Tech SkillsCheckable Tasks

Goal Evaluator Evidence Question Explained

This worked Loop Engineering practice question tests Checkable Tasks. Try the answer first, then review why A is the best choice.

Short answer

The correct answer is A. The evaluator judged the conversation, and no test output appeared in it.

The /goal docs say a separate evaluator model checks the condition after each turn, and it judges against what Claude has surfaced in the conversation. It does not run commands or read files on its own. A sentence saying tests should pass is the only evidence it saw, so it accepted it. Write the goal so the test output itself must appear in the conversation. The correct answer is A. The evaluator judged the conversation, and no test output appeared in it

The Question

You run /goal "All tests pass." Claude edits code, writes "tests should pass now," and the goal is marked met. You run the tests yourself and two fail. What went wrong?

AThe evaluator judged the conversation, and no test output appeared in itCorrect
BThe evaluator ran the tests against a cached copy of the repository from before the edits
CThe /goal command checks the condition once at the start, not at the end
DTwo failing tests are below the evaluator's default failure threshold

Why A is correct

The /goal docs say a separate evaluator model checks the condition after each turn, and it judges against what Claude has surfaced in the conversation. It does not run commands or read files on its own. A sentence saying tests should pass is the only evidence it saw, so it accepted it. Write the goal so the test output itself must appear in the conversation. The correct answer is A. The evaluator judged the conversation, and no test output appeared in it

Why the other options are wrong

Option B: The evaluator ran the tests against a cached copy of the repository from before the edits

The evaluator does not run anything, cached or live. It reads the conversation and calls no tools.

Option C: The /goal command checks the condition once at the start, not at the end

The evaluator checks after every turn, not once at the start. The problem is what it was given to read, not when it read it.

Option D: Two failing tests are below the evaluator's default failure threshold

There is no failure threshold. The evaluator judges the stated condition against the evidence shown and nothing else.

Key idea: Checkable Tasks

Loop engineering is a practice skill. Open Claude Code in a throwaway folder, write one small task in a file, and run it with /loop or /goal while you watch. If you doubt an answer, test it against a real run in that folder.

Ready to test the full skill?

Take the free Loop Engineering quiz and find the exact basics to review next. No signup required.

Practice related questions

Same skill area or adjacent concept. Use these after reviewing the explanation.

Related Loop Engineering Practice Questions