What a reviewable output needs to make visible.
SnapOS's research spans several named concepts — decision identity, continuous legitimacy, semantic stability, mandate stability. Publicly, what matters is not their internal order of operation, but what any reviewable output needs to show.
It is a reference for what a decision needs to remain defensible.
A reviewable finding about an AI-supported decision needs to show, at minimum: what was relied on, whether the mandate behind it is still current, whether it can be witnessed by someone who was not in the room, and whether third parties can independently check the same output against the same rules.
Decision identity
Whether today's decision is still recognizably the one that was authorized. See the concept page →.
Continuous legitimacy
A decision must remain legitimate not only when issued, but for as long as it continues to be relied on — not a one-time approval.
Mandate stability
Whether the operating state still matches the authorized mandate over time. See the concept page →.
Semantic stability
Whether the meaning a system assigns to its own state has drifted, independent of whether its output performance looks unchanged.
The full technical specifications for each — including how they are formally modeled and how they relate to each other internally — are published as citable research artifacts. This page describes what they mean for a decision-maker, not how they are computed.
Conformance Suite
The public output contract, witness requirements and result labels.
Governance Evidence
What must be visible if a decision is later challenged.