Bracket second mount
hard tier · iterate
The requests, in order
1Make an L-bracket, lying in the XY plane with Z up. A base plate 50 mm along X by 40 mm along Y by 4 mm thick, centred on the origin in X and Y with its bottom face on Z = 0, plus an upright wall 50 mm wide, 40 mm tall and 4 mm thick standing on the base's +Y edge, so the bracket's overall height is 40 mm, with the wall running from Z = 0 to Z = 40, its outer face at Y = 20 and its inner face at Y = 16. Put two 4.4 mm holes through the base at X = ±15, Y = −10. Name the sub-part exactly `bracket`.
2Add two more 4.4 mm holes, this time straight through the upright wall along Y, at X = ±15 and Z = 30. Keep the base holes exactly where they are.
3Lighten the upright with one 20 mm diameter hole through it, centred at X = 0 and Z = 20, without touching any of the four existing holes.
What the AI judge checks
- An L-shaped bracket: a flat base plate with a wall standing up along one edge
- Two small round holes through the flat base plate
- Two small round holes through the upright wall, near its top
- One much larger round hole through the middle of the upright wall
- The large hole does not run into or overlap any of the small holes
Measured checks
- one sub-part named bracket (gate)
- base hole at X = 15 is 4.4 mm (gate)
- base hole at X = -15 is 4.4 mm (gate)
- one sub-part named bracket (gate)
- base hole at X = 15 is 4.4 mm (gate)
- base hole at X = -15 is 4.4 mm (gate)
- upright hole at X = 15, Z = 30 is clear (gate)
- upright hole at X = -15, Z = 30 is clear (gate)
- one sub-part named bracket (gate)
- base hole at X = 15 is 4.4 mm (gate)
- base hole at X = -15 is 4.4 mm (gate)
- upright hole at X = 15, Z = 30 is clear (gate)
- upright hole at X = -15, Z = 30 is clear (gate)
- 20 mm lightening hole is clear
- material remains just outside the 20 mm hole
Ranked by hand
One reviewer put one attempt from each model in order, best first, ties left tied.
Ordering by hand decides Human rank and Human rate on the leaderboard; it never affects Overall or Strict.
Every model's attempts
Run baseline-2026-092026-09-10commit 41aec7daspend $339.70 + $48.02 AI judging, 7 unpricedOther runs: 2026-09-04T20-02-10-stage2-hard-cases, 2026-09-04T19-48-31-screen-14-fixed