Public reviewed engineering · October 27, 2026 (UTC)
A16 ear mesh quality improves; original Evelyn head is still visually rejected
The A16 results are executed bounded software and geometry tests, not human-form or Evelyn-likeness acceptance. Private reference photos, recipes, original custody and rejected renders are not published.
What actually ran
One fresh ordinary production attempt passed 31 head software checks and three ear checks, saved 12 head inspection images and an editable original one-mesh scene with 16,385 vertices and 16,320 faces. Saved projected and Blender arrays matched. The 128-point neck boundary and semantic attachment contract remain unchanged, and actual torso registration is unbound.
Ear quality measurement
The localized ear correction changed only medial-back interior depth. The old small-case worst triangle aspect was 333.94948; revised small cases scored 26.48408 and 37.33067, with larger cases about 26.08093, all below the unchanged threshold 60. The scoped contact audit found no nonadjacent contacts under declared shared-index exclusions. This is not proof of realistic ear appearance or global mesh validity.
Visual review and next step
Primary and independent reviewers inspected all 12 actual views and withheld acceptance: under-chin/front-neck geometry remains too abrupt, and mandibular angle, facial depth and whole-head proportions need reference comparison. No eyes, eyelids, ears, nose, lips, skin, hair, body registration, movement, speech or physiology was tested as fitted head components. The earlier standalone original eyeball acceptance remains separate.
The next proposed research corrects submental/jaw/upper-neck geometry while retaining neck-boundary constraints; it has not yet been reported as an executed success.