What the studies have measured, as they measure it
Every study on this site publishes its data as they come in, anonymised, whether or not they say anything
flattering. The studies are designed and run by Dr. Satwant Kumar, a physician and vision neuroscientist.
This page is regenerated from the study database; the numbers below are the numbers the studies hold.
Sessions are labelled S01, S02, … in order of upload, and nothing else about a participant appears.
Study 1Trap occlusion
How early can you call the line?
A clay leaves the trench at a small angle left or right of straightaway, is visible for a fixed
few tens of milliseconds, and is then hidden. The observer says left or right. An adaptive procedure finds, at
each viewing time, the smallest launch angle the observer calls correctly three times in four. That angle is
the threshold; smaller is better. 9 sessions from 9 people are shown:
5 elite or international shooters, 2 club shooters, 1 novice or non-shooter and 1 who did not say, 2026-08-05 to 2026-09-15.
2,862
calibrated trials, 6% rejected for timing
0.09°
smallest launch angle called so far
29 of 37
thresholds measurable; 8 below the task's floor
9
sessions shown, of 11 uploaded, from 9 people
The longer the target is seen, the smaller the angle that can be called
One line per session, coloured by self-reported experience. Both axes are logarithmic. Hover or
tap a point for its confidence interval and trial count.
Every session falls with viewing time, as it should if the task measures what it claims to: more
visible flight is more information about the line. Hollow points sit at the floor of what the task can
present (about 0.1°), so those true values are lower still.
Experience shows in the numbers
The same thresholds grouped by experience at each viewing time. One dot per session; the bar is
the group median of the measurable thresholds.
Group is self-reported at sign-in. For the sessions run in person the investigator has confirmed the
labels. With this few people the medians are indicative, not a finding; the study's stated gate for any
claim about groups is a larger sample and a repeat session per person.
Medians by group and viewing time
Threshold in degrees of launch angle, with the sideways drift it amounts to at the distance where
the target disappears. A clay is 11 cm across.
Visible for
Distance at occlusion
Elite / international
Club / developing
Novice / non-shooter
Not stated
33 ms
17 m
0.38° ≈ 11 cm n = 5
0.60° ≈ 18 cm n = 2
—
—
50 ms
17 m
0.19° ≈ 6 cm n = 4(+1 at floor)
0.35° ≈ 11 cm n = 2
—
—
67 ms
18 m
0.13° ≈ 4 cm n = 3(+2 at floor)
0.29° ≈ 9 cm n = 2
—
—
100 ms
19 m
0.13° ≈ 4 cm n = 2(+2 at floor)
—
—
—
What the numbers mean on the range
Take the 50 ms condition. When the target disappears it is about 17 m from the shooter, and the median threshold of 0.23° is a sideways drift of about 7 cm from the straightaway line, less than one clay width (11 cm). That is what "calling the line" resolves to: a drift of a few centimetres, at that distance, in 50 ms.
What one session looks like underneath: S04, elite / international
The psychometric function behind each threshold at 33 ms, 50 ms, 67 ms: the proportion of correct
calls against launch angle, the maximum-likelihood fit, and where it crosses the 75% criterion.
The adaptive procedure places most trials where the answer is least certain, which is why the large
points cluster around the threshold and the far ends are sparse. Whiskers are 95% intervals on each bin. A
dashed curve means the fit landed below the smallest angle shown and is reported as "at the floor".
Why the measurement can be trusted, and where it cannot yet
The screen is calibrated, not assumed. Scale from a standard plastic card (driving-licence size), viewing distance by the
blind-spot method, refresh rate measured before the first trial. Every angle is in degrees at the eye.
Every stimulus is a whole number of frames, and a trial in which a frame arrived late is
rejected rather than analysed with the wrong duration: 172 of 2,862 main trials so far
(6%). One session (S01) lost 155 of its 318 trials to frames delivered late by a display running slower than it had measured; those trials are excluded and its thresholds rest on the rest, which is why its intervals are wide. The study now stops a session early when that starts to happen.
