Check Shots and Field QA: Measuring the Accuracy You Actually Achieved

An RTK rover can report a fixed solution and a horizontal RMS of 12 mm. That is valuable evidence about the receiver's internal solution. It is not independent evidence that the point landed 12 mm from truth. A wrong antenna height, bad localization, stale control coordinate, multipath environment, or datum mismatch can produce a precise-looking answer in the wrong place.

Estimated accuracy and observed accuracy answer different questions

The receiver's estimate asks: given the satellite measurements and model, how uncertain does this solution appear? A check shot asks: how far is this measured coordinate from a coordinate we already trust?

Both belong in the project record, and they should never be collapsed into one number:

• Estimated accuracy comes from the GNSS solution — GST standard deviations, DOP, fix state, satellite count, correction age, and occupation spread.
• Observed accuracy comes from the residual between a new measurement and a known coordinate.

What a check shot records

Choose a target whose coordinate already exists in the project, typically an imported design point or a control point. Occupy it as a normal field shot. The app compares measured minus known in the active project frame:

ΔN = measured northing − known northing
ΔE = measured easting  − known easting
ΔU = measured height   − known height

horizontal error = √(ΔN² + ΔE²)

The sign matters. It tells you the direction of the miss, not only its magnitude. Several checks drifting east by the same amount suggest a frame or localization problem; random directions suggest a different cause. A vertical residual is reported only when both sides have meaningful heights.

Tolerance is a decision, not an accuracy estimate

A tolerance expresses what the job will accept. It does not change the measured residual, and it should not be chosen merely to turn the indicator green. The appropriate threshold depends on the deliverable, control quality, receiver class, occupation method, and governing standard.

For stakeout, the same idea creates a useful field loop: navigate to the design coordinate, watch distance and north/east offsets, enter the tolerance ring, then store the as-staked position and its deviation. The planned point and the position actually marked remain linked but distinct.

RMS summarizes a set of checks

For multiple independent checks, horizontal root mean square summarizes their magnitude:

horizontal RMS = √(Σ(horizontal error²) / n)

RMS is useful because larger misses receive more weight. It still does not explain the pattern. Always inspect the individual northing, easting, and vertical residuals alongside the aggregate value. One failed point can reveal a blunder that a single project statistic hides.

Checks should surround the work

A strong check plan is redundant in space and time:

• Check known control before production to catch the wrong project, frame, antenna height, or correction source.
• Use points distributed across the site instead of clustering every check beside the truck.
• Check again after a receiver restart, mountpoint change, localization change, or long interruption.
• Close on control before leaving so a failure can still be investigated in the field.
• Do not check a coordinate using the same observation that established it; that proves only that the database can subtract a number from itself.

A useful QA report includes exceptions, not just averages

The reviewer needs the whole argument behind the work. OpenSurvey3D's project report separates the measured check results from the receiver estimates and includes:

• project identity, operator, device, app version, and coordinate-system declaration;
• counts of collected and imported observations, averaged occupations, and single-epoch shots;
• solution types, estimated horizontal and vertical accuracy, occupation spread, satellites, DOP, correction age, baseline, and correction sources;
• each check residual, tolerance, pass/fail result, and project RMS;
• 3D-capture origins, heading sources, heading accuracy, and alignment fit where available;
• exceptions such as missing independent checks, weak solution states, or incomplete capture coverage.

The report can be shared as PDF for a fixed record or HTML for inspection. The raw observations remain in the project exports, so the summary does not become a dead end.

What the report cannot prove

No automated report can decide whether the known coordinate was valid, whether the operator occupied the intended mark, or whether the measurement is legally sufficient for a boundary opinion. Those remain professional judgments. The goal of field QA is narrower and more honest: preserve enough evidence that a competent reviewer can understand what happened and identify what was not verified.