How can an honest chart still mislead you?

Every chart in this course has been accurate. None of them lied. And yet charts mislead people constantly — through framing choices, missing context, and the human talent for finding patterns in noise. Knowing the failure modes is as much a part of chart literacy as reading a candle.

Framing traps

The scale trap. Covered in Unit 1 and worth repeating because it's the most common of all: a decades-long history on a linear axis makes recent moves look enormous and old crashes look trivial. Nearly every viral "this looks exactly like 1929" chart depends on this.

The axis crop. Start the price axis at $95 instead of $0 and a 2% wobble fills the screen as a mountain range. Nothing is falsified; the reader is simply handed a magnifying glass without being told.

The chosen start date. "Down 30% since January" and "up 400% since 2019" can describe the same instrument on the same day. Whenever you see a chart with a striking start point, ask what the chart looks like starting one year earlier or later. If the story changes, the story was about the crop.

Both charts below are the same instrument, drawn from the same rows. Only the window differs.

Data traps

Unadjusted prices. A stock that split 4-for-1 shows a terrifying 75% crash on unadjusted data — a crash that never happened to anyone. Dividends do a smaller version of the same thing. Know which series you're on before you draw anything.

Thin instruments. On a stock that trades rarely, gaps and spikes are as likely to be artefacts of illiquidity as they are to be genuine repricing — the prints are real, but they represent almost nobody. A "level" formed by three trades is not a level; it's three trades.

Survivorship. Chart libraries and screeners are full of companies that made it. The ones that were delisted are simply absent, so any pattern you "discover" by browsing charts is drawn from a sample that already excluded the failures.

The human trap

This is the big one. Humans find patterns in random data — reliably, and with confidence. Generate a chart from coin flips and you will see trends, support levels, head-and-shoulders formations and beautiful trendlines. This is a well-established result, and it should permanently calibrate how impressed you are by any pattern you spot.

Two related biases finish the job. Hindsight bias makes the past look obvious, so historical charts feel far more readable than the live right edge ever is. And confirmation bias means that once you have a view, the chart cooperates: you'll notice the touches that support your level and forget the ones that didn't.

The defences that actually work

1. Fix your settings before you look — timeframe, scale, swing rule, zone width. Decisions made in advance can't be bent to fit a conclusion.
2. Write the read down, with a falsifying condition (last lesson). Unwritten analysis is unfalsifiable analysis.
3. Count the misses. For any pattern you like, deliberately go find the times it didn't work in the same chart. If you can't find any, you haven't looked.
4. Ask "compared to what?" A move is big or small only relative to this instrument's own history and its peers.

None of this makes charts useless. It makes them what they honestly are: a well-designed record of what price has done, read by a brain that is very good at seeing more than is there.

In the data

The adjusted-or-traded choice follows a price into everything computed from it. Here is one session of Coca-Cola, a steady dividend payer, both ways:

The two closes differ by about 7.5% as of 29 September 2026, all of it dividends paid since. An average, a band or an oscillator computed on one series gives a different number from the same recipe on the other: same stock, same dates, two answers. Charts do not always say which series they used. This course's charts name it in the legend under each chart.

Try it now

1. Write one sentence describing each of the two charts above as a headline would. Same rows, two headlines — that is the whole of the crop trap, and nobody had to lie to you to produce it.
2. Now the data trap. Below is Apple on the session before its most recent split and the session after it, the traded close beside the adjusted close, and under it the split itself in Apple's list of splits. One of those two columns shows a collapse that happened to nobody.

3. Now a third crop of the identical rows, below. Pick any pattern you believe you can see in it, then hunt the same instrument for three occasions that pattern appeared and led nowhere. Write down the count.