Attention is checked. Easy probe trials are interleaved throughout; a lapse there means the
observer was not looking. Across every session shown, 0 of 162 probes were
missed.
A number the task cannot measure is not reported as a number. When the fit lands below the
smallest angle presented, the result is "below that angle", shown hollow. 8 of 37 fitted
thresholds are in that state; they are the best shooters at the longer viewing times.
A display that cannot present the durations is refused. 2 sessions ran on a display that measured 30 Hz, below the 60 Hz the durations need: there the 50 ms condition was presented as 67 ms, so two conditions collapsed into one. They are left out of everything on this page, and the study now refuses such a display before the first trial.
What this is not. A handful of people, self-reported experience, one session each, and no
peer review. It is an interim look at a running study, not a finding, and nothing here is a claim that any
training or equipment improves anything.
What the first sessions changed
The study runs on a stated protocol and the protocol is versioned. The sessions above led to
protocol 4.0.0: the 100 ms viewing time was dropped, because for experienced shooters it mostly sat below the
task's floor and taught nothing; the three remaining durations are exact multiples of a 60 Hz frame; a display
under 60 Hz is now refused outright rather than warned about; and a session that starts losing trials to late
frames is stopped early rather than run to the end. Session length is unchanged.
Does the model predict which lens helps whose eyes?
Each observer's contrast sensitivity is measured on three cone-contrast axes and on a clay-sized moving disc; then, for six lens classes presented blind as faint moving discs, the study finds the fraction of each lens's real clay–background contrast at which the observer can just tell the direction of motion. The physical model predicts that fraction from the observer's own thresholds, and the comparison is the point of the study. 2 sessions under the current protocol from 4 participants, and 1 earlier session shown but not pooled.
Across the 2 pooled sessions, the measured multiplier was on median 2.4× the model's prediction from the observer's own thresholds, and the model ordered the lenses with a rank agreement of τ = +0.13 against the study's gate of 0.6. The model is being tested on its absolute scale as well as its order, and a moving clay-sized target is harder than the model's static thresholds imply; with this little data that is a first reading, not a verdict.
3
sessions shown, 2 under the current protocol
4
participants
1,700
main trials, 2% rejected for timing
+0.13
mean rank agreement τ, current protocol (gate ≥ 0.6)
Lenses: what the eyes measured against what the model predicted
For each lens class, the multiplier of the lens's real clay–background contrast at which the observer could just tell the direction of motion (circle), and the model's prediction from that observer's own thresholds (gold diamond). Lower is better; the grey bar between the two is the model's error, and ×n is the ratio.
Every lens condition is a faint grey-field disc carrying that lens's cone-contrast vector, so the observer cannot tell which lens a trial belongs to and expectation cannot enter. Faded marks are a session under protocol 0.1.0, shown but not pooled.
Predicted against measured, with each session's rank agreement
Every usable lens condition, on equal axes. On the diagonal the model was exactly right; above it, the eye found the lens harder than the model predicted. Kendall's τ per session: +1 means the model ordered the lenses exactly as the observer's eyes did, 0 means no relation.
The study's stated gate for a personal lens recommendation is a mean τ of at least 0.6 across observers, against at most 0.3 for a prediction that ignores the observer. Until then the recommender stays a physics estimate.
The red–green threshold is only a colour threshold when the flicker match before it found a minimum: it defines the axis the discs are shown on. In 1 pooled session the match ran to the edge of its adjustable range instead of settling, so that session used the standard observer's axis and its red–green threshold is shown but not pooled. The instrument now requires two settings that found a minimum and agree, and falls back to the standard axis when they do not; earlier records are re-scored by the same rule.
Every observer's sensitivity profile
Threshold contrast on the four measures, one dot per session; lower is more sensitive. The red bands are the study's screening bounds: a threshold above one is flagged to the observer, never diagnosed. Bars are medians of pooled thresholds once there are two or more.
