Experiment 001 / See the signal

Landmark Explorer

See all 21 hand points, confidence, timing, and coordinate changes before an effect hides uncertainty.

Browser asks camera permission only after “Start camera.” Frames stay in this tab and tracks stop when tab becomes inactive.

Choose image Sequence

Overlay Mirror display Inspector

Signal inspector

No hand selected

Hand Landmark

Input normalized: -
Display pixels: -
Depth estimate: -
Frame age: -
Palm scale: -
Pinch / palm: -
Index PIP angle: -
Wrist velocity: -

Confidence is classification evidence, not truth. Depth is relative, not physical distance.

- input - display - inference - median / worst 0 long tasks

Ready. Camera permission has not been requested.

Open record ↓

Continue research spine Experiment record / 001

Experiment record / 001

Signal before effect.

Phase: Observe the signal
Study status: Physical validation pending
Decision: Revise

Question

What does the tracker actually see?

Hypothesis

Exposing normalized coordinates, confidence, timing, and palm scale before any effect will make tracking limits legible.

Observed

Tracker coordinates remain normalized until display.
Confidence visibly changes line weight.
Deterministic jitter, dropout, crossing, and motion remain repeatable.
Camera tracks and temporary image URLs are disposed on teardown.

Measured

Fixture input: 1280 × 720
Display observed: 120 Hz
Blank-image inference: 65.9 ms

Single-machine reference only; not a sustained camera claim.

Not yet proven

Physical camera performance, hand identity through crossings, varied lighting, and real-device behavior still require study.

Decision / revise

Keep the signal architecture. Expand physical evidence.

The explorer explains uncertainty without narration. Retain the fixture library and normalized geometry; do not treat current measurements as device coverage.