Controls, evidence, and blind spots

The chain to investigate is control → execution → outcome → evidence. A configuration describes intent. An execution record describes a bounded event. An outcome observation describes the effect that was actually observed. Each supports a different claim. STABLE

What each artifact establishes

Artifact	Useful for	Does not establish by itself
Policy or charter	Declared purpose, scope, rules, ownership	That the control was installed or executed
Test result	Behaviour under identified inputs and conditions	Behaviour on every production path
Signed execution record	Integrity, signer, declared context, recorded decision	Truth of all inputs, sufficient control design, complete capture
Independent witness	Additional integrity or timing evidence under its trust model	Semantic correctness of the model or absence of harm
Downstream observation	A specified effect or state at the observed boundary	Every other effect or the legal cause of a loss
Population reconciliation	Relationship between counted and attested activity	Complete coverage if the total itself is incomplete

This table is the academy’s reasoning framework. OVERT is the public specification for independently verifiable runtime evidence; a conforming record must still be assessed within its scope and trust model. STABLE GROUNDED · OVERT Published baseline checked September 30, 2026.

Denominators need definitions

For an explicitly defined period and action type, report attested eligible actions divided by total eligible actions. Record who counted each side, how retries and failed actions are treated, and where activity could escape observation. STABLE

If the total comes from the same optional hook that creates receipts, the ratio describes that hook’s activity. It cannot establish the share of the entire deployment that was governed. Missing evidence can indicate a bypass, an outage, a retention gap, or an excluded path; investigate before assigning a meaning. STABLE

Capability stages

Use proposed, implemented, tested, deployed, and accepted for a defined use precisely. A local passing test is evidence of that test. A merged change is evidence of source state. Production effectiveness needs evidence from the deployed version, configuration, and workflow. An insurer’s acceptance needs that insurer’s decision. STABLE

Streaming creates a further boundary: stopping a stream after detection may leave earlier content delivered. Report prevention only when the prohibited output or action was stopped before the relevant effect. STABLE

Field drill

A customer has valid receipts for every event its gateway counted. The total count of direct model calls is unknown. What can you claim? The receipts can be verified for the governed events; deployment-wide mediation coverage remains unknown.

Next

→ Drift and change control

Drill this page →61 bank questions stand behind what you just read. Check it while it’s warm.