Eve library

Phase 170 locally actionable completion evidence

Tracker

Revision 1

On this page

Historical task source. The Eve task board holds current task state.

Date: 2026-07-15

Completion audit#

Phase 170 defines sculpting, topology, layers, brushes, face sets, masks, polypaint, retopology, export, line rendering, cel shading, inking, and validation requirements across thirteen focused Rust workspaces under libs/neith.

The local audit corrected six validation claims that were previously weaker than their checklist text:

  • the 50M-poly gate now executes the production deformation kernel over tiled meshes containing at least 50 million real triangles; it does not extrapolate from a 90K-vertex sample;
  • face-set persistence now saves, clears, reopens, and reads a real Blender .blend binary rather than treating a JSON roundtrip as Blender evidence;
  • the multires bake is checked against both its analytic surface and a separate Blender Catmull-Clark evaluation;
  • cel validation now computes pixel RMSE against a hand-authored golden value study instead of only counting quantisation bands;
  • dogfood binds the real Project Obsidian source package and performs actual Phase-141 X-sheet and Phase-156 Grease Pencil handoffs;
  • ZTL compatibility now has a versioned executable adapter protocol that passes the untouched binary over stdin and rejects malformed, version-mismatched, or topologically invalid geometry responses.

The export crate now exposes a binary-safe export_alembic API. It writes a real Ogawa archive when supplied a Blender build containing the Alembic writer, and distinguishes a missing compiled operator from a successful export. The text API refuses to misrepresent binary Alembic bytes as UTF-8.

Requirement-to-evidence matrix#

Requirements Authoritative package Passing tests
170.1.1–.1.15 sculpt-core 80
170.2.1–.2.15 sculpt-topology 47
170.3.1–.3.10 sculpt-layers 37
170.4.1.1–.4.3.10 sculpt-brushes 84
170.5.1–.5.10 sculpt-face-sets 24
170.6.1–.6.10 sculpt-mask 30
170.7.1–.7.10 sculpt-polypaint 18
170.8.1–.8.10 sculpt-retopo 26
170.9.1–.9.10 sculpt-export 37
170.10.1–.10.20 npr-line 36
170.11.1–.11.15 npr-cel 23
170.12.1–.12.10 npr-inking 13
170.13.1–.13.10 sculpt-testing 26

The audited phase contains 481 passing Rust tests with zero failures and zero ignored cases. Every package passes canonical rustfmt, strict all-target clippy, all-target tests, and an all-target release build. Two stale Rust 1.95 clippy violations discovered during the gate were fixed in sculpt-face-sets and npr-line.

Open external acceptance gates#

The repository-local contracts and negative paths are complete. Two live interoperability captures still depend on software/assets not present in this checkout and cannot be replaced with synthetic evidence.

Requirement Locally complete contract External evidence still required
170.9.2 export_alembic creates an isolated OBJ handoff, invokes an explicit Blender binary, requires a produced .abc, reads the binary bytes, and distinguishes launch failure, omitted Alembic builds, backend failure, and I/O failure. Text formats and binary-API redirection are covered locally. Run the passing Ogawa-signature path with a Blender build compiled with WM_OT_alembic_export. Ubuntu's installed Blender 4.0.2 omits that operator.
170.13.5 import_ztl_with_adapter defines the neith-ztl-v1 stdin/JSON protocol, preserves original bytes, validates protocol version and mesh topology, and has positive plus fail-closed adapter tests. Generic zlib streams are no longer falsely identified as ZTL. A licensed production ZBrush/GoZ adapter and redistributable representative .ztl corpus. The format is proprietary and no licensed fixture/backend is present.

Verification#

For all thirteen libs/neith/{sculpt-*,npr-*}/Cargo.toml phase manifests, the following commands passed sequentially:

bash
cargo fmt --manifest-path <manifest> -- --check
cargo clippy --manifest-path <manifest> --all-targets -- -D warnings
cargo test --manifest-path <manifest> --all-targets
cargo build --release --manifest-path <manifest> --all-targets

Verification was serialized and supervised on a 15 GiB host with 31 GiB swap. Available memory remained near 12 GiB while an unrelated BFF development watcher was active. No full-workspace Nx command was run. No browser, mobile, or customer-facing UI changed, so Playwright and mobile automation are not applicable.