Site Inspection Data: Safeguarding Laser-Scanning Coordinates

When counsel receives a 3D scan deliverable, it usually arrives as a single registered point cloud or a walkthrough model. That polished object is a derivative work. It is the output of a transform applied to the underlying raw scans. Treating it as the evidence is the first custody error, because every judgment call in registration is already baked in and no longer visible.

A terrestrial laser scanner does not measure coordinates directly. From each setup it measures range plus horizontal and vertical angles, producing points in a scanner-local frame with its own origin. Every station is a separate coordinate system. The unified cloud you were handed only exists because software solved how those frames fit together. If you cannot inspect the raw per-station scans and the parameters used to combine them, you cannot test whether the geometry a defect claim rests on is real or an artifact of alignment.

• Demand the native raw scans in the manufacturer format (for example RIEGL RXP, FARO FLS, Z+F ZFS, or Leica), not only the exported cloud.
• Treat registration as a working copy operation. The raw scans must never be overwritten by the registered result.
• Get the project or workspace file from the registration software, which records the transforms and settings, so the combination step is reviewable rather than asserted.

There are two dominant ways to merge scans, and they carry very different auditability. Target-based registration places physical targets, planar checkerboards or spheres, in the overlap zones between setups. The software locates each target centroid in every scan that sees it, then solves a least-squares rigid-body transform, rotation plus translation, that brings the same target to the same location across scans. Cloud-to-cloud registration, typically Iterative Closest Point, aligns overlapping geometry directly by minimizing point-to-point or point-to-plane distances, with no independently identifiable tie features.

The distinction matters for evidence. Target-based registration produces named, reproducible tie points and a per-target residual, the distance between where a given target lands from one scan versus another after the transform. Cloud-to-cloud reports an overall deviation but leaves no discrete features an opposing expert can re-verify, so it is harder to reproduce and easier to dispute.

The calibration evidence lives in the residuals. A registration report should state each target residual and the overall RMS registration error. Those numbers are what let a second expert re-run the alignment from the same raw scans and targets and land within tolerance. A model that has been nudged, cropped, or re-fit to support a conclusion will diverge from that reproducible result.

• Ask which method was used per scan and whether any setups fell back to cloud-to-cloud because targets were missing.
• Require the residual table, not a summary sentence that registration was successful.
• Flag suppressed or disabled targets. Turning off a high-residual target lowers the reported RMS without improving the actual geometry.

The targets are the physical calibration standard of the scan. If they are not documented, the claim that the scene model is calibrated is unverifiable. For active sites, where targets get bumped, moved, or removed by trades between setups, this documentation is also the record that the geometry was stable during capture.

A complete target record ties each physical target to its identity and its measured position. That means a target map, unique target IDs, the target type and size, and, for georeferenced work, the surveyed coordinate of each control target. When the same target appears in multiple scans, its ID is the thread that makes the registration reproducible and lets any reviewer confirm the transform was solved from real, located features.

• Target log per setup: which target IDs were visible from which scanner station, with type and dimensions.
• Target map or site diagram showing placement, so an opposing expert can confirm overlap geometry was adequate.
• Stability notes for active sites: record whether targets were left in place across setups or re-established, and by whom. Moved targets between overlapping scans corrupt the transform.
• Photographs of target placement keyed to IDs, which double as a scene condition record for the defect at issue.

Registration alone produces an internally consistent cloud floating in an arbitrary local frame. Georeferencing ties that cloud to a known datum and projection by surveying the control targets with a total station or GNSS receiver into real-world coordinates, then constraining the registration to those positions. This is what makes measurements in the cloud comparable to the as-designed CAD or BIM, to prior surveys, and to later scans of the same defect over time.

Georeferencing introduces a second, independent accuracy chain: the survey. Its quality is governed by the control network, the instrument, and the surveyor. Absent that chain, a dimension pulled from the cloud is a relative measurement inside the scan, not a coordinate you can defend against the design intent or against a differential scan taken months later.

• Obtain the control survey: the datum and projection used, the control point coordinates, and the method (GNSS RTK, static, or total station traverse).
• Confirm surveyor licensure where the work asserts real-world coordinates, since that is often regulated survey work.
• Check network closure and control residuals, the survey equivalent of registration residuals, to see how tightly the cloud is pinned to the ground.

Once scans leave the instrument, the custody problem becomes an ordinary digital-evidence problem, and the standard control applies: compute a cryptographic hash, MD5 or SHA-256, of each raw file at the moment of offload from the scanner or its card. That hash is the fingerprint. Recomputing it later and matching the recorded value proves the bytes have not changed since capture. A mismatch, or a missing baseline hash, means the file cannot be shown to be what came off the instrument.

Format choices affect verifiability. Raw manufacturer files carry scanner serial number, firmware, capture timestamps, and inclinometer or environmental data in their headers, all of which corroborate when and how the scan was taken. For interchange, the open E57 format (ASTM E2807) and the ASPRS LAS or LAZ formats preserve structured metadata that a reviewer can read without the vendor software. Handling of the media and files should follow recognized digital-evidence practice, such as the guidance published by SWGDE.

• Hash at capture and record the algorithm and values in the field log, before any copy or registration.
• Preserve headers. Avoid export paths that strip scanner metadata or re-timestamp files.
• Log every transfer of the media and files: who, when, from where to where, with the hash re-verified on receipt.
• Hash each derivative too, the registered cloud and any model, so the chain from raw scan to deliverable is unbroken and each step is reproducible.

Active sites defeat custody through ordinary churn, not sabotage. Trades move targets, lighting and dust change between setups, and the scanned condition itself can be altered or repaired before anyone returns. A tamper-evident trail is the assembly of independent records that would all have to be forged consistently to fake the scene, which is far harder than editing a single cloud.

The spine is reproducibility. Given the preserved raw scans, the documented targets, and the control survey, a second expert should re-derive the same registration transform within stated tolerance and reach the same measurements. Divergence between an independent re-registration and the delivered model is itself evidence of manipulation or undisclosed editing.

• Field capture log with date, time, personnel, scanner serial, station count, weather, and site conditions.
• Sequence of custody: capture, hash, transfer, registration, export, each timestamped and attributable to a person.
• Contemporaneous documentation of the condition scanned, since a defect can be remediated after capture and the scan is often the only record of its original state.
• Preserve intermediate registration states along with the final model, so the alignment history is auditable.

None of this guarantees admission, and no vendor should promise it. What a documented custody and calibration record does is let the offering party meet the reliability inquiry courts apply to expert methods under Federal Rule of Evidence 702 and the Daubert framework, or the general-acceptance test under Frye in jurisdictions that still use it. Laser scanning is a known technique with published standards, a knowable error rate expressed through registration and control residuals, and a reproducible workflow. The residuals, the hashes, and the target and survey records are what make those factors demonstrable rather than asserted.

Accuracy claims should be stated in the vocabulary the industry uses. The USIBD Level of Accuracy (LOA) Specification distinguishes measured accuracy from represented accuracy, which prevents the common overstatement that a visually detailed model is dimensionally precise. Report the accuracy the data supports, keyed to the residuals, rather than to how convincing the render looks.

• State error as residuals and survey closure, not as a single marketing accuracy figure.
• Separate measured from represented accuracy in any report a witness will defend.
• Preserve the reproducibility chain, since the ability of an opposing expert to re-derive the result is the strongest reliability showing available.