Review guide

Compliance Evidence Review: Building an Evidence Index

Most compliance review failures are evidence failures: the control exists, but the proof is a screenshot with no date, no owner, and no link to the control it is meant to support.

Gemma4.net review library Updated 2026-11-13 中文版

Evidence is not assertion

A policy statement says what should happen. A control description says how it is enforced. Evidence shows that it happened, on a date, for a population, with a reviewer. Reviews go wrong when a document is accepted as evidence of itself: a signed policy proves a policy was signed, not that access reviews were performed or that exceptions were approved.

The first discipline is to write the claim each item is meant to prove. “We review privileged access quarterly” is a claim. The evidence is the query used, the population, the reviewer, the date, and the outcome, including exceptions and their resolution. If the file set does not contain those elements, the claim is unsupported, however confident the narrative.

Build the index before you read the evidence

An evidence index turns a folder into a reviewable structure. At minimum, record the control reference, the claim, the evidence file name, the period covered, the owner, the preparer, and the review status. Without an index, reviewers spend their time opening files to discover what they are, and the same file is reviewed twice while another is missed.

Keep the index in a format that survives export, such as a spreadsheet or a plain table in the review memo. Each row should be independently checkable: a reader should be able to open the referenced file and confirm the period and the owner without asking you. If a row cannot stand alone, it is a note, not evidence.

Number the evidence files and reference the numbers in the narrative. When a document is replaced, add a new row rather than overwriting the old one; the history of what was reviewed on which date matters if the conclusion is challenged later.

Judge sufficiency and appropriateness

Sufficiency is about quantity: is there enough evidence across the period and the population? Appropriateness is about quality: is this the right kind of evidence for the claim? A management self-assessment is weaker evidence than a system-generated log, and a sample of five tickets is weak support for a claim about a population of thousands unless the sampling method is described.

For population claims, ask for the population definition and the completeness check. A user access review that starts from a list produced by the same team that administers access has a completeness risk. A reconciliation against an independent source, even a simple count, is worth more than a larger sample from the same list.

For period claims, check the boundaries. Evidence dated inside the period does not prove the control operated throughout it. Where a control is continuous, ask for the monitoring or alerting evidence rather than a single screenshot.

Triage a finding with likelihood and impact

Set likelihood and impact to get a triage band and the next action. The scale is a working heuristic for internal review, not a regulatory rating.

Handle exceptions explicitly

Exceptions are normal; undocumented exceptions are the problem. For each exception, record what happened, how it was detected, who approved it, what the remediation was, and whether the control was changed as a result. An exception with a clear owner and a closed remediation is stronger evidence of a working control than a clean sample with no explanation of how exceptions would be handled.

Distinguish design exceptions from operating exceptions. A control designed without a second reviewer is a design gap that no amount of operating evidence will fix. A control that operated outside its threshold once is an operating exception and usually needs a remediation date, not a redesign.

Avoid the temptation to rewrite evidence. If a screenshot lacks a date, the correct action is to obtain a dated export or to record the gap, not to annotate the screenshot as if the date were known. Reviewers who reconstruct evidence destroy the audit trail they were asked to protect.

Triage findings with a consistent scale

Not every gap deserves the same response. A consistent triage scale lets the review reach a conclusion instead of producing an undifferentiated list. Likelihood and impact are enough for most internal reviews: how likely is it that the gap produces a bad outcome, and how bad would that outcome be if it did?

The scale should be written down before the findings are scored, otherwise the scoring follows the conclusion. Record the reasoning next to each score so a reader can disagree with the judgement without redoing the analysis. Where two reviewers disagree by one band, take the higher band and say why.

Access, retention, and confidentiality

Evidence often contains personal data, customer identifiers, or security details. Give access to the smallest group that can complete the review, and record who has it. Share extracts rather than full systems where possible, and redact identifiers that are not needed for the conclusion.

Agree the retention period before the review starts and record the deletion date. Evidence collected for a review should not become a permanent shadow archive. If the evidence is needed for an audit or a regulator, the retention decision belongs to the function that owns the obligation, not to the reviewer.

If an external reviewer or tool is involved, state what they can see, how long they keep it, and whether the material leaves your environment. These are the questions a security team will ask, and having the answers in the review memo saves a later round of escalation.

Write the conclusion so it can be relied on

A compliance review conclusion should state the scope, the period, the claims tested, the evidence reviewed, the exceptions found, and the residual risk. “Controls are effective” is not a conclusion; it is a hope. “Access reviews operated for three of four quarters; Q2 evidence was missing; remediation is scheduled for 30 November with the identity owner” is a conclusion someone can act on.

Separate what you verified from what management asserted. If the review relied on a management representation, say so. That distinction protects both the reviewer and the reader, and it makes the next review faster because the boundary is already recorded.

Keep the review reproducible

A review that cannot be repeated is a review that cannot be defended. Record the population, the sample, the query or filter used, and the date the evidence was extracted. If the systems change, a later reviewer should be able to see whether the difference in the result comes from the control or from the way the data was pulled.

Store the review file in one place with a version label, and keep the original evidence unedited alongside any extracts you create. Extracts are working documents; the originals are the record. When a finding is challenged, the ability to return to the unedited source is what settles the question.

Limits

This guide describes a review method. It does not establish compliance with any specific framework, regulation, or contract, and it is not a substitute for the auditors or counsel who own that determination. Frameworks differ on evidence requirements, sampling, and documentation, so confirm the standard that applies to your organisation before relying on any checklist.

The same evidence can support different conclusions under different frameworks. Record which standard you applied, and keep the version of the standard with the review file so that a later reader can see the criteria you used.

Next step

Scope a private review by email

Send the matter, file volume, formats, deadline, and decision question. A person confirms fit, handling, and timing before files move.