grading

Run grader_fetch.py on the challenge directory, then grader_grade.py --bulk followed by grader_consensus.py for 3-pass consensus grading. Show the generated comments and old-to-new grade preview in chat, then push with grader_push.py after the instructor approves two guardian pop-ups.

grader_fetch.py de-identifies submissions into opaque keys before anything reaches the LLM, and names live only in gitignored keymap files. Re-identification uses grader_reidentify.py keyed by user_id, and Zone-2 name-bearing files are never read or displayed.

No. Grades and comments reach Canvas only through grader_push.py or grader_standing.py. The grade_guardian hook deterministically blocks creating, editing, or running hand-written scripts that write to Canvas, because they bypass the review gate, duplicate-comment protection, and disclosure tags.

grader_push handles assignments that take student submissions and is keyed on submission files with a regrade gate. grader_standing handles No-Submission standing columns, is roster-keyed by user_id, overwrites freely, and accepts --yes because values are instructor-computed.

Rows are skipped when already pushed (use --regrade for genuine resubmissions), when the push would lower an existing grade (use --allow-lower), or for inactive and Test students (use --include-inactive). The fix is always a flag on the tool, never a direct API call.

Yes, for courses on other LMS platforms with local-only inputs. Declare the course's Zone-2 files in .claude/ferpa_zone2.txt, work from roster ID codes, and use local_feedback_join.py to join names back into code-keyed feedback at output time.