Axis crimes

The auto-ranged axis

Both ends fitted to whatever is on screen, so every window is equally dramatic.

a.k.a. autoscale · auto-range · scale to fit · fit to data · the self-fitting axis · the dynamic axis · the rolling window scale · the default axis · no ylim

The other axis crimes are decisions. Somebody chose a floor, typed a maximum, cut a slice out of the middle; a reader who looks can see what was done and, in the first two cases, roughly undo it. This one is what happens when nobody chooses. Both ends of the scale are computed from the data currently in view, so the series fills the frame by construction and the amplitude on the page stops carrying information: a two-unit wobble and a fifty-unit collapse arrive at very nearly the same height, and the quieter the data the harder the axis zooms until the last digit of the instrument is a mountain range. The near relative is the free-scale panel grid, and the two are worth telling apart, because they are one rule applied along two different directions: there, several series sit side by side and each panel is given a ruler of its own, so at least the grid is visible and a reader who checks the axes can catch it. Here it runs along time instead. One series, one panel, one place on the dashboard — and a different ruler every time anybody looks at it, with nothing on screen to compare against. Three consequences follow, and none of them is visible in the picture. The exchange rate between a unit of data and a millimetre of screen is recomputed every time the window moves, so two pictures of one series are not comparable and the change you remember seeing is not one you measured. The scale is recentred as well as resized, so a level difference between two panels — often the actual finding — appears on neither of them. And both ends are set by the two most extreme points in view, so a single spike, a sensor glitch or one flush of a line decides how everything else in the window is drawn.

How to spot it

A series that fills its frame top to bottom with little or no headroom. Real measurements do not land on the edges of a plot; an axis fitted to them does, every time. Most tools pad the fit slightly or round the bounds outward, so the trace stops just short of the frame rather than touching it — the effect is the same, and the near-miss is the tell.
Read the axis numbers before you read the shape, and ask what the whole height of the panel is worth. If the answer is two units, then the mountain range is two units tall, and the drama is the frame.
Flat data drawn as noise. A quiet stretch is exactly when an auto-range has the least to work with, so it magnifies whatever is left — which is usually the rounding of the instrument.
One spike and a flat line along the bottom. Same rule, other direction: the extremes set both ends, so a single glitch takes the entire scale with it and squeezes the real data into a few per cent of the frame.
A panel with no axis numbers at all — a sparkline, a dashboard tile, a thumbnail, a phone widget, a cell in a spreadsheet. There the fitted range is invisible as well as unstated, and the shape is the only thing being read.
The same series in two places that do not look the same. Put today’s dashboard beside last week’s screenshot of it: if the silhouette changed and the underlying numbers did not, the axis moved.
A comparison drawn across panels. Each is centred on its own middle, so a step between two windows is not small on the chart, it is absent from it.
In code and configuration it is usually the absence of a setting rather than the presence of one: no ylim in matplotlib, Grafana’s axis min and max left on Auto, an Excel chart with Auto still ticked, a charting library handed a series and no bounds. Grafana documents the failure inside the description of the cure — soft min and soft max “can prevent small variations in the data from being magnified when it’s mostly flat”.

The fix

Choose the range once, from the question rather than from the data, and then say what you chose. The test is whether you could name the bounds before seeing the series: a blood pressure is read in a band a clinician can state, a latency budget has a target, a percentage has a hundred. Write that range into the caption the way you print a unit, and hold it — across every panel in a grid, across every width of window, across every week the dashboard is looked at — for the same reason a small multiple holds one axis: the whole value of the comparison is that the ruler did not move. Where the range genuinely has to follow the data, and often it does, make it follow slowly and visibly. Use soft limits, which widen for real excursions but will not zoom into a flat stretch. Compute the bounds from a robust range rather than the minimum and maximum, so one glitch cannot set the scale, and tick the excluded points at the frame with a count rather than dropping them. Put something in the plot that does not move — a target line, a normal band, a reference series — so the reader has a fixed thing to measure the wiggle against. Then print the span as a number beside the panel: “range 2 mmHg”, “range 4 ms”, “this window spans 0.3%”. It is one line, it survives being screenshotted, and it is the one line no panel will ever write for you. Above all, resist the reflex to read a shape off a picture whose scale you have not checked. On an auto-ranged axis the shape is the one thing guaranteed to be there whatever the data does, so it is the one thing that cannot be evidence.

