What Xianji Measures

Xianji (先机) is a measurement and training instrument for visual anticipation in fast interceptive sports. Badminton singles is supported today; tennis serve return is planned.

The problem it addresses

At a professional badminton smash or a fast tennis serve, ball flight time is shorter than the receiver’s own movement initiation. The receiver cannot be reacting to ball flight. The information they use arrives before contact — from the opponent’s trunk rotation, shoulder line, elbow height, and racket face during the swing.

Expert–novice differences in this ability are among the more robust findings in the perceptual-motor literature. The standard laboratory method for isolating it is temporal occlusion: show the action, cut the display at a controlled moment relative to contact, and force a prediction. If the display is cut before contact and the athlete still predicts above chance, they are demonstrably reading pre-contact kinematics.

What it reports

The anticipation threshold — the occlusion offset, in milliseconds before racket–shuttle contact, at which the athlete remains about 79% accurate.

A threshold of 120 ms means: cut the vision 120 ms before contact, and this athlete still predicts direction correctly roughly four times in five.

Larger is better. It means the athlete needs less information, earlier.

The result is a single number with a stated standard error, on a physical scale, that can be tracked across a season and compared between athletes within the same sport.

Why the timing engineering is unusual

The number is only as good as the instant the screen goes black. If the blackout lands 40 ms late, every threshold is displaced by 40 ms and the instrument produces fiction that looks like data.

Xianji therefore does not command an occlusion and assume it happened. Every trial measures the actual blackout instant and the actual touch instant from the same hardware clock and stores both. Reaction time is measured_response − measured_blackout, never the intended value.

Measured on the target iPad over 200 trials:

	Target	Measured
Median blackout error	≤ 8 ms	3.4 ms
95th percentile	≤ 17 ms	5.3 ms
Errors over 34 ms	0	0

One screen refresh at 60 Hz is 16.7 ms. The instrument is accurate to well under a single refresh. Trials that miss by more than two refreshes are flagged, excluded from the threshold automatically, and counted on the results screen.

The two subsystems

Part	Runs on	Does
Pipeline	Desktop (Mac / Linux)	Marks contact frames, cuts clips, writes `manifest.json`
Trainer app	iPad	Imports the manifest, runs trials, measures timing, reports the threshold

The app performs no video analysis. It consumes prepared clips. This boundary is deliberate — it is what allows the trial loop to be timing-accurate.

Video never leaves the iPad. The app contains no network code.