Session Protocol

Standardise this and follow it. Without a fixed procedure, your comparisons carry the variance of your testing conditions rather than the athlete’s ability.

Fixed conditions

Factor	Requirement
Device	Same iPad every time. Log the model
Brightness	Same setting, normal range
Viewing distance	Same, roughly arm’s length
Time of day	Same where possible
Fatigue state	Not immediately post-exercise
Trial count	60 for anything recorded
Footage set	Same set for any comparison
Instructions	Same script, read aloud, every time

Establishing a baseline

1. Familiarisation run

A full 60-trial session that you discard.

First exposure measures learning the task, not anticipation. Skipping this makes every athlete’s first recorded session an underestimate, and their second look like improvement when nothing changed.

2. Two baseline sessions within 72 hours

Run two full sessions and compare them.

If the two thresholds agree within 2 × the standard error, the athlete is stable and single sessions can be trusted going forward.

If they do not agree, do not yet trust individual results. Either the athlete’s performance is not yet stable, or something in your procedure is varying. Run a third before concluding anything.

This is the test–retest reliability of your own instrument in your own hands. It cannot be inherited from someone else’s setup.

Retesting

Retest every 4–6 weeks unless studying a specific short-term intervention.

Anticipation is a perceptual-cognitive skill; it does not move week to week under normal training. Testing more often mostly measures noise, and repeated exposure to the same footage introduces familiarity with specific clips rather than general skill.

Recording a session

There is no session history in the app. Record at minimum:

Date            ____________________
Athlete         ____________________
Footage set     ____________________
Device          ____________________
Trials          ______

Threshold       ________ ms  ± ________ ms
   or  DID NOT CONVERGE — reason: _______________

Accuracy (corrected)  ________ %
No-response rate      ________ %
Median RT             ________ ms
Reversals             ________
Timing-invalid        ________

Interpreting change

A change is real only if it exceeds roughly 2 × the standard error.

If a baseline is 133 ± 12 ms and a retest is 141 ± 14 ms, that is noise. If the retest is 175 ± 13 ms, that is a change worth investigating.

Never compare:

Across sports — different chance levels
Across viewpoints — different task
Across footage sets — different stimulus

For research use

If you are writing this up, the method is:

A temporal-occlusion paradigm with a 3-down-1-up adaptive staircase converging on 79.4% correct. Threshold estimated as the mean of the final six reversals; standard error as SD/√6. Sessions with fewer than six reversals or SE > 25 ms were rejected as non-convergent. Stimulus timing was hardware-verified (median blackout error 3.4 ms, p95 5.3 ms at 60 Hz); trials deviating by more than two display refreshes were flagged and excluded from threshold estimation. Response latencies were captured from the platform touch event in the same monotonic clock base as the measured blackout.

Every raw trial is stored: intended offset, measured offset, deviation, both timestamps, response zone, correctness, and timeout flag. Retrieving it currently requires direct access to the device database — see Current Limitations.