A review is a piece of work. Commission it like one.

A person is for judgement, not for checking patterns. If the question is whether a key is in the code or a statistic has a source, a check answers it for free.

Commission a review when the question is whether the argument holds, whether the tone fits, whether a regulated claim is defensible, or whether somebody senior would put their name on it.

That framing also controls cost. Review hours spent on mechanical defects are hours not spent on the parts only a specialist can see.

Discipline matters more than seniority. Application code needs an engineer, front-end accessibility needs someone who works in it, copy claims need an editor or a compliance reader, media needs somebody who handles provenance.

Check three things in a profile: the discipline, years in it, and whether they publish a rate and a turnaround. A reviewer who will not state a rate has not scoped the work.

The rating and completed-review count are useful, but they describe satisfaction rather than fit. A well-rated generalist is the wrong choice for a payments flow.

The brief is five lines, and each one removes a question the reviewer would otherwise have to ask.

Scope is the difference between a review and an open-ended consultation. Name the sections, the files or the flows, and say what is explicitly out of scope.

Where the material is confidential, share only what is under review. A reviewer asked to check eligibility wording does not need the commercial model.

Agree the turnaround as part of the scope. Standard and priority are different commitments, and a priority request that arrives without a deadline is not one.

A written verdict, the findings that support it, the changes the reviewer would make, and their name attached to a stated scope.

The findings matter as much as the verdict, because they are what you act on. A review that returns only a score is a measurement, not an assessment.

For anything consequential, keep the review with the work. It is the artefact you produce if somebody later asks who checked the claim and on what basis.

Three signals. The reviewer disagreed with something. They asked a question you had not answered in the brief. And they said no to a line of enquiry as outside what they could check in the time.

A review that agrees with everything, asks nothing and never narrows its scope was a signature, not a reading.

The counter-signal is a wall of observations with no priority. A good reviewer tells you which two things to fix and which eight do not matter for this audience.

Keep the reports. The next reviewer starts from the findings rather than repeating the mechanical pass, which reduces the hours you buy.

Track what recurring human findings have in common. When the same issue appears three times, it belongs in the brief or in the rule set rather than in the next invoice.

That is the only mechanism by which review cost falls over a year while the quality holds.