Source-bound English reading edition. The original as_of date, source URLs, dates, locators, claim IDs and factual approvals remain in the canonical metadata. Review of this edition is separate and does not refresh those facts or authorize runtime.

SWE-bench: inputs and verification boundaries for one subset

The SWE-bench Lite dataset card at revision 6ec7bb89b9342f664a54a6e0a6ea6501d3437cc2 describes the task as fixing an issue given a repository and a GitHub issue, with results verified by unit tests. Its Data Instances section separates what a system is given from what checks it:

Field	Role
problem_statement	The issue title and body: what the system is given.
base_commit	The repository state before the original fix was applied.
patch	The answer patch from the original fix, without its test changes.
test_patch	The test changes from the original fix.
FAIL_TO_PASS	Tests tied to resolving the issue.
PASS_TO_PASS	Tests that should pass both before and after the fix.

Read the dataset card at that revision, especially "Dataset Summary", "Supported Tasks and Leaderboards" and "Data Instances". It names dev and test splits but declares no dataset license, so link to it rather than copying data or patches.

Use the fields to ask three questions about any reported score: what was supplied to the system, what could the tests detect, and which instances stayed in the report. A passing FAIL_TO_PASS set shows that the named tests pass; it does not show that the fix is the only correct one or that untested behavior is right.

The course's A6 evaluation exercises are course-original synthetic exercises. They are not a sample of SWE-bench, and a result on them is not a SWE-bench Lite score.