Next Gen STEP AP242 Native-ish BREP Geometric Modeling Kernel
Sheet Metal M8 — CTC-03 intent recovery
Verdict: Meaningful progression
M8 removes the previous missing-opening blocker. The canonical NIST CTC-03 STEP now recognizes 15 holes, two slots, seven bends, and 15 planar regions. The source-independent Firmament program regenerates the same opening inventory and all 17 opening comparisons pass. It also models the partial-span service flange and its 101.6 x 12.7 mm outer tab.
Profile-M2 now reconstructs the complex central front and rear mounting-flange free edges through generic semantic edge attachment. It is still not a full CTC-03 reconstruction: endpoint corner ownership, left-wall tapered ends, and right-wall attachment cutbacks remain simplified. The comparison remains NeedsReview, but source-to-generated formed RMS improves from 19.4627 to 10.6140 mm, p95 from 52.4846 to 19.0713 mm, and flat height residual from 12.7078 to 0.00782 mm. The isolated next blocker is cross-edge corner composition, not single-edge fragment placement or hole recognition.
Delivered mechanism
The authored compiler now has a bounded semantic-layout pass between parsing and exact profile lowering:
Concept / Concept Struct
-> Datum, Pattern, Tab, Require
-> resolved SheetMetalSemanticLayout
-> stable generated cut paths + validated constraints
-> exact formed profiles and PlanarContour2 flat blank
Datum names local workpoints. Pattern owns size, count, pitch, and generated member identities. Tab extends a bounded partial flange edge. Require validates required members and explicit equal-size, equal-pitch, or mirror claims. Pattern resolution also records equal-size and equal-pitch constraints automatically. The CLI exposes the resolved layout under sheetMetal.semanticLayout and reports semanticResolve timing.
The M8 CTC program resolves three datums, nine patterns, one tab, 17 generated cuts, and 18 admitted pattern constraints. Stable paths include Ctc03Layout.BaseFastenerPattern[0] and Ctc03Layout.ServiceInnerHoles[1].
Evidence map
- Final source
- Feature and missing-feature inventory
- Semantic layout and real syntax
- Formed and flat comparison
- PMI and datum audit
- Iteration and LLM friction
- Validation, performance, hashes, and code-quality verdict
- Independent formed STEP
- Independent flat STEP
- Independent flat SVG
The non-CTC generalization fixture is m8-semantic-panel.firmament.
Direct architecture answers
Can Aetheris fully reconstruct CTC-03 as the actual manufactured part? No. It now reconstructs the complete recognized bend/opening skeleton and the distinctive service tab, but not every manufactured edge trim.
Did the semantic sketch reduce reconstruction mistakes? Yes. Named datums and regular patterns replaced repeated point edits; contradictions are rejected before topology construction; and the final feature comparison reached 17/17 pass without post-lowering feature matching hacks.
Did it generalize? Yes, within its bounded scope. The independent semantic-panel fixture uses the same datum/pattern lowering and stable paths, and tests prove deterministic output.
Largest remaining blocker? Bounded corner programs that may consume both an owning edge endpoint and the adjacent edge endpoint while preserving one explicit corner owner. Single-edge fragments now compose deterministically and retain correspondence; the remaining CTC wall tapers and mounting-flange end chamfers cross that seam.