Ask which accuracy is being discussed
Instrument specifications describe performance under stated conditions. Registration evidence concerns the assembly of scans. Positional checks assess the resulting dataset against an external reference. A drawing or model then introduces decisions about how that evidence is represented. These are related stages, but they answer different questions.
When a proposal quotes an accuracy, ask what it applies to, how it will be checked and which deliverable it describes. Identify whether the task concerns a local distance, a level difference or a position in the project coordinate system. State the relevant requirement rather than asking for the highest possible accuracy everywhere.
Keep resolution separate from accuracy
Resolution concerns the detail represented in the observations. Accuracy concerns their agreement with the required reference. A dataset may look dense without establishing that an edge has been measured correctly, and a sparse record may omit a small feature even when its visible points are well positioned.
Identify the smallest features that the project must recognise. Ask the provider to explain how the capture and processing choices will preserve them. Review representative locations, especially where the subject changes size or character. A general point count does not establish whether a particular bolt, recess or narrow opening can be used reliably.
Plan independent checks
A checkpoint provides an independent comparison with the dataset being assessed. It should not also be used to adjust that same dataset. The USGS lidar glossary makes this distinction between control points and checkpoints, a useful principle when planning survey verification.
Agree the checking approach before capture. Select locations that reflect the project's important geometry and avoid concentrating all checks in one easy area. Ask the surveyor to explain the reference measurements and their suitability. The report should show where checks occurred, how comparisons were made and how exceptions were handled.
Read quality reports with the project in mind
Look beyond the overall summary and examine the evidence relevant to the intended work. Are important floor connections covered? Were checks made near the proposed interface? Does the report distinguish areas with incomplete observations? Ask for clarification where statistics are presented without a method or location.
Review derived drawings and models separately. A correctly assembled cloud can still be represented by an unsuitable simplified surface. For critical features, compare the output back to the evidence and document any modelling allowance or interpretation. Acceptance criteria should make clear which representation is suitable for the downstream task.
Respond to uncertainty before committing work
If evidence is insufficient, choose a proportionate response. That may be a focused return survey, a different measurement method, additional access or a revision to the proposed detail. Do not resolve uncertainty by removing notes from the drawing or treating a smooth visualisation as proof.
Keep the survey's purpose and limitations with the accepted deliverable. If the intended use changes from general design coordination to a more demanding installation task, reassess suitability. The original result may remain useful, but a new decision should be supported by the checks that decision actually requires.
Common questions
Is a registration residual the accuracy of every point?
No. It describes a particular aspect of the registration solution and depends on the method and observations used. Review it with coverage, reference information and independent checks appropriate to the intended task.
Should a client specify a numerical tolerance?
The client and responsible designer should define the performance needed for the intended use, with survey input on an appropriate verification method. Avoid a number without stating the feature, reference, deliverable and acceptance procedure to which it applies.
Further reading
This article was rewritten in October 2026 around the subject of the original Geobiro post. The original publication date is retained; the update date identifies the current text.