Named at class level. Nothing about how SnapOS detects them.
These are short, public notes on failure mode classes that recur across AI-enabled decisions. They name and bound a class of problem. They do not describe how SnapOS finds it — no detection logic, scoring, trigger chains or the failure library behind them.
| Document type | Public Canon Reference |
|---|---|
| Status | ACTIVE PUBLIC SPECIFICATION |
| Version | v16.3 |
| Last revised | 2026-09-19 |
| Maintainer | SnapOS Foundation / Marko Chalupa |
| Canonical URI | https://snapos.org/en/failure-modes.html |
| Cite as | SnapOS.org, Failure Mode Classes — Public Boundary Notes, v16.3, accessed YYYY-MM-DD. |
Mandate Drift
Definition: Execution continues after the authority or mandate that originally justified it has changed or expired.
Public relevance: One of the most common ways an AI-enabled decision quietly stops being authorized without anyone deciding that it should stop.
What this does not prove: It does not identify who is at fault, nor that the underlying system is technically incorrect — only that its original authorization is no longer current.
Unsupported Transfer
Definition: A claim, finding or output moves across workflow, regime or decision context without a record of how or why it moved.
Public relevance: Relevant wherever a result produced for one purpose is later relied on for a different one.
What this does not prove: It does not establish that the transfer was made in bad faith, or that the destination use is invalid — only that the move itself is unrecorded.
Evidence Boundary Failure
Definition: A claim is supported by evidence within one context but is reused beyond the boundary that evidence actually covers.
Public relevance: Relevant wherever a finding validated for one dataset, population, version or jurisdiction is applied to another.
What this does not prove: It does not show that the original evidence was flawed — only that its boundary was crossed.
Green-Light Failure
Definition: Local controls and checks all report as passing while the overall decision trajectory has drifted from its original basis.
Public relevance: Relevant because dashboards, checklists and individual reviews can each be accurate in isolation while the bigger picture has already changed.
What this does not prove: It does not mean any individual control was misconfigured or dishonestly reported.
Semantic Surface Replication
Definition: SnapOS-like vocabulary or narrative is reproduced without the underlying audit-output contract that gives those terms meaning.
Public relevance: Relevant for telling apart language that is SnapOS-inspired from an output that is actually SnapOS-compatible or SnapOS-conformant.
What this does not prove: It does not imply bad intent — SnapOS-inspired use of the vocabulary is legitimate and expected on its own terms.
Mandate Drift is discussed at length, as a standalone concept article, on Mandate Drift →.
These notes are boundary-level companions to the Conformance Suite, which defines the witnesses, result labels and conformance gates an output must satisfy — independently of any specific failure mode.