Skip to content

Expected Evidence & Freshness

Most evidence libraries fail the same way: they fill up, then quietly age. Nobody notices, because a full library looks like a healthy one.

Expected Evidence turns the implicit expectation — "there should be a quarterly access review for this control" — into a governed record, and the freshness engine tells you when reality has drifted from it.


Expected Evidence

An expected evidence requirement states, for a control: what artefact or observation should exist, covering what, at what cadence.

ElementPurpose
ControlWhat it evidences
DescriptionWhat should exist
CadenceHow often it should be produced
Acceptance stateWhether the requirement has been accepted into governance

The proposal and acceptance flow

Expected evidence follows the same maker-checker pattern as everything else governed in OrviQ.

Route inPermission
A person defines it manuallyexpected_evidence.propose
AI suggests it from an approved design adequacy assessmentexpected_evidence.ai_suggest plus the AI entitlement

Either way, it must be accepted by someone with expected_evidence.accept — and never by the person who proposed it.

Why AI suggestion is separately permissioned

expected_evidence.ai_suggest generates recommendations. expected_evidence.accept adopts them. Holding the first without the second means you can generate suggestions and adopt none of them — the same split that keeps AI advisory throughout the platform.


The freshness engine

Freshness evaluation is deterministic and AI-independent. It compares, for each accepted expected-evidence requirement, what should exist against what does, and reports the operational state.

It requires no AI entitlement. Manual freshness governance works on core control assurance alone.

States

StateMeaningWhat to do
CurrentEvidence exists within its freshness windowNothing
DueEvidence is approaching the end of its windowSchedule its production
OverdueThe window has passed with no new evidenceChase; the control's evidence is degrading
MissingNo evidence has ever been produced for this expectationInvestigate whether the control operates at all
StaleEvidence exists but is beyond its windowSame as overdue, viewed from the evidence side

Running an evaluation

The freshness sweep is idempotent and tenant-wide. Running it twice produces the same result; it changes no compliance position and creates no evidence.

Requires evidence_freshness.run.


Three distinct problems

These are frequently conflated and have completely different remedies.

ProblemWhat it meansRemedy
MissingThe evidence never existedThe control may not be operating. Investigate the control, not the evidence.
StaleThe evidence existed and has aged outThe control may still operate; the evidence collection has lapsed. Fix the collection.
InsufficientThe evidence exists and is current but covers too little of the populationCoverage problem. Extend the collection to the uncovered subjects.

Insufficient evidence is the hardest to see

Missing and stale evidence announce themselves. Insufficient evidence looks like healthy evidence — a current, passing, recent artefact — and is only visible through coverage. This is why OrviQ always reports coverage with its numerator and denominator rather than as a bare percentage.


How freshness reaches assurance

Freshness feeds the evidence status dimension:

Underlying assertionsEvidence status
All active and within windowcurrent
Mixture of fresh and stale or missingmixed
All beyond windowstale
None presentmissing

Evidence status then contributes to control effectiveness and requirement satisfaction — but never on its own. Stale evidence produces not_assessed or partially_effective rather than ineffective, because aged evidence is an information gap, not a control failure.

See Control Effectiveness.


Setting sensible cadences

Match the cadence to the control, not to a reporting rhythm.

Control operatesSensible expected cadence
Continuously, automatedDaily observation
Quarterly, proceduralQuarterly, with a window slightly longer than the quarter
Annually, governanceAnnual, 12-month window
On an eventPer event, with expectation defined by occurrence rather than date

A quarterly review with a 90-day window will show as overdue the moment the quarter turns, before anyone has had a chance to perform it. A slightly longer window — 100 days, say — reflects how the work actually happens without letting evidence drift by a quarter.


Permissions

ActionPermissionEntitlement
View expected evidence and coverageexpected_evidence.readcontrol_assurance
Manually define an expectationexpected_evidence.proposecontrol_assurance
Generate AI-recommended expectationsexpected_evidence.ai_suggestai_risk_intelligence
Accept or reject a suggested expectationexpected_evidence.acceptcontrol_assurance
View freshness states and due datesevidence_freshness.readcontrol_assurance
Run the freshness evaluation sweepevidence_freshness.runcontrol_assurance

Example

Control CTL-2026-0044 — Privileged Access Management. Three accepted expected-evidence requirements:

ExpectationCadenceCurrent state
Daily MFA enforcement observation for all privileged accountsDailyCurrent — 214 of 214 fresh
Quarterly privileged access recertification per in-scope systemQuarterlyOverdue — last produced 118 days ago
Annual attestation by the Head of IT SecurityAnnualCurrent — produced 4 months ago

Reading this: the automated evidence is healthy. The procedural quarterly review has lapsed by nearly a month. The annual governance attestation is fine.

Effect on assurance: evidence status is mixed, not current. Control effectiveness is partially_effective, not effective. Requirement satisfaction on the obligations this control serves is partially_satisfied.

Effect if nobody had declared the expectation: the daily collector would still be passing, effectiveness would read effective, and the lapsed quarterly review would be invisible until an auditor asked for it.

That difference is the whole argument for expected evidence.


Troubleshooting

"Everything shows Missing." Expectations have been declared but not accepted, or no assertions have been recorded against them. Check acceptance state first.

"Evidence is current but the expectation shows overdue." The evidence is linked to the control but no assertion was recorded from it, or the assertion's observation timestamp predates the window.

"The freshness sweep changed nothing." It is idempotent and read-only with respect to compliance state. It reports; it does not remediate.

"An expectation cannot be accepted." You need expected_evidence.accept, and you cannot accept your own proposal.


OrviQ Enterprise Governance, Risk & Compliance Platform