How generation and quality validation work

CodeTeach treats generated files as a candidate assignment, then checks the package before GitHub publication. The instructor reviews and approves the result.

Generation

Forward uses your starter, Backward uses your solution, Both Provided preserves both, and From Document creates code from your prompt. Instructor-owned uploaded or approved artifacts are not silently rewritten. CodeTeach creates missing artifacts, including tests and student instructions.

Expected results come from execution

CodeTeach runs the working solution against proposed inputs and records what it actually does. AI can suggest scenarios, but final expected test output must be derived from observed solution behavior. Review the solution itself: an incorrect solution can produce internally consistent yet pedagogically wrong tests.

Quality validation and Deploy

After you approve the current package, Tier 1 reruns the solution and all tests in a sandbox and checks package alignment. Edits can invalidate that approval and result. During Deploy, Tier 2 builds the GitHub Actions workflow and grading metadata deterministically and requires an internal passing run before publication. CodeTeach then watches the real GitHub Actions run. A post-push failure remains visible and should be fixed before student distribution.

There is no partial-pass shortcut. Follow the validation recovery guide if any check fails.