threejs-landmark-placement
Done — solution written to /workspace/SOLUTION.md.
Self-contained markdown with root cause, one-line fix, and a verification table demonstrating the off-by-half bug and the corrected section lands.
# Evidence - Problem class: threejs-landmark-placement - Model: openrouter/deepseek/deepseek-v4-flash-0731 - Solved: 2026-08-22T23:13:28.928Z - Verification: solution produced by pi in sandbox; see signatures.json
{"description": "NEON RUSH TR-004 post-debug: section landmarks placed one section early because landmarkGiFor computed (start+len)/2 instead of start+len/2 \u2014 a pure math grouping bug (off-by-half-section). CAUGHT BY: independent vision QA of probe screenshots (construction crane visible at 790m inside HARBOR; tunnel portal at 1100m inside HARBOR; harbor arch at 575m at the CITY boundary) \u2014 the worker harness only asserted the landmark EXISTS at the probed gi, never that its section matched, so its present:true checks passed while every landmark after the first was misplaced. FIX: one-line paren fix + probe asserting landmark gi AND sectionIndexAt(center) match, plus negative assertions on old gis. LESSON: when verifying per-section features, assert section membership, not just existence; and vision-check screenshots at each probe independently.", "environment": "", "language": "", "model": "openrouter/deepseek/deepseek-v4-flash-0731", "problem_class": "threejs-landmark-placement", "provider": "openrouter", "solved_at": "2026-08-22T23:13:28.929Z", "version": ""}