What calibration is, and why three points is the minimum

Your image is a picture. A phone reports a latitude and a longitude. Calibration is the bridge: it works out the single relationship that turns any pair of coordinates into a spot on your picture, so that a person standing at the food court is drawn at the food court.

You build that bridge by giving examples. Each example is an Anchor Point: this spot on my image is that place on Earth.

What an Anchor Point is

Two halves, and both are needed:

A spot on your image, in pixels — where you clicked.
A place in the world, as a latitude and longitude.

From several of those, the editor fits one transform that stretches, rotates, scales and skews your whole image onto the world at once. One relationship for the whole picture — not a per-corner correction, and not a warp that varies across the image. That is the thing worth understanding, because everything else on this page follows from it.

Why three is the minimum

The transform being fitted has enough freedom that it takes exactly three examples to pin down. With one point the picture could be any size and any rotation; with two it could still be flipped or sheared; with three there is one answer.

So three is where calibration starts working. It is also where it starts lying to you.

⚠️ Three points always score perfectly, and the score is meaningless

With exactly three Anchor Points the fitted transform passes exactly through all three, because there is a transform that does. So the editor reports an error of 0.0 px — and that number is telling you nothing at all about whether your Map is right. It is telling you that three points can always be fitted by three points.

The editor says so where the number is: 3 points = perfect fit. Add 2+ more to measure actual accuracy.

Put a fourth point in and the arithmetic changes character. Now there may be no transform that hits all four, so the fit is the one that comes closest overall — and the leftover distance is real information about your Anchor Points. That is the number worth reading.

What the number means once you have four or more

The editor shows a Calibration Quality figure in pixels: the average distance between each Anchor Point and where the fitted transform predicts that point should be.

The figure	What it reads as
3 px or less	Good
3 to 10 px	Fair
More than 10 px	High error — and the editor says a misplaced anchor is the likely cause

It is a measure of how well your anchors agree with each other, not of how right they are. Five anchors that are all wrong in the same way agree perfectly. That is what the heat map is for — read the accuracy heat map.

How many to place

Five is the number the editor is built around: it shows a five-dot meter, and while you have fewer it tells you how many more to add for better accuracy. Past five it reads as well calibrated.

Anchor Points	What you get
Fewer than 3	No calibration. The Map cannot place anybody
3	A working Map, and no way to tell whether it is right
4	The first honest error figure
5–6	A figure you can act on, and a heat map worth reading
More	Diminishing returns, unless a particular corner is bad

Where to place them

The editor’s own Anchor Placement Tips are the short version, and they are right:

At least three.
Spread far apart — covering the image’s corners is ideal.
Not all in a straight line. Three points along one road cannot define the transform at all — why three points in a line will not work.
More anchors improve accuracy, because the fit is a best-fit across all of them.
Use the heat map afterwards to check accuracy across the whole image.

The reason spread matters so much: the transform is fitted from your points and then applied everywhere. Anchors clustered in the middle of a site leave the edges to extrapolation, and a small angular error at the centre becomes a large positional error at the perimeter.

A worked example

A festival site, roughly 600 by 400 metres, on a 2100-pixel-wide site plan.

You place three anchors: the main gate, the north-west corner of the car park, the bridge. The editor reports 0.0 px and says three points are a perfect fit.
You do not believe it, and place a fourth at the far south-east gate. The figure becomes 2.4 px — Good. Now you know the first three were consistent.
You place a fifth near the centre. 2.9 px, still Good, and the meter fills. The hint changes to say the Map is well calibrated.
You switch on the heat map and see the whole plan shaded evenly, with no warm corner. Done, in about ten minutes.

Had step 2 given you 31.0 px instead, the editor would have told you a misplaced anchor is likely and to check the heat map — and one of the four would have been in the wrong place, not all of them.