Decision Coverage: the missing layer after approval

Approval answers whether you may start. Coverage answers whether you may continue.

5 September 2026 · 7 min read

Most governance stacks are built for the meeting that grants permission. They are weak at the Tuesday six weeks later when a vendor model changes, a prompt is edited, or the exception rate doubles.

Decision Coverage is the layer that keeps the original conclusion honest. It watches the conditions that made the conclusion valid. When those conditions move, the record moves with them.

From the research

Source finding

McKinsey's trust work on agents argues that you cannot govern what you cannot see, and that inventory, ownership, approvals, monitoring and reconstructability have to be explicit. The 2026 trust survey finds agentic controls lagging wider responsible-AI maturity.

Solarascope analysis

Coverage is not a dashboard of model metrics. It is a state machine on a decision: Ready, Conditional, Limited, Suspended. The inputs are evidence, boundary, owner and conditions. The output is whether the organisation is still justified in relying.

Sources

McKinsey. Trust in the age of agents (2026)
McKinsey. State of AI trust in 2026: Shifting to the agentic era (March 2026)