Why the top of the turn bends

A divider only divides when nothing is attached

The divider's arithmetic, SIGNAL = VCC × (track below ÷ whole track), rests on one assumption: the same current flows through the track above the wiper and the track below it. Nothing leaves at the wiper.

On this board, once the LED conducts, something does. The upper track now carries two currents: the one that carries on down the lower track, and the LED's. More current through the same resistance means more voltage dropped across it, so the wiper sits lower than the arithmetic says.

How much, and where

The dashed line is the plain divider. The solid line is the wiper with the LED attached, and the red band between them is the error. Until the wiper reaches about 2.0 V, the two are the same line: the LED is dark and takes nothing.

On 5 V that point comes about two-fifths of the way round, and from there the solid line flattens. The error grows to about 1.3 V, roughly 260 counts on an Uno, near 85 % of the turn. Then it closes up fast: at the very end the wiper sits on VCC with no track above it, nothing drops, and the reading jumps to 1023. So on an Uno most of the upper half of the turn crawls, and the last few degrees sprint.

On 3.3 V the LED lights later, about three-fifths of the way round, and draws less, so the error is smaller: about 0.46 V at worst, near 90 % of the turn. That is still about 14 % of the range, roughly 580 counts on a 12-bit board. The same crawl and sprint, over a shorter stretch.

Why the wiper gives way

Seen from the wiper, the potentiometer behaves like a voltage source, the plain divider's value, with a resistor in series: the two halves of the track in parallel. In the middle of the turn that is 2.5 kΩ; near the ends it is less. Anything that draws current from the wiper loses voltage across that resistance. Where the error peaks on 5 V, the LED draws about a milliamp through a little over a kilohm of it, and that is more than a volt.

An analog input draws almost nothing, which is why it reads the divider faithfully. The LED is the only load here, and it is on the board.

What it costs, and what to do

A knob that reads crooked is not broken, but a sketch that assumes a straight line is wrong over the upper part of the turn, on every board. If you map an Uno's reading to a percentage with map(), the knob at three-quarters reads about 52 %. The first 30 % of the turn moves an Uno's reading about 300 counts; the 30 % from 60 % to 90 % moves it about 210.

Three honest ways round it:

• Run VCC from 3V3. On an Uno, the range becomes about 0 to 675 and the bend shrinks to about 0.46 V, roughly 95 counts. The LED is a little dimmer. On the three 3.3 V boards this is already the wiring, and the bend is already that size.
• Calibrate. Record the reading at a few marks and draw straight lines between them. Calibrating the turn does it with five.
• Use the lower part. If the fine control only matters at the low end, the reading is straight until the LED lights: the first two-fifths of the turn on 5 V, the first three-fifths on 3.3 V.

All of this is a model, not a measurement: the LED as a fixed 2.0 V drop with its 1 kΩ, on a 10 kΩ linear track. Real LEDs turn on gradually, so the corner at 2.0 V is softer in life. Seeing the bend draws your own board's curve.