ForgeFX SimulationsREAD THE 100-WORD VERSION →
FORGECAD / DEEP-DIVE PLAN

CAD in. Editable real-time asset out.

The target is not merely “opens in Unity.” ForgeCAD exists to produce intelligent, human-editable, VR-ready geometry with correct structure, units, pivots, UVs, and textures—and to prove quality with a repeatable 0–100 judge.

Halliburton industrial CAD assembly
HOLY GRAIL / D00416063-Tier4-I_2026_07_07.zip / 3.43 GB
8Dave-derived score dimensions
0–100One weighted final score
8 stagesPedal to complete machine
1 real targetHalliburton Tier 4 assembly
01 / Problem

“Converted” is not the finish line.

CAD tessellation can preserve the shape while destroying the model’s usefulness: dense cloth-like topology, meaningless object trees, bad pivots, arbitrary scale, overlapping UVs, and baked-looking materials. Dave’s standard is tougher: the output must run in VR and remain legible and editable to an artist who needs to select a part, move an axis, add a label, revise a texture, or optimize a subassembly.

Mission: teach ForgeBot and ForgeFX artists a repeatable conversion workflow that gets better as asset complexity rises, without hiding manual cleanup or accepting one-click mesh soup.
02 / Pipeline

A measured conversion bench, not one magic exporter.

Use CAD-aware ingestion first, art-aware cleanup second, runtime proof third. The same manifest and judge follow every candidate so tool changes remain comparable.

01 / INGEST

CAD-aware import

Primary: Unity Asset Transformer / Pixyz. Fallbacks: CAD Exchanger, SimLab, Okino. Preserve assembly references and source metadata.

02 / OPTIMIZE

Purposeful geometry

Retessellate, remove hidden internals where allowed, generate LODs, merge only static groups, and retain interactive parts separately.

03 / FINISH

Artist handoff

Use Blender or the winning DCC lane for naming, pivots, UV layout, materials, texture authoring, collisions, and final inspection.

04 / PROVE

Runtime + judge

Import into a clean Unity test scene, profile target VR behavior, run automated checks, then complete a blinded artist review.

03 / Curriculum

Earn complexity one asset at a time.

Each rung advances only after the prior rung passes. This isolates failure causes and turns every conversion into reusable evidence instead of one heroic cleanup.

Pedalsingle object
Consolecontrols + labels
Cabinterior hierarchy
Boomarticulated assembly
Chassislarge structure
Harnessthin routing
Enginedense mechanics
Machineholy grail
04 / Judge

Every Dave constraint gets its own score.

The roll-up uses fixed weights. A high texture score cannot conceal broken geometry; hard gates cap a result that is unusable in practice.

25%
Intelligent geometryCorrect silhouette and parts; clean normals; no non-manifold or cloth-like topology; deliberate edge flow; no critical omissions.
15%
VR runtime qualityTriangle, draw-call, memory, load-time, LOD, culling, and target-frame-time evidence under a declared hardware profile.
15%
Human editabilityLogical hierarchy; interactive parts remain separate; artist can select, modify, replace, and export without reconstruction.
15%
UV qualityLow distortion, intentional seams, useful texel density, no unintended overlap, padding, and reserved label/text editing space.
10%
Texture qualityReadable PBR materials, correct assignments, useful resolution, consistent channels, no baked lighting or missing maps.
8%
NamingStable, descriptive, unique names for assemblies, parts, materials, textures, LODs, and collisions.
7%
Axes + pivotsDeclared up/forward convention; pivots sit on real joints, hinges, controls, wheel centers, or intended selection origins.
5%
Real-world scaleCentimeter-authored dimensions match source measurements and arrive in Unity at correct physical size.
05 / Test Method

Automation measures facts. Artists judge usefulness.

Each candidate produces the same evidence bundle: source manifest, conversion logs, mesh statistics, UV checks, material inventory, runtime profile, turntable captures, and a human-editability task sheet.

Automated inspection

  1. Count parts, meshes, triangles, materials, textures, UV sets, invalid geometry, and missing references.
  2. Check unique names, naming schema, transform cleanliness, scale tolerance, axes, and pivot placement rules.
  3. Measure UV overlap, distortion, bounds, padding, and texel density.

Runtime proof

  1. Import into a clean Unity scene with known settings.
  2. Measure load time, memory, frame time, draw calls, and LOD transitions on named hardware.
  3. Exercise selection and articulation on candidate interactive parts.

Blinded artist review

  1. Give the output—not the tool name—to Brandon, Ed, or another artist.
  2. Ask them to rename a part, move a pivot, add a label, change a material, and export.
  3. Score task success, time, repair count, and confidence.
06 / Source Asset

The Halliburton target is identified.

Dave’s July 8 message points back to the Halliburton “Elite phase 1 batch.” Direct Graph inspection resolves that folder to the 3.43 GB archive below. We do not start there; we preserve it as the graduation test.

Prior CAD conversion route research
EvidenceVerified state
Slack sourceVERIFIED Dave: “grab this CAD file” for AI geometry, initial texture, and UV mapping.
FolderVERIFIED Elite phase 1 batch; 3,957,000,076 bytes total.
Holy grail archiveVERIFIED D00416063-Tier4-I_2026_07_07.zip; 3,433,599,308 bytes (3.43 GB / 3.20 GiB).
Source formatsTHREAD-EVIDENCED SolidWorks SLDASM and SLDPRT with assembly references.
Current conversionNOT CLAIMED Identified, not yet judged as a ForgeCAD pass.
07 / Delivery

What ForgeCAD will make visible.

The site becomes the operating scoreboard: mission, benchmark ladder, current candidate, evidence, per-constraint judge, final score, failures, and the exact next experiment. No generic product marketing and no unverified “AI converted it” claims.

Every run publishes

Input manifest · converter and settings · elapsed time · before/after stats · turntable images · Unity profile · artist task results · eight subscores · weighted final score · hard-gate status · fix list.

Promotion rule

Advance to the next asset only when the current asset scores at least 80/100, triggers no hard gate, and a human artist confirms the geometry and UVs are editable without reconstruction.