OneCAD/ROADMAP.md

SolveCraft roadmap

Status as of 2026-10-07. SolveCraft targets Fusion-style parametric modelling (sketch → features → parametric timeline) as an open, pure-Rust, agent-drivable application.

At a glance

Question	Answer
Where are we?	M0 (vertical slice) done. Sketch with solver, extrude/revolve/fillet/chamfer/cut, parameters that rebuild the timeline, STEP/STL export, desktop app, CLI, oracle harness.
Command parity (SOLID + SKETCH toolbar)	46 / 291 (16%) — docs/parity.md
Fusion oracle	2 / 29 parts match Fusion's measurements (16 need features not built yet) — docs/oracle.md
Tests	60 (solver, profiles, kernel booleans/blends/measures, expressions, timeline, file formats, camera, engine end-to-end, hostile-input fuzz over every command)
Gates	cargo xtask ci: fmt, clippy -D warnings, tests, asset attribution, layering — green
Weighted parity estimate	≈ 5% of Fusion's Design workspace by importance (sketch + basic solids are the core, but surfaces, assemblies, sheet metal, CAM, drawings are untouched)
Time to a useful alpha (M0–M6)	≈ 150 agent hours remain

Milestones

#	Milestone	Status	Hours left
M0	Vertical slice	done	—
M1	Sketch depth, oracle sketch cases (arcs, taper, two-sided, sketch modify tools)	in progress	16
M2	Robust modelling: coplanar booleans in the kernel, holes, patterns, mirror, shell, draft, split, construction geometry	planned	24
M3	General fillets and chamfers (curved edges, chains, vertex blends)	planned	30
M4	Files: STEP import, mesh import, DXF sketches, 3MF	planned	16
M5	Persistent naming of faces and edges	planned	20
M6	UI depth: profile shading, selection filters, measure tool, section view, edit dialogs, drag-solve	planned	24
M7	Sweep, loft, rib, web, emboss, thread, coil, pipe	planned	24
M8	Surface workspace	planned	30
M9	Assemblies and joints	planned	40
M10	Sheet metal	planned	30
M11	2D drawings	planned	30
M12	Release: installers, signing, docs, performance	planned	16

Known limitations

• Fillets and chamfers: straight convex edges between planar faces with perpendicular planar end faces; two blended edges may not share a corner yet.
• Booleans of bodies with coincident faces fail unless they come from extrudes (where the tool is extended automatically); fully internal voids are not supported.
• Mass properties come from fine tessellation (curved faces within ~1e-4 relative).
• No STEP import yet; no persistent naming (fillet edges are re-found by position).

Kernel

truck (Apache-2.0) behind our own kernel crate, with our own blend operation and boolean retry/validation; see plan/adr/0001-geometry-kernel.md (local) — summary in README.