SYNCLOGICSL-DG-001
How do you detect scope drift during an AI session?
Drift Guard compares the evolving reasoning chain with the declared Scope Lock, source set and confidence boundary.
AuthorWalter Shepherd
Published26 July 2026
Updated26 July 2026
Version2.1.0
StatusPublic Research and Pilot Release
MethodSL-DG-001
Failure mode
What this method prevents
The session gradually operates against a different scope without a visible boundary crossing.
Illustrative example
A session declared for adults begins introducing paediatric evidence without declaring the change.
How it works
A controlled sequence
- Recheck current claims against Scope Lock.
- Verify sources remain registered.
- Audit confidence against evidence quality.
- Test inferential continuity.
- Repair or reset when drift is detected.
Reliance note
The method supports governance; it does not self-authorise a decision.
Application must be proportionate to the task, evidence and consequence level. The output should state what has been registered, what remains uncertain and which authorised human or organisation controls release.