CCAR-F / Free hands-on exercise / Version 1

Can a passing helper hide a broken caller?

Repair a small synthetic repository, execute its tests and bind review evidence to the source you actually checked. Then change the interface from kilograms to grams and investigate why the old local check still passes.

This exercise runs on your computer with Python 3.10 or newer and its standard library. No sign-in, package installation, Claude account, API key or network connection is needed after downloading. Familiarity with basic Python functions, dictionaries and exceptions is helpful. Your edits and test results stay local; this page does not collect submissions or save progress.

Download the Python starter (ZIP)

Start with failing tests

  1. Check Python with python3 --version. On Windows use py -3 --version.
  2. Extract the starter into a new empty folder. Open a terminal in that folder, then read README.md and CONTRACT.md.
  3. Run the command below. The unfinished validator, missing-input case and changed caller intentionally fail. Passing helper checks alongside those failures are part of the lesson.
python3 -m unittest -v

Windows equivalent: py -3 -m unittest -v. Use a disposable local folder, separate from customer or production repositories. The download contains source, instructions and executable tests; it needs no setup script.

Make two different repairs

  1. Original contract: the fee helper accepts kilograms and must return zero for missing input. Reproduce the failure, make a focused fix and retain the two-kilogram result of eight synthetic credits.
  2. Changed interface: a caller now receives grams while its helper still accepts kilograms. Execute the actual caller with 2000 and 250 grams. Fix the caller boundary without silently changing the helper's units or breaking missing input.
  3. Review the current revision: implement the finite synthetic report validator in review.py. Reject malformed or failed output, stale source digests, incomplete check declarations and unsupported line evidence. Reconcile duplicates only after validation.

Run all tests without changing them to obtain a pass. Write a short MY_REVIEW.md with your current digest, reproduced source locations, separate helper/caller check results and remaining uncertainty. The README gives the digest command. A person must still review your explanation; the tests do not judge its reasoning.

Compare after your attempt

Keep your work. Extract the worked solution into another new empty folder, run the same tests and compare its walkthrough with your evidence. It is a separate reference, available when you deliberately open the section below.

Reveal the worked solution download

The solution repairs both interfaces and validates the local envelope. Its walkthrough explains why a passing job or matching source line is insufficient on its own.

Download the separate worked solution (ZIP)

What the checks establish

Python executes the supplied synthetic helper and caller. The validator checks this lesson's local envelope and literal source locations. Its statuses are declarations, not authenticated evidence from a CI provider. A matching line cannot prove a finding is true, an empty report cannot prove every issue was found, and passing these finite tests cannot establish correctness for other inputs or systems.

The instruction, path-rule and forked-skill files are teaching artifacts. This exercise does not execute Claude, load Claude Code instructions, invoke a skill, register an SDK hook or test the real CLI output schema. These original CertLabs scenarios are not official exam questions, a full exam simulation, calibrated assessment or evidence of mastery, transfer effectiveness, exam readiness or passing probability.

Connect the practice to current documentation

Primary sources checked October 11, 2026. The fee and envelope contract is original to this lesson. Provider behavior and schemas may change.

  • Claude Code project memory and rules describes instruction scope and path-specific rules. Instructions supply context; they do not enforce actions.
  • Claude Code skills describes reusable tasks and forked context. Merely placing the example artifact in this ZIP does not demonstrate invocation.
  • Programmatic Claude Code documents actual CLI execution and output. This exercise uses its own explicitly synthetic envelope.

Keep practicing with evidence