Skip to content

Timing Verification

The T1 Timing Harness is not a training feature. It is how you verify that the instrument is still measuring what it claims to measure.

  • On a new iPad, before any real session
  • After an iPadOS update
  • Whenever a threshold looks wrong
  • If a session reports many timing-invalid trials

Run on the physical iPad, unplugged from any computer, at normal screen brightness, with no other apps active. A device connected to a debugger does not perform like a device in the field.

Verifies that the screen blacks out when commanded.

  1. Tap T1 Timing Harness from the home screen
  2. Select T1 · Timing
  3. Set trials to 200
  4. Tap Run

The screen shows the device model, the real refresh rate, and the clip duration while it runs.

MetricTarget
Median |deviation|≤ 8 ms
p95 |deviation|≤ 17 ms
Any deviation > 34 ms0

A green PASS banner means the instrument is sound. A red FAIL means thresholds measured on this device are not trustworthy — stop and investigate.

Reference measurement on the target iPad (iPad15,7, 60 Hz): median 3.4 ms, p95 5.3 ms, zero deviations over 34 ms.

Verifies the touch-timing path.

  1. Select T2 · Response
  2. Set trials to 50
  3. Tap Run
  4. Tap any zone the instant the screen goes black. Which zone does not matter — this measures the response path, not anticipation
CheckTarget
Negative RT0 — a negative value means a clock defect
RT over 1000 ms0
Median RTPlausibly 200–350 ms

Three outcomes: PASS, Invariants OK but RT range implausible (amber — the clock is sound but the operator wasn’t tapping consistently), or INVARIANT VIOLATED (dark red — a real defect).

RowMeaning
Median |deviation|Typical blackout error. Red if > 8 ms
p95 |deviation|Worst-case in normal operation. Red if > 17 ms
Max |deviation|Single worst trial
Mean (signed, = bias)Systematic offset. Correctable
SDScatter. Not correctable
Over 34 msHard failures. Must be 0
Invalid (>2 refreshes)Trials that would be excluded from a real threshold

Read bias and scatter separately. A consistent 5 ms late bias is a different problem from a 5 ms standard deviation, and only one of them can be tuned away.

Tap Export report. This writes a CSV plus a Markdown report to the app’s Documents folder and opens the iOS share sheet.

  • CSV — one row per trial with every measured value. Timeouts are left blank, not zero, so a spreadsheet cannot confuse them with an instant tap
  • Markdown — metadata, an acceptance table with per-metric PASS/FAIL, a distribution table, a histogram of signed deviation, a per-offset breakdown, and display-link health