Do-not-change locks
Tell reviewers which wording, parties, or figures must not be altered. DraftLens passes that intent into model context to reduce accidental edits.
Last updated 2026-05-11
Do-not-change fields are injected into reviewer context so models steer away from renaming parties, altering defined terms, or “cleaning up” numbers you need verbatim.
Locks reduce risk; they do not guarantee zero suggestions touching adjacent text—human review still matters.
Why it exists
The fastest way to lose trust in AI review is an accidental party rename or a “small” numeric edit. Locks make non-negotiable strings explicit so reviewers spend attention where humans actually want discretion.
When it matters most
- Contracts and deal documents with sensitive proper nouns and financial figures.
- Regulated boilerplate that must remain character-stable even if surrounding prose improves.
Where it can be limited
Adjacent wording may still change in ways that affect meaning near a locked span. Treat locks as strong guardrails, not a formal verification system.
What to verify manually
- That lock lists are complete (missing a lock is worse than having too many).
- That punctuation around locked spans still reads correctly after accepted nearby edits.