Get Started with Evaluation and Troubleshooting

In Part 1 you learned to see a workflow clearly: to map it end to end and to spot where it is likely to break. That is half the skill. The other half begins the moment a workflow actually produces something, a total, a draft, a confirmation, and you have to decide whether you can trust it. Part 2 builds that second half, and this short chapter sets you up for it before the method starts.

See What Part 2 Adds to Your Method

Analysis told you where a chain might strain. Evaluation is what you do when it does. Think back to the weekend food drive: mapping showed you that the AI-drafted thank-you notes were a weak point, but catching the note that thanked the wrong donor is a different move. It means looking at the actual output, deciding whether it is really right rather than just well-written, and knowing what to do when it is not.

That is the work of Part 2. Where Part 1 kept you a step back, reading the shape of a workflow, Part 2 puts you right at the point where results come out, judging them and acting on what you find. A workflow you have mapped but never checked is only half understood. This part closes the gap between seeing a problem coming and actually handling it.

Preview the Three Moves You Will Learn

Part 2 teaches three moves, in a deliberate order. You do not need to master them yet; this is just the map of where you are headed.

First, you will learn to evaluate an output against explicit criteria, checking whether it is actually correct, complete, and appropriate. Second, you will learn to troubleshoot, tracing a problem back from the symptom you noticed to the real source that caused it, rather than patching the surface. Third, you will learn to make a judgment call, deciding whether to accept, fix, escalate, or override a result, and then turning a one-off fix into a change that keeps the problem from returning.

The order matters for the same reason it did in Part 1. You cannot troubleshoot a result you have not yet judged to be wrong, and you cannot make a sound decision about a problem whose true cause you have not found. Each move sets up the next.

Know What You Need for Part 2

You need nothing beyond what you already have. If you worked through Part 1, the mapping vocabulary will feel familiar, but this part defines what it needs as it goes, so a fresh start still works. There is no prior experience with technology or AI assumed, and every example stays with everyday group efforts rather than any one job or industry.

Everything here can still be practiced with free, commonly available tools, and the checks at the end of each part remain formative: they are there for you to confirm what has landed, not to grade you. Work the chapters in order, read the worked cases closely, and attempt each practice prompt before you check its model solution. By the end of this part, you will be able to look at what a workflow produces and say, with reasons you can defend, whether to trust it, fix it, or pass it on.

With the goal of Part 2 in view, let’s begin the method by learning how to evaluate an output against clear, explicit criteria, so that “it looks fine” gives way to “here is why it holds up.”