Why can't you save a theory by adding an excuse every time it's wrong?

You can, technically, and that's the problem. An excuse added just to rescue a theory from a failed test is called an ad hoc hypothesis, and there is no limit to how many a theorist can add: the theory gets more and more complex but is never falsified, usually at the cost of its power to predict anything. Carl Sagan made the point with a story. He tells a visitor a fire-breathing dragon lives in his garage; when the visitor sees nothing, Sagan says she's invisible. Flour to catch footprints? She floats. An infrared camera for her fire? It's heatless. If no conceivable experiment could count against the dragon, he asks, what's the difference between it and no dragon at all?

Each of Sagan's answers is an ad hoc hypothesis: an extra assumption bolted onto a claim purely to save it from being proven wrong. Such patches are often used to explain away anomalies the original theory never anticipated. Someone determined to believe in leprechauns can keep the belief alive forever by adding 'they're invisible', then 'their motives are complex', and so on, forever one step ahead of any test.

That's why scientists are often skeptical of theories that rely on frequent, unsupported adjustments, and why ad hoc hypotheses are often characteristic of pseudoscience. As Sagan put it, being unable to disprove a hypothesis is not at all the same thing as proving it true.

The flip side is a standard the philosopher Karl Popper called falsifiability: a theory has to exclude some possible observations, making predictions that could turn out wrong. Even great scientists have reached for an ad hoc fix: Albert Einstein added a cosmological constant to general relativity to allow a static universe, later called it his 'greatest blunder', and yet something like it may correspond to today's theories of dark energy.