ISO 5725-1 defines accuracy as the combination of trueness and precision. Trueness is the closeness of agreement between the average of a large series of results and an accepted reference value. Precision is the closeness of agreement between independent results obtained under stipulated conditions (ISO, 1994). A scanner can be precise and untrue, returning the same wrong arch width every time, and the two properties have different clinical meanings. ISO 12836 sets test methods for laboratory digitizing devices (ISO, 2015), and ISO 20896-1 extends accuracy testing to intraoral devices (ISO, 2019). Neither standard is written for implant scan bodies on an edentulous arch, which is the gap this protocol addresses.

Figure 1 groups the sources of deviation in a bench study under four headings. Device factors are the object of study. The other three are conditions that must be controlled or reported, and each has been shown to matter.

Object factors include the distance between implants, their angulation and depth, and the geometry and material of the scan bodies (Giménez et al., 2014). Precision falls as the distance between scan bodies increases, because errors accumulate as successive images are stitched across featureless tissue (Flügge et al., 2016). Operator and room factors include the scan path (Müller et al., 2016), experience, and ambient lighting, which altered the accuracy of several scanners in a controlled comparison (Revilla-León et al., 2020).

Analysis factors are the least visible and may be the most influential. Many studies superimpose the test scan on the reference by a global best-fit algorithm of the iterative closest point type (Besl and McKay, 1992) and report the mean surface distance. Best-fit alignment distributes error across the arch and can understate a systematic distortion at its ends, a problem analysed for sequential dental scans by O'Toole et al. (2019). Surface distance is also a poor proxy for what the prosthesis experiences, which is the relative position and axis of each implant. Comparative studies that report inter-implant distances and angles are easier to interpret (Vandeweghe et al., 2017).

The model represents an edentulous arch of realistic form with implant analogues placed in a distribution typical of a fixed complete-arch prosthesis, including at least one tilted posterior implant. It is made from a dimensionally stable material, and the analogues are fixed so that they cannot move between sessions. The surface between implants should offer as little texture as an edentulous ridge does, since a richly textured model flatters the stitching algorithm. Scan bodies of a single make and batch are tightened to the manufacturer's torque and left in place for the whole study, because reseating introduces its own error.

The reference positions are measured with an instrument at least several times more accurate than the devices under test, such as a tactile coordinate measuring machine or an industrial optical scanner with traceable calibration. The uncertainty of the reference is stated in accordance with the Guide to the Expression of Uncertainty in Measurement (JCGM, 2008). A trueness result smaller than the reference uncertainty has no meaning, and this simple point is often overlooked. The reference measurement is repeated at the end of the study to confirm that the model has not changed.

Room temperature and lighting are fixed and recorded. Each scanner is calibrated according to its manufacturer's instructions before each session, and hardware and software versions are noted, since updates can change performance. The model is mounted in a phantom head or behind a mask that restricts access to the dimensions of a mouth opening. A model scanned freely on a bench allows scanner movements that are impossible in a patient.

A scan path is defined in advance for each device, following the manufacturer's recommendation where one exists, and is described in enough detail to be repeated. Operators are trained to a stated criterion before data collection. The number of repetitions per scanner and operator is fixed beforehand, and the order of devices is randomised to spread any drift in conditions. Scans that the software rejects or that the operator restarts are counted and reported, since the frequency of failed scans is itself useful information. Files are exported in an open format without smoothing or hole filling.

The position and axis of each scan body are obtained by fitting its library geometry to the scanned surface, and the residual of each fit is recorded. Test and reference data sets are then brought into a common coordinate system by aligning on one terminal scan body only, so that deviation accumulates visibly along the arch as it would beneath a rigid framework. Global best-fit superimposition of the whole surface may be reported as a secondary analysis for comparison with earlier work, on the method described for complete-arch scans by Ender and Mehl (2013). It should not be the primary one.

Three quantities are computed for every implant relative to the reference implant: the deviation in inter-implant distance, the angular deviation of the implant axis, and the three-dimensional displacement of the platform centre. Trueness is the difference between the mean of repeated scans and the reference value. Precision is the dispersion of repeated scans about their own mean, reported both within and between operators. Signed deviations are retained alongside absolute values, since a consistent contraction or expansion of the arch is hidden when only magnitudes are averaged. Deviation data are often skewed, so medians and ranges are given with means, and the analysis plan is fixed before scanning begins.