How Jonathan Guides

Patterns from debugging intersection line occlusion (Feb 2025).

1. "You already wrote the code" — stop reinventing; the algorithm exists, find it and use it
2. "Stop / undo" — prevent premature changes when the approach is still unclear
3. "We are going in circles" — re-read the conversation instead of asking the same question
4. "Analyze the image" — look at actual visual output instead of theorizing
5. "Right length, just offset" — visual observation that pinpoints the bug class faster than code inspection
6. "Those tiny lines are edges of SO" — correct wrong assumptions about which code is responsible
7. "Could this be another world/screen t mismatch?" — guide toward the shared root cause instead of treating symptoms separately
8. "Can we increase precision instead?" — steer away from band-aids toward real fixes
9. "Try removing it entirely" — eliminate variables to narrow the search

Anti-pattern: discarding a working fix

The collaborator had the correct fix (project world-space crossing point to screen, work in screen-space t throughout) but then layered a half-measure on top — projecting bs/be correctly but still mapping the polygon clip's t back to world-space linearly. This introduced a second bug (tiny stray edge fragments) that took additional rounds to diagnose. Lesson: when a fix is structurally correct, carry it through to completion. Don't mix the old approach with the new one.

The pattern

Jonathan diagnoses visually, collaborator translates to code. He keeps the collaborator honest by demanding evidence, stopping premature action, and pointing at what he can see rather than letting the collaborator speculate.