Model usage boundaries
Where AI drafts are helpful — and where ClosePilot stays deterministic.
How we're different
- LLMs draft text; policies and humans govern actions.
- Deterministic recon + validation for safety-critical steps.
- Audit trail links drafts, approvals, and resulting actions.
Trust rule: AI drafts, humans decide for sensitive outcomes.
Allowed (drafting)
- Evidence request drafts (email/slack)
- Variance note drafts from structured inputs
- Executive summaries for pilot reports (illustrative)
Outputs are reviewable and can be edited before sending.
Not allowed (deterministic)
- Posting journal entries
- Approving an audit pack export
- Recon matching decisions without human review thresholds
Humans approve sensitive decisions. No silent overrides.
Safety mechanisms
- Policy gates with explicit approvals.
- Schema validation for structured outputs.
- Exception inbox when uncertain.
- Immutable audit log with evidence links.
Not legal advice. Your controls and approvals remain your responsibility.