Known issues and limits

What the add-on does not do, or does imperfectly. Some are ours, some are the CAD side — either way, better known than discovered mid-job.

Imported CAD shades badly where the CAD itself is clean

You see shading artefacts on an imported STEP body in places the original CAD renders perfectly smooth.

The STEP export hands over a raw CAD solid with no error-correction layer, so whatever the modeller papered over internally arrives unpapered. Import repairs what it can — collapsed triangles, inverted normals, outliers judged against their own CAD face — but it cannot invent a good surface where the solid has a bad one.

Do: rebuild or regenerate the offending surface upstream, then re-export. See also CAD sources.

Large, complex curved patches with no subdivisions

One enormous curved face bounded by a few long curves is the hardest possible input: the fill has almost no boundary information to derive an interior from, and no algorithm covers every shape.

Do: split such a face into logical pieces before export. The fills follow the boundary, so giving them a sane boundary is the single most effective thing you can do — far more effective than fighting the result afterwards.

This is a limit, not a bug. It will not be "fixed" by a smarter classifier.

Auto can switch fill type while you tweak curves

You change a curve's spans and the patch comes back as a different strategy than before.

That is Auto working as designed: the strategy is judged by the result, and your tweak changed the input, so the winner can change. It is not random and it is not a regression.

Do: if you want a strategy pinned, press it — an explicit override sticks until you clear it back to Auto. See Fill types.

C is a region split, not a cut

It draws a new boundary and re-decides both sides; a tube takes no split at all. See CURVE → Region split.

Shift+Wheel skips flat patches — even flat strips

A flat strip between two arcs looks like a band of quads (the triangle fill dissolved into n-gons for viewing), shows no U/V readout, and Shift+Wheel over it does nothing; the status line reports it as planar.

That is the rule, not a threshold: the mass-span wheel drives a patch's transverse curves, and a flat patch has no transverse direction — the strip test alone would fire on half of a part's flat faces. Curvature is not "insufficient", the face is flat.

Do: tune the strip's arcs one at a time in CURVE — the wheel on a single curve works on any built patch.

No live streaming — re-imports are explicit

Hardcraft does not follow a live CAD source, but an edited CAD can be re-imported mid-retopo: Re-import CAD (STEP, one click) or Rebind Target (Plasticity) migrates your patches, spans, strokes and freezes onto the new source — see Re-importing an edited CAD.

What it cannot do: a patch whose CAD surface was rebuilt outright loses its old fill (there is nothing to hang it on — it rebuilds fresh), and the STEP path needs a current converter binary for face identity. The gesture is explicit by design; there is no watch-folder.

Merge is off the panel for now

The patch-merge workflow is contentious enough that shipping a one-click door to it does more harm than good. The operator still exists (F3 → HC: Merge Patches) but is at-your-own-risk until its problems are worked through; the button returns when that happens. UV Cut, which used to sit next to it, is gone from the build since 0.8.5 — the seam of a wrapped surface is a curve now (CURVE).

Tweak cost grows with the number of built patches

Each tweak writes parametric state back, so a fully-built complex part responds slower than an empty one. The heavy part was cut in early August; if a notch still bothers you, work in a region and build the rest later.

Not this: periodic viewport stalls while orbiting. Measured three ways, a bare cube in an empty scene stalls the same — that is the driver or Blender, not Hardcraft.