How to find edge cases in Python interview code

How do I use Break My Code?

Break My Code is available on all 13 interview exercises, including Two Sum, Number of Islands, and LRU Cache. It tests a snapshot of the code currently in the editor, so results describe that version rather than later edits. The free rule-based tests do not require sign-in or an AI key.

Run Tests checks the exercise's existing suite. Break My Code adds a separate set of targeted cases for common implementation mistakes. Neither replaces understanding the problem statement or checking your algorithm's complexity.

1. 01Open Two Sum or another problem from Interview Problems.
2. 02Keep the starter function name and implement its required behavior.
3. 03Select Break My Code and wait for the case results. The first run downloads Python runtime files.
4. 04Expand failures to inspect the input, expected and actual values, and edge-case lesson.
5. 05Edit your solution, rerun Break My Code, then select Run Tests to check the original suite.

Which Python edge cases should I test?

Start with the smallest valid input, boundaries where a loop starts or stops, repeated values, zero and negative values when allowed, and cases that reach the last element. Then test a larger input to expose unnecessary repeated work. Only test values allowed by the exercise contract: an invalid input is not evidence that a valid solution is wrong.

For Two Sum, nums = [17, 17] and target = 34 require two different indices, [0, 1] or [1, 0]. Returning [0, 0] reuses the same element and is wrong. Also test nums = [6, 2, 10] and target = 12: the valid answer is [1, 2] in either order, not the first index used twice.

Do not assume every problem accepts any output ordering. The comparison follows the exercise contract; deterministic output requirements still matter even when two results contain the same values.

How do I read expected and actual results?

Expected is the result from a separate reference run. Actual is your implementation's return value for the same input. A failing case may instead show an exception or an execution timeout. The input and lesson help you distinguish a boundary error from a performance problem.

Your candidate and the reference run in separate browser workers with independent input copies. A reference execution error is skipped rather than counted as a failure in your code. Passing these cases does not prove correctness for every possible input.

How are LRU Cache operations tested?

An LRU Cache case is a sequence of constructor, put, and get operations, not one ordinary function call. Check the per-operation expected and actual table to find where behavior first differs. put produces no return value, displayed as null, while a missing get returns -1.

Important cases include capacity zero, capacity one, reading an item before eviction, and updating an existing key. A successful read or update changes recency, so an error may only become visible in a later operation. Large tables show a bounded selection around a mismatch instead of rendering every operation.

What happens if my Python code hangs?

Select Stop to terminate the current adversarial run. Once a case's Python runtime is ready, a roughly two-second watchdog terminates an implementation that takes too long. Runtime downloading has its own longer loading timeout; cold downloads do not use the execution budget.

Some rule-based probes use about 10,000 items. Browser speed and available memory affect results, so a timeout is a reason to inspect your algorithm, not a universal performance benchmark. Use smaller examples and the visualizer to understand a failure before trying a large case again.

Are AI-generated cases required, and is my code uploaded?

No. Free rule-based cases and Python execution run in the browser. Optional AI controls appear only when the site has a server-side provider configured. With your consent, AI generation sends your current code and problem statement to the provider; an explanation request can also send the selected input, output, and error. Do not submit secrets or private data.

AI suggestions are validated against the problem's input contract but can still be unhelpful or incorrect. They are labeled separately from rule-based cases and are subject to usage limits. Read the privacy policy before opting in; access to an AI explanation is not a guarantee of a correct diagnosis.