Here's the uncomfortable truth about data in this field: if your numbers aren't trustworthy, every decision built on them is guesswork. A BCBA looks at your graph and decides to keep an intervention, change it, or drop it. If your data is sloppy, they might fade a program that was actually working, or keep grinding on one that wasn't. Reliability is what lets everyone trust that the line on the graph reflects the kid, not the observer's mood that day.
Summary: Reliable measurement starts with an operational definition clear enough that two people count the same thing, then gets checked with inter-observer agreement (IOA). Know how to write a good definition, know which IOA formula goes with which measurement system, and know how to do the arithmetic. The exam tests all three.
Operational definitions
An operational definition is a description of the target behavior so specific that two people who've never met could read it and count the same instances. That's the whole test. If you and I watch the same session and come up with wildly different numbers, the definition failed, not the kid.
What goes into a good one
A complete definition tells the observer the behavior's name, what it physically looks like (the topography), and exactly when to start and stop counting. It also gives examples of what counts and non-examples of things that look similar but don't. The non-examples are the part people skip, and they're often what saves you in the gray areas.
That's the free preview of Measurement Reliability. The complete guide, plus all 851 practice questions and 3 full exams, is in the study pack and the book.