← Avatar Builder status

Research evidence · October 28, 2026 · Original independent head geometry

Why the latest head correction still failed visual review

Codex saved an actual bounded geometry correction with real Blender outputs. The correction improved some lower-jaw support but did not alter measured front-to-back depth. The owner, primary and independent reviewers rejected the appearance. This document reports the completed experiment, not a new public head model.

65Software checks passed in stated scope
12Actual inspected neutral renders
6,324Geometry vertices moved laterally

Actual measurements

MeasurementSaved reviewed result
Original scene geometry16,385 vertices / 16,320 faces
Editable Blender files3
Head neck-boundary correspondence128/128 vertex positions unchanged
Vertices affected by lateral change6,324
Maximum displacement6.944 mm
Measured front/back depth change0 across all nine tested profile sections
Visual accept/rejectREJECTED by owner, primary and independent reviewers
Female torso runtime calls7 planned calls unrun

The final explicitly projected neutral geometry round-tripped exactly in the saved evidence, but an intermediate raw symmetry failure remains documented. A finite, successful geometry audit does not establish that the facial anatomy or proportions are realistic.

The visible defects remain

The head still has a bulbous cranium, elongated and insufficiently deep facial plane, unclear mandibular profile and a visibly unnatural submental/skull-to-neck transition. A predominantly lateral widening adjustment cannot fix a front-to-back depth deficit. The next candidate should use the original B-spline shape controls to adjust the cranium, forehead and midface depth and mandibular angle, with fixed-camera front and profile references.

The neck attachment metadata and 128 original boundary positions were preserved. Seven female-torso calls were not run. No accepted individual head, fitted eye, mouth, expression rig, original body, organ simulation, motion or medical anatomy is demonstrated. The previously accepted standalone eyeball component remains separate from this failed full-head fitting exercise.

Research boundary

Private identity recipes, raw renders, source photos and desktop records stay private. The 25,214-frame desktop custody capture is not a public avatar video or proof of rendered animation. No copied ready-made head mesh was substituted. The saved report and measurements are an internal reviewed source, not an externally replicated scientific publication.

Source: KiraWorld retained review, docs/research/original-shapes22-2026-10-28/verified-status.txt. The mirrored copy in Patient Horizon represents the same experiment, not independent clinical validation.