TrimSrfCache::get_tree_with_face is SmSurfaceCache::GetTree with the virtual
BuildTree dispatched to build_tree_with_face. Hcr::is_point_visible builds a
face-owned cache through its face and runs the tracer's three GetTree calls
(AddSurfaceToGrid twice, ExpandHierarchy once) with the face, in upstream
order, before the tracer borrows the surfaces; the tracer then reads that tree
as is (RayTracer::add_surface_with_fetched_cache,
Grid::expand_hierarchy_fetched). blend.hcr: 434/434 pass, 0 deferred.
The native-face trim-cache probe gains build modes that call GetTree (after an
editing-on build, with editing turned off first, on a trimmed tree, and with no
tree), with 7 new fixture cases.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEtgaA54W2nr8dK1EYp
Local checkpoint only: user chose to wait for #537 to merge, not stack or continue duplicate shared preparation. Reconcile against its CacheFace/BrepCacheFace mechanism on fresh main before further implementation or readiness.
Verified local native checkpoint: creator 898 cases / 717 passed / 181 deferred / zero failed, all original 713 passes retained; trim cache 628 / 568 passed / 60 deferred / zero failed. All seven new native boundary probes compare values with zero relative error and no floors; six selected native units have zero failures. Executive reports remain byte-identical to baseline.
Earlier kernel run: 1904 tests passed. Formatting and whitespace checks pass. Full final workspace Clippy/tests, wasm, native green gate, sensitivity/fuzz and post-537 verification remain outstanding. PR stays draft; remote claim-only CI is not implementation validation.
Co-Authored-By: GPT-6.1 Sol Fast <noreply@openai.com>
Brett Lawson