Expert hard surface is controlled light on edges: every edge catches a highlight from a bevel or a support loop, flat faces read perfectly flat, and cuts stay editable until the design is settled. The experts disagree on method (n-gons and booleans vs quad cages and subdivision), not on the checks: order of operations, cutter hygiene, planar faces, a shiny matcap orbit. The agent builds the stack with the toolkit and lets the measurable gates find what a human finds by orbiting. If a sibling skill named here is missing from your available skills, ask the user to install it (npx skills add scenario-labs/skills --skill <name>); unattended, proceed from tool schemas and flag the gap.
REQUIRED BACKGROUND: scenario-blender-expert (execution channel, review loop, 5.2 API changes). Toolkit: scripts/bx_hardsurface.py (import sys; sys.path.append(\"<skills>/scenario-blender-hard-surface/scripts\"); import bx_hardsurface as HS), world-space meters, stock Blender only.
HS.order_stack does it and HS.add_boolean always lands above the tail (verified).HS.clamp_culprits names the cut.| Input | \nWhy it changes the plan | \nDefault when silent | \n
|---|---|---|
| Purpose | \nrender: 3+ micro segments (Gambrell); game mid-poly: 1 to 2 segments, triangulated, custom normals; bake: subD or boolean high + separate low | \nrender still | \n
| Real size | \nbevel widths are absolute: du Mont uses about 3 mm on a small robot body, 1 mm on small parts, 0.1 mm on a collar | \nmodel at real size; if unknown, micro width 0.5% of the largest dimension [added] | \n
| Surface | \nmostly planar vs curved vs organic-mechanical | \nSchool A for planar parts, B for curved shells, hybrid when mixed | \n
| Budget and engine | \nAugusto: 15k tris, one 2K map; a quad-only pipeline forces School B (Gambrell [00:24:35]) | \nno budget: report tris | \n
| Later stages | \nsculpted damage or multires detail needs a quad cage (PzThree); rigging needs collapsed copies (du Mont) | \nnone | \n
| Symmetry | \nmirror the target when the part is symmetric; mirror the cutter around the target when only the cut is (Gambrell [00:14:53]) | \nask the reference | \n
HS.apply_scale(ob) (keeps children in place, flips winding on negative scale). Check proportions against a human-scale box through a long lens (du Mont [00:10:01]). GATE: scale is 1 on every object with a bevel, solidify or array; silhouette sheet reads (bx_review.review(..., modes=(\"silhouette\",))).HS.find_edges(ob, min_angle=80, inside=...) + HS.set_edge_attr(ob, edges, 1.0), then HS.hard_surface_stack(ob, micro_width=..., design_width=...).HS.subd_stack(ob, support_width=...) (Bevel Weight, even segments, profile 1.0, Subdivision with UV Smooth Keep Corners) or explicit HS.support_loops.\nGATE: HS.stack_problems(ob) == []; A has custom normals on the evaluated mesh.HS.make_cutter(target, \"BOX\"|\"CYLINDER\"|\"SLOT\", size, location, ...) builds a cutter that meets the checklist (scale 1, smooth, Cutters collection only, wire, hidden from render, parented, EXACT, wired above the tail). HS.as_cutter for custom shapes, HS.slice_boolean for Slice. Rules: extend past every surface entered; keep cutter edges far from target edges or exactly on them; offset a round cutter by half a segment against a target with the same count (same phase gave 10 non-manifold edges and streaks [verified]; make_cutter default); round cutter corners in facets well under the bevel angle (4+ segments per 90 degrees). GATE after each cut: HS.cutter_report(c, target) clean, HS.clamp_probe(target)[\"ratio\"] < 0.1, a HS.closeup of the cut looked at.r = HS.shading_audit(ob); HS.verdict(r), then HS.review([ob], out, modes=(\"matcap\",\"reflect\",\"hsgrey\",\"wire\")) and closeups. Triage with references/critique.md: smear on flat faces, non-planar faces, cuts on curvature (render first: Harden + WN already shaded 24/48-segment cylinder cuts cleanly; on a sphere HS.transfer_normals cut the normal error p95 from 2.79 to 0.34 degrees [verified]), mirror seam (HS.mirror_seam_fix). GATE: verdict empty or every remaining line explained, and the closeups agree.r[\"subd_cage\"] quads near 100%, no long rectangles on a cage that will be sculpted, shiny-matcap closeups free of pinches.g, tris = HS.game_version(ob, micro_segments=1) (Triangulate with keep custom normals after WN; stays in place because it shares the cutters); glTF kept the custom normals 100% [verified]. Baked asset: the low poly and bake go to scenario-blender-retopology and scenario-blender-uv-baking. Rig or export: HS.collapse(ob) on a copy (du Mont [01:07:29]). GATE: tri count within budget, triangles only, matcap closeup of the game copy.| Value | \nSetting | \nRelative to / source | \n
|---|---|---|
| Micro bevel | \nangle 30 degrees, 3 segments render, 1 to 2 game, Harden Normals, Miter Outer Arc | \nGambrell [00:17:20], du Mont [00:14:40] | \n
| Micro width | \nabout 3 mm body, 1 mm small parts, 0.1 mm collar | \nreal size, du Mont | \n
| Design bevel | \nLimit Method Weight, 8 to 12 segments | \ndu Mont [00:13:32] | \n
| Big rolling curve | \n46 to 50 segments; cutter edge bevels about 8 to 10 | \nRyuu [00:12:53], [00:05:42] | \n
| Min feature | \nwall, step, groove, facet width > 2 x micro width | \n[added, verified] | \n
