Project Stargate: Visualisation · Signals from the Periphery

World

CHANCE 25%

Trials

Hit rate

—

z vs 25%

—

PROJECT STARGATE · THE STATISTICS BENCH
1972-1995 · SRI → Fort Meade → SAIC
a blind protocol, a chance machine, a leaky lab
the claim: 32.2% where chance says 25
one dataset · two signed verdicts, 1995
▶ runs trials · z is always against 25%

Station

…

Evidence

fact ⇄ what kind of claim · live

Chance is exactly 25%: rank one of four, by construction

the one number in this file with no error bar at all

MEASURED

Autoganzfeld: 106 hits in 329 sessions, 32.2%

Bem & Honorton 1994 · the cleanest claimed margin

REPORTED

Utts: effect size ≈ 0.2 for the star SRI/SAIC viewers

AIR 1995 · mapped to 33.7% here, and that mapping is modelled

REPORTED

Utts: "far beyond what is expected by chance"

her verdict on the same dataset her co-reviewer read

READING

Hyman: real numbers, "does not warrant the conclusion"

same data · the other signature on the same review

READING

The customer: vague, ambiguous, never used alone

AIR programme review, Sept 1995 · the programme closed

REPORTED

The leaky lab's five switches and their sizes

documented mechanisms, illustrative magnitudes

MODELLED

23 years, ~$20M, six code names, 89,000 pages public

the declassified Star Gate collection, CIA reading room 2017

REPORTED

View

The world

…

True hit rate25.0%

The dial nobody in the real story could ever see. Here you can set it, then check how long it takes the statistics to find it.

▶ One number runs everything here. Chance is 25% by construction; the entire claimed effect is the sliver between 25 and about 32.

Run the experiment

Series length25 trials

Sessions were judged in series. Changing the length resets the tallies: a new experiment starts.

The leaks · leaky-lab view

Sensory cueshandled targets carry fingerprints of use

Sloppy randomizationHyman's 1985 ganzfeld finding

Judge not quite blindfeedback and familiarity leak in

Optional stoppingend the series while ahead

File drawerfailures never get filed

Both labs live in a zero-psi world. The sizes are illustrative and tagged MODELLED; the mechanisms are from the published critiques.

Display

Labels

Glow

Theory curvethe binomial the pile should become

Readout · live from the tallies

True rate in force

…

Trials run

Series filed

Observed hit rate

—

z against 25%

—

One-sided p

—

Left in the drawer

…

What is measured vs modelled here

Measured: the 25% baseline (arithmetic of ranking four), the binomial machinery, and every z this bench prints from its own runs. Reported: the autoganzfeld 106/329, Utts' effect sizes, the two 1995 verdicts, the operational stories and the programme facts. Modelled: the leak sizes, the ES-to-hit-rate mapping, and the theatre of the trial view, which draws the sketch after the outcome, inverting the real protocol's order for the sake of the show. Reading: whether the claimed margin was psi or procedure. This instrument prices both readings and buys neither.

The statistics model

Each trial is Bernoulli with the world's true rate p; a series of n trials files into the bin of its hit count, and the histogram converges on Binomial(n, p). The bench always tests against the chance baseline: z = (X − Np₀)/√(Np₀q₀) with p₀ = 0.25, one-sided p from the normal tail. The claimed markers are 32.2% (autoganzfeld, 106/329) and 33.7% (Utts' ES 0.2 via p = p₀ + ES·√(p₀q₀)). Optional stopping ends a series when the running rate clears the 90% one-sided band after ten trials; the file drawer refuses series under 18%. Both are honest cheats: each mechanism is in the critique literature, and their magnitudes here are illustrative.

Where this drawing takes liberties

No remote-viewing lab looked like this room: SRI's viewers sat in ordinary offices, and the ganzfeld's red light and headphones are represented here by one violet booth. The sketch is drawn after the judge's verdict is decided, which reverses the real order of events; real transcripts were pages of fragments, not one tidy pictogram. The dashed question mark is an editorial mark, not apparatus. And the leaky lab compresses twenty years of methodological argument into five switches with round numbers. The histograms, the z arithmetic and the quotations take no liberties at all.

…1 / 10

…

drag = orbit · scroll = zoom · ▶ runs trials