wording you want, link the exact clause.
Search your clause library by name, or create a new entry inline with a name and the clause text.When a clause is linked, drafting reproduces that text verbatim and adapts the surrounding section to it, rather than generating a paraphrase.
The preferred position field then becomes optional notes, which is the right place for deviation rules or rationale that the clause text itself does not carry.Update the clause once in the library and every playbook that links it stays current.
Preferred position
Preferred position
The clause wording you want by default. The AI drafts to this exact position.Write it as language, not as a wish.
“Mutual cap at 12 months of fees paid, with a super-cap for data-breach liability” is actionable.
“Reasonable liability cap” is not.An Improve action sits under the field and rewrites a rough note into a cleaner position statement, with an undo.
Acceptable range
Acceptable range
How far you will flex in negotiation.
Give numeric ranges or alternative phrasings.
Anything inside this range passes review without escalation.
Guardrails
The limits that trip review.Fallback ladder
An ordered list of acceptable retreats, best first. When the preferred position is rejected, drafting takes the highest acceptable rung and review reports which rung a counterparty term actually lands on, instead of jumping straight to your worst case. A three-rung liability ladder might read: 12 months of fees, then 24 months of fees, then total contract value. This is the single highest-value optional field in a playbook. It is what turns “we want X” into a negotiation you can delegate.Walk-away floor
The non-negotiable line for this clause. A drafted or counterparty clause at or below it is forced to the most severe rating in review and to Needs attention in drafting, and it requires senior sign-off regardless of anything else.Escalation triggers
Words or phrases that need sign-off if they appear. The AI avoids them when drafting, and review flags them if a counterparty slips one in. Keep them short and literal, for example “uncapped liability” or “no cure period”.Numeric limits
Where a clause has typed constraints, the editor shows structured number fields instead of asking you to bury the number in prose. These are checked deterministically against generated text, so they are far more reliable than the same figure written in a sentence.
Clauses without applicable constraints show nothing here.
Governance
Who signs off, and how loudly a miss is flagged.Priority
How negotiable the clause is.Rationale
Why the firm holds this position: the precedent, the regulator concern, or the past deal that drove it. This is not decoration. It is injected into the drafting prompt so the model can argue the position, and it is shown to reviewers in compliance results. The same Improve action is available here.Approval level on deviation
If a draft strays from this position, who has to approve before it can go out.Conditional escalation
Raise the required sign-off automatically when the deal matches a rule. Each rule reads as a sentence: if an attribute compares a certain way to a value, then require a given level of sign-off.
Rules are evaluated at review time against the deal context you supply on the review form.
A rule that escalates to a level at or below the clause’s base approval can never fire, and the structure check reports it as an error.
Risk weight
How much a deviation on this clause matters, on a scale of 0 to 5. It drives how loudly review flags the clause and how much it moves the overall compliance score.
Standard clause types come with a sensible default weight.
Setting a value overrides it, and the field shows both the default and your override so you can reset.
Limitations
- Free-text fields are interpreted by a model. Precision in the wording is what makes enforcement precise.
- Numeric limits only exist for the clause types listed above. Everywhere else, a number has to live in the prose.
- Conditional escalation understands four attributes and nothing else. It cannot branch on counterparty identity, industry, region beyond governing law, or renewal status.
- A deal value threshold must be a plain number. Anything else is flagged as an invalid rule.
- Custom clause types carry no default risk weight and are not compared against the market-standard baseline.
- A linked library clause is reproduced verbatim, which means a stale clause in the library becomes a stale clause in every draft. Keep the library current.
- Escalation triggers match on language, so a counterparty who phrases the same concept differently will not trip them. Fallback ladders and numeric limits are the more robust guard.
Related
Building a Playbook
Create a playbook and add rules like these.
Sign-Off Gates
What approval levels and conditional rules do at review time.
Structure Check
The linter that catches missing ladders and dead rules.
Clause Library
Store vetted clause text and link it into positions.