| Clamp probe | \ndisplacement over 10% of width = clamping | \n[added] | \n
| Solver | \nEXACT; MANIFOLD for speed with manifold operands; FLOAT only to diagnose | \nverified in 5.2.1 | \n
| Round cutters | \nvertex count set at creation (40 for screw holes); half-segment phase vs target | \nRyuu [00:29:43]; [verified] | \n
| Bevel before Subdivision | \n2 or 4 segments, profile 1.0, Weight, UV Smooth Keep Corners | \nLampel [00:08:06], [00:11:15] | \n
| Creases | \nfine for small details or rounding a coarse cylinder at level 1; 10x tris for primary edges | \nLampel, du Mont [00:52:18] | \n
| SubD cage | \ncircle 18 verts body, 36 on a denser part; square quads | \nrileyb3d [00:01:34], PzThree [00:03:15] | \n
| Mirror seam fix | \nbevel Face Strength Affected + WN Face Influence, weight 100 | \nGambrell [00:06:57] | \n
| Game budget | \n15k tris, one 2K map, animated parts separate | \nAugusto [00:11:48] | \n
| Cost measured | \ncase 1,984 tris (A) vs 4,608 (B, level 2); barrel 7,308 (A), 25,728 (B), 3,716 (A game) | \ntest_03 [added] | \n
In code (all in HS.shading_audit, readable through HS.verdict): scale applied; stack_problems empty; custom normals present when expected; 0 non-manifold edges; 0 flat-shaded faces; 0 non-planar base faces (tolerance 1e-4 x size); no smear (corner normal leaning over 3 degrees on a flat face [added]); no hard edge (over 45 degrees) smoothed across; clamp ratio under 0.1; every boolean changes the mesh; cutter reports clean (coplanar, near-parallel, thin wall, narrow gap, near-miss, rounding facets, pokes out, hygiene). School B: quads %, quad aspect, crease length. Game: HS.tri_count, triangles only.
Visual: HS.review rows matcap (forms), reflect (check_reflection_horizontal: stripes run continuous over bevels and rounds), hsgrey (bevel highlights), wire (real boolean topology); views front, threequarter, top, low. Then one HS.closeup(ob, point, radius, ...) per cut in shiny (metal_carpaint). Look for gradient smears on flat faces, a bevel highlight broken or narrower at a cut, stretch at cut corners on curves, a butterfly on a mirror seam, subD pinches, facets on cut walls. A straight-down shiny view saturates white: use matcap or hsgrey from the top.
| Mistake | \nWhat it looks like | \nFix | \n
|---|---|---|
| Boolean below the micro bevel | \ncut edges razor sharp, rest beveled | \nHS.order_stack(ob) | \n
| Weighted Normal added after pinning the bevel | \nWN lands above the bevel, smears return | \npin bottom-up (order_stack) | \n
| Cutter near but not on a target face or edge | \nwhole-object bevel narrower, thin steps | \nmove clearly away or snap exactly; cutter_report, clamp_culprits | \n
| Groove or wall thinner than 2x bevel | \nclamped bevels everywhere | \nwiden the feature or thin the bevel | \n
| Round cutter in phase with the target facets | \nstreaks, black patches, non-manifold edges | \nhalf-segment phase (default in make_cutter) | \n
| Rounded cutter corners at 30-degree facets | \nmicro bevel catches facet edges, clamps | \nmore segments (beveled_narrow_facet_edges) | \n
| Flat-shaded cutter | \nfacets on the cut walls | \ncutters smooth (make_cutter does it) | \n
| \"Switch to Fast\" | \nnon-manifold result on coplanar faces | \nEXACT; repair the operand instead | \n
| Unapplied scale | \nuneven bevel widths | \nHS.apply_scale | \n
| Stacking normal fixes on a non-planar face | \nstretch that nothing fixes | \nHS.flatten_faces | \n
| Data Transfer placed before WN, or scoped by a vertex group | \nno effect / cut walls shaded like the skin | \nHS.transfer_normals (after WN, distance-scoped) | \n
| du Mont's support cut placed through a round cut | \nrim edges halved, clamp 0 to 1.2 [verified] | \nonly where sliver_report shows boolean slivers; measure before and after | \n
| Creases on primary subD edges | \nsoft, pinched, heavy | \nsupport loops or bevel before subdivision | \n
| Moving a game copy or slice | \ncuts vanish (cutters stay with the original) | \nkeep it in place, or HS.collapse then move | \n
bpy.ops.object.shade_auto_smooth adds a pinned Smooth by Angle node modifier (works headless with temp_override); WN, Harden Normals and Data Transfer need no Auto Smooth.'EXACT' (default), 'MANIFOLD' (4.5), 'FLOAT' (the old Fast, renamed 5.0).bevel_weight_edge / crease_edge (4.0); the Bevel modifier reads any attribute through edge_weight.ob.modifiers.move(i, j) silently refuses moves that involve pinned modifiers; modifier_move_to_index returns CANCELED. Unpin, move, re-pin bottom-up.bmesh.ops.bevel defaults are segments=0, profile=0.0: always pass both.mesh.circularize, mesh.flatten, mesh.space_edge_loops_evenly); Relax and Round Cube are not.references/procedures.md: full tested code for the three parts (panel, case both schools, barrel both schools), every audit case, game routes. Load before scripting.references/critique.md: the rubric and triage table to judge a render and a verdict. Load at every shading pass.references/expert-notes.md: the experts' principles, disagreements and deciding conditions, with timestamps. Load when a choice is not covered here.references/sources.md: the nine videos, who, what each is best for, best timestamps.