Make the acceptance criteria observable
Agree a brief version and a small set of defect categories before production. Useful categories include missing output, wrong variant, incomplete mask, texture loss, colour mismatch against an agreed reference, inconsistent crop and incorrect filename. Each should describe an observable result, not simply “looks wrong”.
Separate severity from category. A wrong product variant can make an image unusable; a small removable dust mark may be a local correction. Agree which problems block release, and keep examples of disputed decisions so reviewers can apply the same standard later.
Calculate first-pass acceptance with a clear denominator
Hypothetical example: a team submits 200 images for first review. Eighteen need corrections because they miss the approved brief. The first-pass acceptance rate is (200 − 18) ÷ 200 × 100 = 91%. If a customer also requests a different background on ten otherwise compliant images, record those as changes rather than adding them to the 18 production defects.
If five source photographs are rejected before editing, report them separately as intake exceptions. If only a sample was reviewed, describe the figure as the acceptance rate within that sample; do not present it as a measured rate for every delivered image.
Count affected images and repeated work separately
An image with three defects is one failed image for first-pass acceptance, but three recorded defects can still be useful for diagnosis. Keep those measures distinct. Record correction rounds and reviewer handling time as well: a supplier may eventually deliver acceptable work while creating considerable coordination cost.
A short review record can contain asset ID, brief version, reviewer, defect category, severity, correction note and final disposition. For an aesthetic disagreement, attach a source crop and the relevant reference rather than escalating an unexplained score.
Look across the collection
Detail inspection catches a damaged edge. A contact sheet can reveal inconsistent product occupancy, shadow direction or colour treatment across a family. Use both views. Where possible, keep the checker separate from the person making the edit; self-review alone is a weaker control.
Compare similar work over a representative period. A batch of glassware is not equivalent to a batch of simple cartons. Read the figures alongside service type, source quality and brief changes. Then choose a specific intervention: clarify a crop rule, improve a mask technique or ask the studio for a missing angle. A score without that next decision has limited operational value.
