eduos-rotation-probe-teacherless-class-fixture-gap-teacher-leg-unprovable

Answer

Diagnosed and solved. The environment contains no application codebase — only /workspace/problem.json — so I reconstructed the probe's candidate-selection logic, reproduced the false-FAIL and the fix with a faithful SQLite model, and wrote the solution.

Deliverables (both in /workspace): - solution.md — the self-contained writeup (title, root-cause analysis, exact fix, general pattern, verification). - verify_fixture_gap.py — runnable reproduction; exits 0.

Root cause

The probe needs both RBAC legs provable on the same fixture row: - student leg: assignment not in enrollments → expect 404 - teacher leg: classes.teacher_id IS NOT NULL (teacher owns the class) → expect 200

The old candidate select only guaranteed "unenrolled":

SELECT a.id FROM assignments a
WHERE a.id NOT IN (SELECT assignment_id FROM enrollments WHERE student_id=:student_id)
LIMIT 1;

After the reseed, the only unenrolled assignments sat in teacherless classes (teacher_id IS NULL). The endpoint correctly 404s a non-owning teacher, so the 200 was unprovable by fixture, not broken by product — a false FAIL.

Fix

Join classes and require an owning teacher, and treat zero candidates as a named SKIP rather than a FAIL:

SELECT a.id
  FROM assignments a
  JOIN classes c ON c.id = a.class_id
 WHERE a.id NOT IN (SELECT assignment_id FROM enrollments WHERE student_id = :student_id)
   AND c.teacher_id IS NOT NULL
 LIMIT 1;

if candidate is None:
    return TestResult(status="SKIP",
        reason="fixture-gap: no unenrolled assignment in a teacher-owned class "
               "(classes.teacher_id IS NOT NULL); student=404/teacher=200 unprovable")

Verification

The script reproduces all four discriminating cases and passes:

[OK ] old select, teacherless-only fixture                 -> FAIL (assignment 100: got student 0/teacher 0)
[OK ] fixed select, teacherless-only fixture (the bug case) -> SKIP (fixture-gap: no unenrolled assignment in a teacher-owned class)
[OK ] fixed select, teacher-owned candidate available      -> PASS (assignment 101: student 404 / teacher 200)
[OK ] old select, mixed fixture (picks teacherless first)  -> FAIL (assignment 100: got student 0/teacher 0)
All fixture-gap semantics verified.  exit=0

This matches the live result: 26 PASS, 0 FAIL, 1 SKIP, exit 0.