Public synthetic proof
Relevant context can still be wrong for the exact decision.
Run four bounded scenarios and inspect the decision record. The demonstration separates the task, the context supplied to the workflow, applicability checks, provenance and the resulting verdict.
Interactive proof
Change the boundary. Watch the decision change.
Each scenario keeps the basic topic relevant while changing a boundary that matters: version, geography, authority conflict or a clean current match.
Task context
- Question
- Product
- Plan
- Version
- Region
- Decision date
Carried answer
Why the context does or does not fit
| Boundary | State | Reason |
|---|
Verdict
Decision record
What this demonstrates
A retrieval result can remain relevant while failing the boundary required for reliance.
All four scenarios can retrieve text that appears relevant. The operational question is whether that text belongs to the exact entity, version, plan, geography, time period and source role required by the task.
Semantic relevance is insufficient
A source can discuss exactly the right feature and still be wrong for the release or region in front of the user.
Uncertainty must survive
When authoritative sources conflict, the defensible output is not an invented winner. The decision should remain unresolved until the conflict is adjudicated.
Evidence changes the conversation
A reviewable record lets an operator see what was supplied, what boundary failed and why the workflow accepted, held or rejected the context.
Explore the Context Integrity Failure Atlas for eight reference failure classes →
Claim boundary
This public proof is deliberately narrower than a commercial assessment.
It does show: explicit task scope, context scope, provenance, applicability checks, preserved uncertainty and a reviewable decision record.
It does not show: customer-system access, production deployment, universal compatibility, compliance certification, autonomous remediation or a guarantee that a source is factually true.
Run this discipline against one real workflow.
The Context Integrity Assessment applies the method to one defined AI, RAG, agent, support, retrieval or knowledge workflow using agreed, authorized sources and human-reviewed findings.