A bar says its whole length. That’s the deal the reader signs when they look at one — length is value. So the moment you cut the axis off at 96 instead of 0, you’ve broken the deal: a 2% difference stretches into a bar that looks three times taller than its neighbour. The numbers didn’t change; the drawing lied about them. Tufte gave the measurement a name — the lie factor, the ratio of the effect shown to the effect in the data. Chart B’s is enormous.
The honest catch, so this is a receipt and not a rule chanted from memory: truncated axes aren’t always a crime. A line chart of already-large values (a stock at $412–$418 all week) can start above zero, because a line encodes position, not length, and a flat line honestly reads as “not much happened.” The sin is specific to bars, whose length is the entire message.
So: bar charts start at zero. If the real difference is too small to see that way, that’s information — show it with a dot plot and label the range, don’t stretch it into a lie.