Step one: define the claim in assessable terms
A milestone statement should not depend on an unspoken interpretation of “done.” Specify the work package, location or boundary, relevant scope revision, assessment date and condition being reported. If the project uses staged completion—such as installed, inspected, tested, commissioned or accepted—name the stage precisely. Do not combine several stages into one label unless the approved reporting rules explicitly do so.
Ask what evidence is required by the project’s own definition. Some conditions can be observed visually; others depend on measurements, inspection records, certificates, commissioning results or an authorized decision. An image may support the first category while saying little about the rest. The review should not promote a visual observation into a broader claim without the required supporting information.
Step two: inspect the imagery as an observation
Check image date, source, location and viewpoint before interpreting content. Compare the view with prior captures where the framing is genuinely comparable. Identify the visible boundary, obstruction, missing angle, temporary work and relevant conditions. Ask whether the image is sufficiently clear to support the particular observation. A wide view may demonstrate that an area is accessible but lack detail to establish an installation feature; a close view may show a feature but fail to locate it.
Describe observations in neutral language. “Three visible members appear installed within the marked bay” is more precise than “the structure is complete” when the evidence only covers a portion. Distinguish what a reviewer sees from any machine-generated label. If image analysis detects a likely object or change, treat the output as a prompt for review. Document the source image and the person who confirms whether the output is useful.
Step three: connect non-visual project records
Identify the controlled documents and records that the milestone definition calls for. Depending on the work, these may include inspection and test plan entries, quality records, material traceability, delivery or installation records, approved changes, test results and sign-off. Check their status and scope. A record that refers to a neighboring zone, superseded drawing or different work package cannot quietly be treated as proof for the claim under review.
Keep each evidence item’s role clear. Programme data can show planned and reported dates, but a planned activity is not evidence that work happened. A status field may reflect a team update rather than an inspection. An image can document visible state but not necessarily workmanship or concealed elements. The chain is stronger when sources complement each other and limitations are explicit.
Step four: record the human decision
Give the reviewer a concise evidence pack: the claim, included boundary, relevant images, record links, capture dates, scope version, known exceptions and the decision requested. The reviewer should be able to accept, reject, qualify or defer the claim and explain why. Record reviewer identity or role, decision date and any conditions. If authority is delegated, refer to the project’s agreed responsibility process rather than assuming a model or tool can approve a milestone.
A qualified reviewer may determine that a site inspection is necessary, request a new viewpoint, ask for a test result or accept a carefully qualified statement. Preserve that judgment separately from automated suggestions. A review trail should also allow later correction if a source record is withdrawn, an error is found or scope changes. Avoid overwriting the earlier assessment as though it never existed.
Step five: write a bounded report statement
Report only what the evidence supports. A useful statement can identify the observed portion, the evidence date, the records considered and any exclusion. If a condition remains open, name it and who will resolve it. Avoid unsupported percentages, implied certification or language that suggests an image-reading model has independently verified the work. Where no sufficiently reliable evidence exists, say that assessment is pending.
These controls make reporting more useful, not more cumbersome. Project managers gain a clear basis for decisions; site engineers can see where a field observation needs follow-up; clients can distinguish visible progress from acceptance. The common discipline is to trace each completion claim to a definition, comparable imagery, relevant project records and an accountable human review. A defensible result is not the most confident-sounding result: it is the one whose reasoning can be retraced.
All diagrams and examples here are illustrative only and are not actual site evidence.