Read the accuracy heat map

The Calibration Quality figure gives you one number for the whole Map, and it cannot tell you where the trouble is. The heat map is drawn over your image, region by region, and answers that question instead: which part of my venue is wrong. It is a different measurement from the figure, not the same one spread out — which is why the two are read together and why neither replaces the other.

It is the instrument for the commonest complaint a map creator gets, and almost nobody finds it on their own. If people are appearing in the wrong place, this is the page.

Switch it on

On the Map’s Calibration tab, use Heat map. It is disabled until you have three Anchor Points, because there is nothing to compute before that.

The overlay divides your image into a grid and works out, for each cell, how far a spot moves when the Map turns its pixel position into a real-world coordinate and turns that coordinate back into a pixel position. Where your Anchor Points agree with one another the round trip lands almost exactly where it started; where one of them is wrong, it can miss by hundreds of pixels.

Each cell is then shaded by which band its own difference falls in — the same three bands the Calibration Quality figure is read against:

Band	The cell is out by	Shown as
Good	3 px or less	The calm tone, labelled Good in the readout
Fair	more than 3 px, up to 10 px	The middle tone, labelled Fair
High error	more than 10 px	The strongest tone, labelled High error

Under the overlay the editor shows the numbers behind it, as text: Accuracy overlay, then the grid’s worst cell, its mean, and the band that worst cell falls in. Anything smaller than the readout can print appears as < 0.01 px, which is what a Map with nothing wrong looks like.

Because every cell is measured against the same fixed thresholds and never against the rest of your Map, two Maps can be compared, and a calm image genuinely means a calm Map.

That is a Map reading 0.0 px on four Anchor Points, and the whole grid is inside the Good band — one flat calm tone, with nothing warmer anywhere on it. Calm everywhere is the state you are aiming for, and it is reachable.

What it is actually showing — and what it is not

It shows where the Map’s own two conversions disagree with each other — pixels to coordinates and back. Your Anchor Points are what those conversions are fitted to, so anchors that contradict one another are what makes the two disagree, region by region. That is not the same as truth:

• ⚠️ Anchors that are all wrong the same way agree perfectly, and the overlay reads Good. A missing minus sign on every longitude does this: the two conversions still match each other exactly, so every cell is inside the Good band and the picture is calm. A calm overlay means your anchors are consistent, not that they are right. The Calibration Quality figure cannot see this fault either; only walking the venue, or checking your coordinate list against a second source, can.
• An intense region does not mean the venue is wrong there. It means your anchors cannot be reconciled there, which is usually one misplaced anchor rather than a locally bad image.
• ⚠️ A Map in trouble is not uniformly intense. The two conversions drift apart smoothly across the image, so somewhere between a badly wrong region and a calmer one they briefly agree — which paints a thin calm line through an otherwise intense Map. A few calm cells in a bad picture are a property of the measurement, not a part of your venue that is fine. Read the readout’s worst-cell figure, and the overall pattern, rather than any single cell.
• A band is a range, not a measurement. Two cells shaded the same are somewhere in the same band; they are not equally accurate. The readout gives you the grid’s worst cell and its mean, and those two numbers are the magnitudes — the shading is where.

Reading it, in three steps

1. Look for the one intense region. A single warm area with calm shading elsewhere is one bad anchor, and it is usually the one nearest the middle of that region. Check its two numbers against your list, and check where you clicked.
2. Look for a gradient across the whole image. Calm at one end and intense at the other, with no single culprit, is a source-image problem — the picture is not a uniform view of the place. That is choose and prepare a venue image and no number of extra anchors fixes it.
3. Look for large calm areas with no anchors in them. These are not evidence of accuracy. They are regions where nothing has been checked, and they are where the next Anchor Point goes.

Where the fourth, fifth and sixth points go

The overlay answers this directly, and the rule is simple: put the next anchor where the Map is currently guessing.

What you see	Where the next point goes
One intense region	Fix or delete the suspect anchor first, then re-read
A large area with no anchors near it	In that area, near its far edge
Intensity along one edge	On that edge, ideally at two corners of it
Calm everywhere, and five or more anchors	Nowhere. You are done

The editor nudges you towards five: while you have fewer it says how many more to add for better accuracy, and at five or more it reads as well calibrated. Past six, extra anchors rarely change anything unless they are in a region that had none.

Remember the third step above when you read that last row: calm where you have never placed an anchor is an unchecked region, not a verified one. Five anchors clustered in one corner give a calm overlay across a venue nobody has measured.

When the figure is high

Over 10 px the editor tells you outright: High error may indicate misplaced anchors. Check the heat map. Take that literally — a high figure is much more often one anchor entered wrongly than a genuinely difficult venue.

Work in this order:

1. Switch on the heat map and find the intense region.
2. Check the anchor at its centre: are latitude and longitude the right way round, is the sign right, is it the feature you thought?
3. Correct or delete it. Watch the figure, and watch the overlay go calm.
4. If deleting it drops the figure into the Good band, that anchor was the problem. If nothing changes much, the problem is spread — which points at the source image rather than at one point.

A worked example

A campsite map with five anchors reads 18.3 px — High error.

1. You switch on Heat map. The whole north half is calm; the shading intensifies sharply around the shower block in the south-east. The readout reads worst cell 41.60 px, band High error.
2. You check that anchor: latitude 55.61204, longitude 12.61890. Your list says 12.16890. A transposition.
3. You correct it. The figure drops to 2.6 px — Good — and the overlay goes calm across the whole image, with the readout following it into the Good band. Both instruments now agree that the fault is gone.
4. You add a sixth anchor in the south-west, which had none. 2.8 px, still Good, and you stop.

Ten minutes, and the alternative was rebuilding the calibration from scratch.

Related

• Diagnose “my friends show up in the wrong place” — the same problem arriving from the other end, as a complaint from somebody using your Map.
• What calibration is — why the figure means nothing at exactly three points.