# Change one thing and compare

Ask a useful question without losing your first attempt.

Canonical: https://roboty.dev/guide/use/change-one-thing
Content revision: 7
Platforms: iphone, mac, browser, agent

## Name your question

Start with one concrete question: will changing the target height make this goal reachable? Record your expectation before the second attempt.

## Keep the baseline

Retain the first executed revision and result. Return with Edit goal. Change one supported value so that you can explain what is different. The new draft does not change the old result.

## Review another attempt

Prepare and explicitly confirm the new attempt through the selected owner. Compare like tasks and methods. If the method, starting conditions or success criteria also changed, name those differences rather than attributing the result to height alone.

## Interpret carefully

Useful evidence includes failure. A better result suggests a next question, not universal superiority. Training, automatic improvement and physical generalization require their own workflows and evidence.

## Lesson 3: a fair comparison

### Example

Return to the same table-mounted arm. Trial A uses our target (20, 0, 15) cm and the 1 cm success rule. A **controller** is the program that chooses how the arm moves.

The following fictional results are supplied for this exercise. You do not need to run a simulation.

| Trial | Target height | Controller | Final distance from target |
| --- | --- | --- | --- |
| A | 15 cm | Original program | 2 cm |
| B | 17 cm | Different program | 0.5 cm |

In both trials, X = 20 cm and Y = 0 cm, the arm starts in the same position, and the success rule stays within 1 cm.

### Your turn

Write or choose your answers before checking:

1. Which trials meet the success rule: A, B, both or neither?
2. Name the two things that changed from A to B.
3. Can these results show that raising the target caused the improvement?
4. You want to investigate height alone. Choose a fair comparison with A: **C:** height 17 cm, original program, same starting position and success rule. Or **D:** height 17 cm, different program, same starting position and success rule.
5. Before looking at the explanation, predict whether C will succeed, fail, or remain uncertain. What measurement would you need to check your prediction?

### Check your reasoning

1. **B only.** Its 0.5 cm error is within 1 cm. A’s 2 cm error is outside it.
2. **Target height and controller.** Both changed, so the improvement could relate to either change or their combination.
3. **No.** These two trials do not isolate the effect of height.
4. **C.** Compare A and C with only target height changed. Keep the controller, X and Y, starting position and success rule fixed.
5. **Any of those predictions is a starting hypothesis, not an established result.** C has not been run and no result is supplied. Measure its final distance from its target, then compare that distance with the 1 cm rule. Record both your prediction and the outcome, even if they disagree.

You have designed a comparison that isolates one change. Repeated trials would still be needed to assess consistency. Fictional examples and simulated results do not establish how a physical robot will perform.

Ready to apply the three lessons? [Choose your next step](/guide/get-started).

## From this written example to a live task

The fictional 1 cm final-position rule above is a teaching example. A live task has its own displayed success criterion. Roboty’s supported Reach criterion requires the tool point within 12 mm for 250 continuous simulated milliseconds. Read the rule shown with the selected task before comparing an actual result.

Content ID: compare-one-change
Compiled digest: 7679d151b41236402446a8c037755e8b6c07be166b2add83293e3c702d4f3a39
Language: en