Working from Whole to Part

No measurement is exact. A well-used steel tape might be good to ±10 mm and a modern total station to ±2 mm, but neither is perfect, and the question is not how to remove error but how to stop it building up.

Suppose twenty features must be located. The intuitive method measures P1 to P2, then P2 to P3, and so on. Every measurement is short, and short measurements are accurate — so it feels efficient.

But each position is built on the one before it. Random errors grow as √n rather than n, so they partly cancel — while any systematic error, such as a tape 2 mm short, adds every single time. And there is no check anywhere: every measurement is consistent with every other, so a large final error is invisible.

1. 1Establish a few widely spaced control points covering the whole site.
2. 2Measure the lines and angles between them with far more care than any individual detail requires.
3. 3Arrange them in a closed figure — a triangle, or a traverse that returns to its start.
4. 4Check the closure against its known geometric condition, and adjust the small error that remains.
5. 5Only then locate every detail directly from the control.

The framework measures nothing anyone asked for. It is pure overhead, and it is the whole method: each detail point now carries only its own error, so the twentieth point is exactly as accurate as the first.

A closed figure has a condition it must satisfy — the interior angles of a triangle sum to 180°, a traverse's latitudes and departures each sum to zero, a level run between two benchmarks must reproduce their known difference.

The failure to satisfy it exactly is the closing error, and it is not a nuisance. It is the survey reporting its own accuracy in a number you can compare against a specification.

The chained method had no such condition, which is why its error was invisible rather than absent. A measurement you cannot check is a measurement you cannot trust, however carefully it was taken.

Once the closing error is within tolerance it is distributed across the measurements — by Bowditch's rule for a traverse, which apportions it in proportion to leg length on the reasoning that longer legs accumulate more error. If it exceeds tolerance, the work is redone rather than adjusted.

Mistakes (blunders)

Reading 6 as 9, recording the wrong station, omitting a whole tape length. Not errors at all — they must be found and removed, and the closing check is how they are found.

Systematic errors

Same sign every time: a tape 2 mm short, sag, temperature, incorrect tension. These add linearly with the number of measurements, and they are the real danger in a chained survey. They can be corrected for once identified.

Random errors

Equally likely either way: small variations in reading and pointing. They partly cancel, growing as √n, and are dealt with by adjustment rather than correction.

This taxonomy is why the principle works. A framework has few measurements, so systematic error has few chances to add; and being closed, it reveals both blunders and the size of what remains.

Two centuries of instrument development have not changed the principle at all. Even a satellite system establishes the global frame first and positions everything within it.

• Well-conditioned triangles. Angles between about 30° and 120°, because a sliver triangle turns a small angular error into a large positional one.
• Measure the framework better. Higher-order instruments and repeated observations on control; ordinary care on detail. Effort goes where it propagates.
• Never extend a survey from detail. A new area is tied to control, never to the last point of the previous run.
• Record the closing error. It is the evidence the survey met its specification, and the first thing a checker will look for.