Adjust from Feedback

What you need to know

When you change several things at once, you cannot tell which change helped. Adjust one variable at a time, test, and keep what works. Use feedback from reviewers and users as the signal for what to change next.

One variable at a time

Change the prompt, or the knowledge, or the model, or the format — then compare. Keep a baseline output so you can tell whether a change improved anything. Use the same input each time so differences come from the change.

Turn feedback into edits

Translate reviewer comments into specific instruction or knowledge changes.
Test on several representative inputs, not only the one that failed.
Record what you changed and why — that log becomes the team’s tuning notes.
Revert changes that do not help; combine only the ones that do.

Worked example

A weak-but-usable draft gets “too technical” feedback. List possible changes (add context, tighten format, change model, add an example). Apply the first alone, compare to baseline, then the second alone. Keep only what helped. The final prompt includes only proven changes — and the log proves it.

Exam traps

Change several things at once
You cannot tell which change helped. Isolate one variable.
Test only on the one case that failed
Try several representative inputs — a fix that helps one case may hurt another.
Rewrite everything from vague feedback
Translate each reviewer comment into a specific prompt or instruction change.
Keep tweaking after the bar is met
Stop when the output meets the success criteria. Further tweaking is wasted effort.

Practice scenario

A draft is acceptable but a reviewer says it is too technical. You also wonder about adding an example, changing the model, and rewriting the whole prompt. What is the disciplined next step?

Option AChange the model, rewrite the prompt, and add an example all at once.Option BTurn the feedback into one specific change, keep a baseline, rerun on the same input(s), and only then combine changes that helped.Option CKeep tweaking forever even after success criteria are met.Option DIgnore the reviewer and swap to the largest model as the only experiment.

Build exercise

Change One Variable at a Time

7.2 · Intermediate · ~35 min · claude.ai

What you will practise

Why changing several things together hides what worked.
How to turn user feedback into one testable change.
How to compare before and after on the same input.
When to revert a change.

Keep a Domain 7 change log that shows each change applied alone with keep or revert.

Open Domain 7 exercises

Path: /learn/claude-associate/exercises/7-troubleshooting-optimization

Sources

CCAO-F Exam Guide v1.0 — Domain 7 task statement 7.2 — Adjust from Feedback
preporato/claude-certification-guide — optimization rule — one variable at a time (rewritten)
evggzzz/ccao-f-guide — task 7.2 Knowledge/Skills
docs/ccao-f-implementation-plan.md — Domain 7 lesson map

View progress