Results · SIMULATED data · generated 2026-09-06T08:59:43+00:00

Results

Everything on these pages was computed from synthetic participants produced by the simulator described in the methods. It demonstrates that storage, censoring, estimators and figures behave correctly; it says nothing about human vision yet.

60participants

2469main trials recorded

1674in the primary analysis set

37.8%detection (change trials)

3.4%false alarms on catch trials

62.2%censored

01 Exclusions

Flagged, never deleted

Every rule is pre-specified; every flagged record stays in the raw data and the sensitivity analysis re-fits the primary model on all trials.

Trial-level flags (main phase, n = 2469)
Flag	Rule	Trials
timing_invalid	dropped-frame share > 5%, gap > 250 ms, or tab hidden	182
too_fast	response < 200 ms after onset	2
premature	response before change onset	7
mislocalized	click > 1.5× target radius away	97
no_change_trial	catch trial (false-alarm analysis only)	353

Participant-level flags (n = 60)
Flag	Rule	Participants
incomplete	< 80% of 42 trials	2
high_false_alarm	> 50% of catch trials pressed	0
inattentive	< 20% detection on largest-magnitude fastest trials	2
repeat_suspected	device signature seen before	3
touch_device	coarse pointer or viewport < 900 px	2
withdrawn	withdrawal requested	0

02 Headline models

Speed, magnitude and place

Forest plot of Cox hazard ratios — **Hazard ratios for detection.** *Cox proportional-hazards model, robust SEs clustered by participant; grey = not significant at 0.05. Concordance 0.838.*

Detection

Detection probability by magnitude, speed, type, location, clutter; mixed-effects logistic model.

Detection probability →

Latency

Survival curves, Cox with time-varying magnitude, accelerated-failure-time models, PH diagnostics.

Latency →

Features

Nested feature sets, grouped cross-validation, permutation importance, calibration.

Feature modelling →

03 Hypotheses

Pre-specified tests

H	Verdict	Evidence
H1	supported	Cox HR per doubling of duration = 0.226 (p = 4.4e-54); log-rank across durations p = 1.4e-62.
H2	supported	Magnitude-at-detection AFT: +100 u eccentricity multiplies threshold magnitude by 1.131 (p = 2.5e-12).
H3	supported	Cox HR per SD of spectral-residual saliency = 1.359 (p = 1.1e-11).
H4	supported	Cox HR for semantic importance (0→1) = 2.025 (p = 8.2e-13); LRT vs low-level-only logistic: χ²(1) = 16.525, p = 4.8e-05. Out-of-sample: see prediction.json.
H5	supported	Cox HR for high vs low clutter = 0.573 (p = 1.3e-10); threshold ratio = 1.240 (p = 2.2e-06).

Semantic LRT (H4, in-sample): χ²(1) = 16.525, p = p = 4.8e-5; AIC 1568.26 → 1553.73.