A threshold above the screening bound is reported to the observer as a flag to see an optometrist, never as a diagnosis: the display's colour is assumed, not measured.
What one session looks like underneath: S02, clay shooter
The psychometric function behind each threshold: proportion correct against contrast or multiplier, the maximum-likelihood fit, and where it crosses the criterion.
Point area scales with the trials in each bin; whiskers are 95% intervals. The adaptive procedure puts most trials near the threshold, so the ends of each curve are sparse by design.
The profiler is a five-minute sighting-dominance test, run without an account: which eye keeps a target framed in an aperture made with both hands, across five positions on the screen, with either hand in front, at arm's length and near the face. Contributing the record is optional, anonymous when signed out and linked to the account by an internal id when signed in. The study asks how often sighting dominance is cross to the gun shoulder, how often it switches with gaze direction, and how often people's belief about their dominant eye is right.
49
contributions, 2026-09-12 to 2026-10-12
59%
right-eye sighting dominance; 22% left, 16% mixed
13%
cross-dominant among the 38 who named a gun shoulder
2%
switched eye with gaze direction
Sighting dominance, overall and by gun shoulder
Each bar is everyone in that group; the segments are the share whose sighting eye was the right, the left, or neither consistently (mixed). Undetermined means fewer than six of the thirteen trials gave a clear answer.
A person counts as right- or left-eye dominant when that eye kept the target on at least three trials in four; cross-dominant means the sighting eye is on the other side from the gun shoulder.
Of the 37 people who thought they knew their dominant eye before the test, 100% were right.
How much lead do shooters see, against the physics?
A crossing clay has to be shot at where it will be when the pellets arrive; how much lead that is, is arithmetic, and how much of it a shooter sees as right is a judgement nobody measures. The test shows a clay crossing at a known angular speed and has the observer set a ring where they would point the gun at the instant of firing, with no feedback; the setting against the shot's time of flight (a clay at 30 m, 0.095 s of flight) is the perceived lead, its error the bias, its spread the precision, by speed, direction and geometry. The study asks whether perceived lead differs by method, discipline and experience, and later whether it predicts anything on the ground.
6
contributions, 2026-09-12 to 2026-09-19
-6%
median bias against the physics (interquartile -16% to -4%)
33%
see less lead than the shot needs; 50% about right, 17% more
0%
saw relatively less lead on the fast targets
Perceived lead against the physics, by target speed
One point per person and speed: the lead they set divided by the lead the shot needed. Above the dashed line is more lead than the physics, below is less.
A person counts as seeing less or more lead than the shot needs when their mean setting is more than 15% below or above it; precision is the spread of repeated settings within a condition.
Median bias by how people say they shoot a crosser: Swing-through -4% (n = 3).
Where in the field do shooters pick up a target late?
A target is picked up some tenths of a second after it appears, and where it appears relative to the look point decides how late. The test has the observer hold a look point above a cluttered horizon while a small target appears at one of five positions across and one of three heights, and press the instant they see it, with empty trials to measure guessing and a localisation check on every press; the viewing distance is measured by the blind-spot procedure, not just asked for. The result is a map of median latency by side and height, the asymmetries in it, and the look-point adjustment it suggests. The study asks what these maps look like across shooters, whether the slow side follows the dominant eye or the gun shoulder, and, later, whether a map predicts anything on the ground.
3
contributions, 2026-09-12 to 2026-09-21; 3 with the viewing distance verified
396.0 ms
median pick-up latency across people
33%
even across the field; 0% slower on one side, 67% at one height
+0.1 ms/°
median growth of latency per degree out from the middle
The pooled pick-up map
Each cell is the median across people of their median latency at that position; darker is later.
A person counts as slower on a side or at a height when the difference in their own medians is 40 ms or more; absolute latencies carry a fixed display lag of a frame or two, comparisons between positions do not.