How your stats are measured

Every number in the stats section derives from one primitive: blind rank-order judging. Nothing is self-reported, nothing is machine-scored, and nothing is inflated to feel encouraging. This page explains what is measured, what is deliberately not measured, and how to read your own numbers honestly.

Where a score comes from

When your session is judged, each judge receives your transcript and a lineup of candidate images: the real target and several decoys, in a randomized order. The judge does not know which is real. They rank the candidates from best match to worst.

Your score for a judgment is the position of the real target in that ranking. If the judge placed the real target first, the session registers a hit for that judgment. That is the whole definition. A hit means a blind judge could pick your target out of a lineup using nothing but what you produced.

Rank 2 is shown as a partial in your feedback. It is a nuance, not an achievement: partials are never counted in your acquisition rate, your calibration position, or your belt criteria. Everything that matters counts strict rank-1 hits.

Chance

With a lineup of N candidates, a judge ranking the real target first by pure luck happens 1 time in N. That number is the floor under every rate you see. A 25% acquisition rate against 4-candidate lineups is not evidence of anything. It is the sound of coin flips.

The calibration bands on your acquisition curve come from the reference: newcomers distinguish major gestalts about 25% of the time, and demonstrable talent with months of practice reaches about 50%. The bands exist to place your curve against those documented figures, not to promise a trajectory.

Mean rank

The binary hit rate throws away information: ranking the real target second out of five is different from ranking it last, and neither is a hit. Mean rank keeps that information. Across judgments, the average position of the real target has an expected value of (N+1)/2 under chance. With 5 candidates, chance sits at 3.0. A mean rank reliably below that line means judges tend to place your target high even when they do not place it first.

Why descriptors are not scored

The stats do not tell you which of your descriptors were right. That is deliberate, twice over. Scoring your own descriptors against the revealed target is self-analysis, and the reference is unambiguous that self-judging retrains exactly the assumption-making habit the practice exists to remove. And the app has no automated judge today: a per-descriptor accuracy layer may return later, under the same blinding rules as everything else.

What you get instead is a production profile: how much you produce in each descriptor category, how often you flag AOL, how often you mark impressions tentative. Volume and habits, not accuracy. Read it as a picture of how you work, and use it to notice categories you rarely touch.

The timing view

The timing histograms show when in a session your output happens: descriptors and strokes across the session's duration. The reference's model of contact is a rapid taste followed by internal processing, which is why early output tends to carry signal and late output tends to carry decoding. The view shows your distribution so you can see your own pattern. It makes no claim about which of your impressions were accurate.

Conditions: LST and Kp

Local Sidereal Time is star time: a clock tied to the Earth's rotation relative to the distant sky rather than the sun, so a given LST hour corresponds to a fixed orientation of your longitude toward the galaxy. Published free-response data shows effect sizes rising several-fold in a window around 13.5h LST.

Kp is a global index of geomagnetic disturbance, updated in 3-hour bins. In the same published data, geomagnetic activity correlates negatively with performance, but only inside the LST window. Outside it, the correlation is near zero.

That interaction is why the app logs both and reads them jointly, never Kp alone. The stats view shows your acquisition inside versus outside the window, split by Kp level within it, and shows today's window in your local clock time. It is a soft suggestion. Nothing in the app gates on conditions, and building a list of required conditions is explicitly warned against in the reference.

Plateaus

Progress in this practice arrives as plateaus lasting weeks or months, sometimes followed by sudden jumps. That pattern is documented in the reference, and the stats engine treats it accordingly: a flat curve renders a note saying so. A plateau is not failure and does not demote anything. If the practice stops being interesting during one, the reference's advice is to step away rather than grind.

Above chance, in the statistical sense

The highest belt criteria use the reference's definition of a remote viewer: a documented above-chance hit rate sustained over about a year under protocol. Above chance does not mean a rate numerically higher than the chance line. Short runs above the line happen constantly by luck. It means enough hits over enough sessions that luck stops being a plausible explanation for the total, which is a moving bar: the fewer sessions in the window, the higher the rate must be. The belt view shows you the current bar in plain numbers: how many hits out of how many sessions it takes, right now, for your record to be distinguishable from chance.

The numbers here are conservative by design. That is the point of the instrument.