project-bug-huntinglisted

# Project Bug Hunting Build an auditable chain: **what the project promises → what happens under applicable conditions → why the difference is a defect**. Explore the whole project, including its documentation, examples, interfaces, implementation, configuration, tests, and delivery artifacts. Machine learning frameworks are one application, not the scope boundary. Report in the user's language. Use this for open-ended discovery. A supplied crash, failing test, or localized regression usually calls for focused debugging instead. Discovery delivers evidence and coverage; fixes, commits, and issue publication follow the user's requested scope. ## Keep Three Dimensions Separate | Dimension | Decision | | --- | --- | | Contract relationship | What should agree, and which promise makes that agreement necessary? | | Runtime scenario | Which version, dependencies, configuration, platform, operation sequence, installation mode, or cache state makes it applicable? | | Verification method | What source analysis, static check, behavior test, or trace can distinguish the candidate from its alternatives? | Start with a relationship, select a meaningful scenario, then choose sufficient evidence. Tools do not define additional contract groups. Do not expand these dimensions into an exhaustive Cartesian product. ## Exploration Loop 1. **Fix the baseline and budget.** Inspect repository instructions, revision and local changes, support claims, dependencies, build/test entry points, and