What is Chebyshev's inequality actually telling me, in a form I can picture?

Your instinct about the trade-off is exactly right, and the picture makes it precise.

The integral is a fixed budget of area. If you want the function to be very large — lambda big — then the set on which it achieves that must be correspondingly small, or the rectangle would not fit under the curve.

So: a function cannot be big on a large set without having a large integral. Which is intuitively obvious and is exactly what the inequality formalises.

That also tells you when the bound is tight and when it is useless. It is tight when the function actually looks like that rectangle — constant at lambda on a set and zero elsewhere. It is very loose when the function is enormous on a tiny set, because then the integral is dominated by a spike the bound knows nothing about.

There is a picture and it makes the whole thing obvious in about ten seconds.

Draw the graph of your function. Draw a horizontal line at height lambda. Look at the set of points on the domain where the graph is above that line — call its measure m.

Now draw the rectangle sitting on that set, of height lambda. Its area is lambda times m.

That rectangle is entirely underneath the graph, because on that set the function is at least lambda by definition. And the integral of the function is the total area under the graph.

So the rectangle's area cannot exceed the integral:

lambda × m  ≤  ∫ f

Rearrange and you have the inequality. That is the whole content: a box that fits under the curve is no bigger than the area under the curve.

One formal note that makes the proof feel like the picture rather than a separate thing: the standard one-line argument is to observe that lambda times the indicator of the set is pointwise less than or equal to f, and then integrate both sides.

That inequality between functions is the rectangle sitting under the curve, written symbolically. Once you see that correspondence, the proof stops being a trick and becomes a transcription of the drawing.

Worth connecting it to the version most people meet first, because they look unrelated and are the same statement.

The probability version — that a random variable is far from its mean with probability bounded by variance over distance squared — comes from applying exactly this inequality to the squared deviation. The measure is probability, the function is the squared distance from the mean, and its integral is the variance.

So the abstract measure-theoretic statement is the general one, and the familiar probability result is one substitution away.

The same move generates a whole family: apply it to different functions of the variable and you get different tail bounds, including the exponential ones used everywhere in concentration inequalities. All of them are the rectangle argument with a different curve.