← Development updates · NewBrain · Avatar Builder

Reviewed November 5, 2026 · Research dates distinguished below

Completed experiments, measured limits and preserved failures

No newly accepted model, Evelyn head, complete body or functioning organ system. These are reviewed public-safe aggregates from committed status records, not a fresh independent rerun of private research. Private lessons, model states, reference photographs and videos, raw responses, rejected renders and unpublished source files are not reproduced.

NewBrain: four-condition V3 study closed with negative learning results

The controlled pathway V3 experiment, still active in the November 4 review, is now recorded as closed and failed. Each of its four separate 15-response conditions scored 1/15, with the credited answer limited to an unsupported-question control. Correct evidence selection scored 0/10. These are this experiment's denominators—not Original12 or evidence of factual learning.

80/80 outputs ended with EOS and 24/24 post-restart response pairs matched exactly. Execution completion and repetition of answers, including errors, do not imply correct learning, generalization or conversational ability. A separate contextual-practice study has a recorded completed native execution and saved state, but its complete scored outcome has not been qualified for this public report.

The earlier V9 native qualification remains negative: taught facts 0/6, rules 0/2, paraphrases 0/6, compositions 0/3 and correct abstentions 1/3, with 20/20 repeatable post-restart answers that included mistakes. Accepted Original12 remains 2/12, not rerun.

Avatar Builder: two real head renders rejected; body V9 failed

Original Evelyn V12 and a separate commissure V2 revision each produced 17 saved review views. Visual review rejected both for pronounced cheek/mouth folds, mismatched jaw/skull/neck and oval ears. The strict native-form check also failed. The previously approved blue eye geometry and materials were preserved, but this component is not an approved head.

The actual original full-body V9 run failed its unchanged work guard after four coarse regions. Partial saved metadata are not a finished mesh or export. Later V10 source/environment reviews and a shared-cheek diagnostic are not successful body construction or a usable head. No accepted separate fictional-character reconstruction, digestive/urinary/reproductive route, functioning organ, completed movement rig or whole-body export exists.

Separate October 31 partial torso/arms experiment

An October 31, 2026 regional shoulder study, whose factual status was later mirrored, records a successful third native attempt after two earlier failures. Static GLB reimport passed a strict one-mesh check at 16,244 vertices and 32,144 triangles. Original source coordinates and a 128-vertex neck connection were retained. Three limited native poses were measured; the combined pose's maximum edge-length ratio changed from 1.8852 to 1.1938 under revised spatial weights.

This is an original partial torso and arms without accepted head, hands, legs or internal physiology. Its torso silhouette and overall deformation remain unaccepted. No exported animation, complete anatomical body or clinical function was qualified. Mirrored status records describe one experiment, not independent replications.

Verification and publication boundary

This public summary distinguishes completed native runs, rejected outcomes and source-only successors. It does not grant public access to private repositories, reference materials or model snapshots and does not independently replicate all measured studies. The November 4 review remains unchanged for historical comparison.

Ask Iris about reviewed research → · Development timeline →