In the gallery

Dark navy briefing slide. The masthead reads “Anaesthesia quality desk · intraoperative stability review” on the left and “Mean arterial pressure · VitalDB case 4732 · 7,856 readings” on the right. The headline is “Unstable every time we looked.” Beneath it sits a bordered gold chip reading “Demonstration · drawn by us from an open dataset”. A standfirst says this is one dashboard tile showing the last ten minutes of arterial pressure, screenshotted four times across the 4 h 23 min the recorder ran; that every reading is VitalDB’s own — mean pressure off a Solar 8000 monitor, one every two seconds, 300 to a screenshot, nothing smoothed or dropped; and that the tile’s vertical axis is auto-ranged to what is in it, the panel default, so four looks give four times the same shape, a mountain range that never settles. The main panel, headed “Mean arterial pressure · one tile, screenshotted four times” and subtitled “The tile shows the last ten minutes · 300 readings in each screenshot · vertical axis auto-ranged to what is in the tile · nothing smoothed, binned or dropped”, is a two-by-two grid of gold line charts against a dark ground, each with a horizontal axis running 0 to 10 labelled “Minutes into the tile”. Each carries the time it was taken in bold above it and, in smaller grey type, the window it shows and the range its axis was fitted to. Top left, “Screenshot at 55 min”, showing 45–55 min, axis auto-ranged to 69–126 mmHg, ticked at 80, 100 and 120: the line runs flat along the bottom for three minutes, climbs in steps to a peak at the very top of the frame around the six-minute mark, and falls back to the floor by minute eight. Top right, “Screenshot at 95 min”, showing 85–95 min, axis auto-ranged to 81–87 mmHg, ticked at 82, 84 and 86: a jagged square-edged trace that starts at the bottom of the frame, spikes repeatedly, climbs to a plateau and touches the top twice. Bottom left, “Screenshot at 145 min”, showing 135–145 min, axis auto-ranged to 74–76 mmHg, ticked at 74, 75 and 76: a dense picket of vertical strokes running the full height of the panel throughout. Bottom right, “Screenshot at 215 min”, showing 205–215 min, axis auto-ranged to 72–76 mmHg, ticked at 72, 74 and 76: another restless trace that fills its frame, touching the top just after minute one and the floor around minute three. All four traces run edge to edge, top to bottom. A caption along the foot reads “One patient · one arterial line · one monitor · one tile · every reading the recorder wrote down, in the order it wrote them”. Six cards run down the right. “The record, in one figure” prints a large gold “4 of 4” beside “looks that never settle”, and notes that the four screenshots were taken 40 to 70 minutes apart across 4 h 23 min of anaesthesia and that every one comes back restless, with peaks, troughs and no plateau anywhere. “Read-out for the desk” says the trace does not settle, that whatever moment you catch the tile at the pressure is climbing or falling, and that the pattern holds from the start of surgery to the end of it, so this is not one bad stretch but how the whole case ran; its recommendation is to narrow the alarm band on mean pressure, add a stability index to the theatre dashboard, and put the case on the list for haemodynamic review. “What one screenshot is” explains that it is the last ten minutes of Solar8000/ART_MBP — the mean pressure the monitor computes from the arterial waveform — at the recorder’s native two-second interval, 300 readings in each, every one of them plotted. “The four looks” tabulates in monospace the axis each screenshot was fitted to: at 55 min, 69–126 mmHg; at 95 min, 81–87 mmHg; at 145 min, 74–76 mmHg; at 215 min, 72–76 mmHg; readings in each 300; sampling interval 2 s; readings dropped 0. “Why nothing here is bent” says it is one tile, one series and one unit throughout, that no axis has been truncated by hand, nothing is indexed, rebased, logged or smoothed, no reading has been dropped, every screenshot is the same size, and each one prints the range its axis was fitted to — so whatever makes these four look alike, it is not the drawing. “Basis of measurement” credits the VitalDB open dataset (Lee & Jung, PhysioNet, 2022; Lee et al., Sci Data 9:279), case 4732, track Solar8000/ART_MBP, retrieved 17 September 2026, and says VitalDB publishes the readings while the tile, the screenshots and the chart are ours. A method line runs along the foot: published readings, as recorded; one patient, one line, one tile; 300 readings per screenshot; nothing smoothed, binned or dropped; axis fitted to what is in the tile. A footnote credits the source and its Creative Commons Attribution 4.0 licence, notes that every plotted value is a reading the recorder wrote down, states that the slide is a demonstration drawn by us in the manner of a theatre quality-review dashboard — the tile, the desk, the read-out and the recommendation are invented, and every figure behind them is real — and adds that each screenshot is fitted exactly to its own minimum and maximum, where a real panel pads a little or rounds the bounds, which softens the effect without changing it.

The 2 mmHg look and the 57 mmHg look, drawn the same height

The auto-ranged axis