7.2 Adjust Approach Based on Feedback and Results
7.2.1 Troubleshooting Is a Loop, Not a Single Move
Diagnosing a symptom (Lesson 7.1) tells you what's wrong. Adjusting your approach is what you do next, and it is iterative and evidence-driven rather than a single corrective action. The core idea: treat the actual output as feedback, not as a verdict on whether the task is hopeless. Every weak result is data about what to try next.
Adjusting approach is a three-step loop that repeats: read the gap, change one variable, re-run and compare — feeding back into the next iteration.
7.2.1 — Key Concept
Troubleshooting is iterative: read the actual result against the intended outcome and name the specific gap, change one variable, then re-run and compare before deciding whether to keep or revert the change.
7.2.2 The Three Steps in Practice
- 1.Read the result — compare the actual output against the intended outcome and name the specific gap (missing detail, wrong format, stale fact, wrong tone).
- 2.Change one variable — the instruction, an example, the context provided, the task breakdown, or the model. Pick exactly one lever per iteration.
- 3.Re-run and compare — confirm the change actually helped before keeping it; revert if it didn't.
Notice that step 2 offers five different levers, and the discipline is picking only one of them per attempt. Adding a missing example is a different variable than swapping the model, which is different again from re-supplying context or splitting the task into smaller steps. Whichever lever you pick, the output itself is usually already telling you which one is missing — a vague answer points at the instruction or an example, a stale answer points at context, a partially-completed answer points at task breakdown.
7.2.3 Why One Variable at a Time
Changing a single variable per iteration is what makes the fix diagnosable. If five things change at once between one attempt and the next — a new instruction, a new example, a different model, more context, and a different task breakdown — there is no way to attribute the resulting change in quality, better or worse, to any one of them. The next attempt becomes a fresh guess rather than a confirmed improvement, and you're right back to the blind-retry pattern this whole domain is trying to move you away from.
7.2.3 — Exam Trap
Changing many variables at once to save time feels efficient but destroys the ability to isolate what actually caused an improvement or a regression. It also means ignoring the feedback the output itself provides about its own specific gap.
7.2.4 Turning Lessons Into Standards
Over several tasks, patterns emerge: a certain kind of ask always needs a worked example, a certain report always benefits from a fresh chat rather than a long thread, a certain output format only lands with an explicit schema spelled out up front. The payoff of disciplined, one-variable-at-a-time iteration is folding those recurring lessons into standard prompts and Project instructions, so the next similar task starts from the improved baseline instead of repeating the same diagnosis from scratch every time.
Where this shows up on the exam
7.2 questions describe a sequence of attempts to fix an output. The correct pattern is always: name the gap, change one thing, re-run and compare. Any option that changes several things at once, or that never checks whether the change actually helped, is the distractor.
Key Takeaways
- ✓The feedback loop is: read the gap, change one variable, re-run and compare.
- ✓Changing one variable at a time is what makes a fix diagnosable — change several and you can't tell what worked.
- ✓The output itself is feedback about what's missing; use it rather than guessing at an unrelated change.
- ✓Recurring lessons should be folded into standard prompts and Project instructions, not re-discovered each time.
- ✓Adjusting approach is iterative and evidence-driven, not a single corrective action.
Check Your Understanding
Test what you learned in this lesson.
Q1.What best describes sound troubleshooting when adjusting a prompt after a weak result?
Q2.After reading a weak output, an analyst decides to simultaneously rewrite the instructions, add a new example, switch models, and supply additional context, then re-runs the prompt. The new output is better. What is the problem with this approach?
Q3.A team notices that every time they ask Claude to produce a specific compliance report format, they end up manually adding a filled example before it lands correctly. What should they do with this recurring observation?
Q4.Which statement correctly describes the first step in adjusting approach based on feedback?
Practice This Lesson