nesting-orientation-review-gate

Context

Use this when a nesting or packing heuristic changes orientation preference, tie-break ordering, or rotation defaults.

Pattern

1. Separate the preference case from the necessity case.

   • Preference: both orientations fit, but one orientation should now win.
   • Necessity: only rotation fits, so rotation must still happen.

2. Prove the rule on the real service path.

   • Assert actual placement geometry and Rotated90, not just a helper-level fit decision.

3. Add a preserved-rotation counterexample.

   • Show a case where the old rotate-to-fit behavior still works after the preference change.

4. Re-run the same request twice and compare placement order, coordinates, dimensions, and rotation flags.

   • Orientation tie-break changes can silently break determinism even when utilization stays the same.

5. Keep one no-fit regression check.

   • Preference changes must not rewrite impossible parts into misleading reason codes.

6. Prefer a narrow priority key over a heuristic rewrite.

   • If one explainability rule should outrank the existing orientation order only in a special case, add a dedicated leading sort key for that case and leave the rest of the ordering intact.
   • This keeps the change deterministic and prevents unrelated panels from inheriting a new rotation policy.

Anti-Patterns

• Approving the slice because a single happy-path panel no longer rotates
• Asserting only Success/Fits without checking Rotated90
• Treating a tie-break tweak as low-risk without a repeat-run determinism check