partforge

Mesh socket mount

hard tier · create

The request

Attached is the STL of an L-bracket. I need a bulkhead panel the bracket slides through. Build a flat rectangular panel standing in the XZ plane: 6 mm thick in Y, spanning y = 17 to y = 23 (that band of the bracket is clear of its holes and slots, so nothing is left standing inside the socket), and 10 mm larger on every side than the bracket's own silhouette seen along Y — its extent in X and in Z — centred on it. Cut a socket right through the panel's full thickness, and cut it with the attached mesh itself scaled up 5 % about the origin, rather than with a shape of your own: the socket has to follow the bracket's real L-shaped outline, not its bounding box, and the bracket has to pass through it without touching. Leave the mesh exactly where it sits — do not move or rotate it — and keep it imported in the part while you work. Put everything in one sub-part named exactly `mount`. Measure the mesh first; do not guess its size.
A render of the mesh the model was given
Mesh · made with FreeCAD 1.1.0 (Part workbench, MeshPart tessellation at 0.01 mm linear deflection) · CC0-1.0

What the AI judge checks

  • The overall form matches the reference: the same kind of object, the same silhouette from the matching angle, allowing for the requested change: the result is not the bracket but a flat rectangular panel with a socket cut through it in the bracket's own L-shaped outline, slightly enlarged; the bracket's holes, slots and thickness are not expected to appear in the panel
  • Proportions match: the relative sizes of the major dimensions agree with the reference within what the eye can judge, allowing for the requested change: the result is not the bracket but a flat rectangular panel with a socket cut through it in the bracket's own L-shaped outline, slightly enlarged; the bracket's holes, slots and thickness are not expected to appear in the panel
  • Every major feature visible in the reference (holes, slots, bosses, tabs, lips) is present in the result, allowing for the requested change: the result is not the bracket but a flat rectangular panel with a socket cut through it in the bracket's own L-shaped outline, slightly enlarged; the bracket's holes, slots and thickness are not expected to appear in the panel
  • Those features sit where the reference has them, allowing for the requested change: the result is not the bracket but a flat rectangular panel with a socket cut through it in the bracket's own L-shaped outline, slightly enlarged; the bracket's holes, slots and thickness are not expected to appear in the panel
  • Nothing is present that the reference does not show — no decorative or unexplained geometry, allowing for the requested change: the result is not the bracket but a flat rectangular panel with a socket cut through it in the bracket's own L-shaped outline, slightly enlarged; the bracket's holes, slots and thickness are not expected to appear in the panel
  • The opening is L-shaped, following the bracket's own profile, rather than round, rectangular or a rough approximation of it
  • There is exactly one opening through the panel — no extra holes, slots or pockets
  • The material around the opening is unbroken — no part of the opening runs out to an outer edge of the panel

Measured checks

  • one sub-part named mount (gate)
  • the imported bracket is still declared as an import (gate)
  • the panel is 95 by 65 mm, 6 mm thick, and stands where the prompt puts it
  • the socket clears the bracket grown 3 % (gate)
  • the panel is still solid in the L's open corner (gate)
  • the socket is open through the panel where the upright passes (gate)
  • asked the oracle at most once

Ranked by hand

One reviewer put one attempt from each model in order, best first, ties left tied.

  1. #1
    Gemini 3.8 Flash
    Gemini 3.8 Flash's ranked attempt
  2. #2
    Fable 5.1
    Fable 5.1's ranked attempt
  3. #3
    Opus 5
    Opus 5's ranked attempt
  4. #4
    GPT-5.6 Sol
    GPT-5.6 Sol's ranked attempt
  5. #5
    GPT-6 Astra
    GPT-6 Astra's ranked attempt

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