October 20, 2026 · Actual captured frames · Independently reviewed
Evelyn: eyes close and reopen
A fresh controlled graphical-motion test passed the fixed eyelid-closure and restoration criteria. Unlike the earlier failed test, this run shows actual recorded open, blink-peak and reopened eye views and retains its negative controls and independent evidence trail.
Original recorded engineering frames
What the fixed test verified
First accepted common eyelid setting. Smaller settings did not pass.
Measured eyelid reopening error. Sampled geometry and pose returned exactly to neutral after the test.
Changed eyelid-control pixels across the recorded frames.
Changed mouth-control pixels. This is graphical control, not verified speech synchronization.
How the evidence was independently checked
Owned run 1047 completed the blink, jaw-control, playback-clock and restoration checks. A separate saved-evidence review, 1048, recomputed the closure and pose measurements; checked all 15 PNG captures against their complete recorded browser responses; and verified process ownership, closed resources, empty process groups, full output streams and available I/O bounds. The accepted run was not replayed to replace a failure.
The precise results reported by the source are 69.8719% closure, left and 72.4625%, right. The original 65% acceptance requirement was not changed to obtain the passing result.
What still needs to work
Evelyn's generated voice has not been demonstrated. The loud noise heard during development was a controlled engineering test tone, not her voice. Later calibration should be silent.
Actual generated conversation, natural speech-to-mouth/phoneme alignment, memory persistence across fresh restarts, a verified user installer, the original high-fidelity appearance and public-release readiness remain separate milestones. Passing controlled movement does not validate human physiology or establish a live NewBrain connection.
The developer intends private installation and repeated personal user testing before a free public release. This report does not offer an installer.
Previous failure is preserved
The earlier motion study, run 1042, measured only 1.68175% closure on the left and 1.74329% on the right, against the same 65% requirement. That failed result remains in the research archive; the new accepted run is a separate experiment and does not erase the failed original.
Read preserved earlier failed test → · Read earlier three-angle static head test →
Attribution: base character listed as Base Female Character by BlackProject, labeled CC BY 4.0 in the local source record. Attribution and redistribution permissions require final release review. This page presents limited research screenshots, not original character assets.