# Sweep29: independent front single-sweep reference

The first Wan3 trial is selected as a front-view motion reference candidate, with limitations.
It delivers one readable sweep and return while retaining one broom, the same holding hand,
recognizable face and stable visible support. Owner visual acceptance remains pending.
No native rig or adopted motion registry was changed.

## What changed in the method

This trial starts from an exact front image instead of asking a video editor to change the camera
of the original oblique performance. The same image is supplied at the end. The request is reduced
to one inward floor sweep and one relaxed return in four seconds. No reference video conditions
the model, so agreement with the oblique source's exact timing or hidden pose is not claimed.

The image is copied without redrawing, cropping or recoloring from Sweep25's hold-front.png.
It fixes a neutral front camera, the viewer-right holding paw, one broom and the feet's starting
placement. Both exact input files are available in the viewer. They remain honestly UNREGISTERED;
the owner authorized this experiment under the current platform spend policy. They do not imply
acceptance of an earlier native performance.

## Original and review

Provider: Runway Wan3. Requested four seconds, 720p, 1:1. Actual original: 960x960, 30fps, 120 frames, 4s.
Provider task: 6065c921-bdd9-41a7-828d-a86ff3afcced.
Original SHA256: f47916172a0d827d3e68527cfdd167d2dea7b3a357f5cca19f0ae804b861cd6a.
Input SHA256: 9530a9397383316a5be923194e683399f28751f0e229c73f5cd3c83ea1b136b4.
No trimming, retiming, frame interpolation, reversal or other media edit was performed.

All 120 frames were inspected in four chronological sheets. Native-size frames 0, 45, 60, 78 and seven
feature-mask keys were enlarged. Actual playback at 1x, 0.5x and 0.25x, loop, end and frame stepping
were checked in the browser. The scoped decision is pass-with-limitations, by the main agent only.
No independent reviewer or subagent was used.

## Observed performance

- Approximately 0–1.3s: neutral hold.
- Approximately 1.3–1.5s: small preparation.
- Approximately 1.5–2.1s: one inward sweep ahead of the feet, with modest head/torso inclination
  and turning, supported by the same holding arm.
- Approximately 2.2–2.8s: a relaxed return toward the original pose.
- Final interval: quiet neutral settle through 4s.

The camera stays fixed in front while the character itself turns slightly during work. Both eyes
remain readable. No second broom, hand switch, extra limb, cut or abrupt attachment loss was found.
The broom shaft remains related to the same hand; its upper section is progressively occluded
by the arm/body during the inward pass. The visible face markings and neutral expression remain
recognizable. There is no observed stepping gesture.

The all-frame color diagnostic finds one qualifying green binding component in every frame;
seven mask images confirm that identity at the inspected keys. The exposed lower dark left edge
ranges 1 pixel and the bottom ranges 1 pixel at native 960x960. These support visible support stability,
not an anatomical or 3D contact certificate. The rightmost dark bounding box is excluded from paw
motion inference because occlusion and other dark regions change it. Full measurements retain
this failed marker rather than presenting it as a foot trajectory.

## Limits and intended reuse

This is an independent front performance, not a synchronized alternate view of the original oblique
movie. Use it to consider a readable front sweep, holding-arm coordination and the return gesture.
It does not uniquely determine ankle, hip, pelvis, chest or neck angles, contact force or balance.

A brown tail silhouette appears as the body turns. That newly revealed shape is not independently
validated by the front input, and must not replace the existing tail shape or pigment authority.
Use the existing native/reference appearance if this performance is later interpreted in a rig.

The approximately 1.3s initial hold remains in the provider original. It can be judged as a pacing
preference, but was not removed to make the source look more active. Identical endpoint inputs
do not prove an exact pixel/velocity loop. The source's 30fps is not a 60fps real-time rig deliverable.

## Cost and preservation

One paid request, estimated $0.40 before submission. The response records 40 credits; actual billed
USD is not returned. No second request or alternative-provider charge was made for this study.
The original video, prompt, request/response provenance and all 120 native decoded frames are kept.
Sweep25, earlier generated sources and adopted releases are preserved. No mechanical fitting,
new native controls or motion adoption is included in this trial.