Project Stargate

The Protocol view is one remote-viewing trial as a physical machine, on a dark stage under a violet lamp. A randomizer ring spins over a pool of four targets; one card travels, sealed, to the site pedestal; in a translucent booth the viewer's pen fills a sheet; the transcript then flies to the judge's table, where four candidates rise and take their ranks, and a beacon flashes mint for a hit, coral for a miss. Every station wears a label you can click: the booth explains sensory isolation, the randomizer explains why honesty lives there, the judge explains rank-of-4 and why chance is exactly 25% by construction. One line is drawn dashed with a question mark over it: the claim itself. Single trials, you will notice, look identical in every world.

The Chance Machine is the heart of the bench. Every series of trials drops as a glowing ball into a histogram bin, and the pile converges on the binomial bell around the mint 25% line. Two markers stand over the board: manila at 32.2%, the autoganzfeld record, and violet at 33.7%, Utts' effect size 0.2 mapped onto the same axis. Choose your world, pure chance, the claimed rate, or a free dial, and run: at small N the worlds are indistinguishable, and as trials mount, the z readout against the 25% baseline hardens. The whole historical argument is visible as geometry: the claimed margin is one bin wide, and its significance is entirely a function of how many times you sample it.

The Leaky Lab is the counter-argument, built. Two laboratories run side by side in a world with zero psi: the clean one keeps the protocol and its pile sits obediently on the mint line; the leaky one has five switches, each a documented critique mechanism, sensory cues on handled targets, sloppy randomization, a judge who is not quite blind, series stopped while ahead, and failures left in the file drawer. Flip them and watch the coral pile drift onto the manila line: a headline effect, manufactured from procedure alone. The panel is explicit about what this proves and what it does not: leaks can buy the margin; whether they did is exactly what the two 1995 verdicts dispute.

The File view lays the program out card by card. The code-name lineage runs along a timeline, SCANATE through GRILL FLAME and SUN STREAK to STAR GATE. Below it hang the two verdict cards, quoting Utts and Hyman on the same dataset, side by side and at equal size, which is this site's entire editorial stance rendered as layout. Three smaller cards hold the celebrated ops stories at the arm's length the file itself holds them, the customer's verdict that closed the program, and the archive: about twenty million dollars, six code names, 89,000 pages, public since 2017. Click any card to read more.

On this site's culture shelf sits a review of Stranger Things: a show that weaves MKUltra, sensory-deprivation tanks and a psychic child into small-town horror. That review carries three tags: MKUltra, remote viewing, Stargate. This instrument opens the most computable of the three. Project Stargate is the real file that fiction loves to cite: the government really paid, the experiments really ran for twenty-three years, and the archive really went public in 2017. And its lasting legacy is not any single viewing; it is a statistical argument, in which two respected scholars read one dataset and signed opposite conclusions.

It is also this station's first instrument to touch the government-programs half of the archive, and its first instrument of pure statistics. There is a deeper reason to pick it: the side-by-side Utts and Hyman conclusions in the 1995 AIR review are the real-world prototype of this site's neutral-curation stance. This site's method is to summarise, attribute and leave the verdict to the reader; that review is what the method looks like inside a federal document. So this bench does not adjudicate remote viewing. It builds the 25% baseline as a machine you can crank, draws the claimed 32% as a line you can weigh with your own eyes, turns the critique literature's flaws into five switches you can flip, and hangs the two verdicts side by side, quoted.

One protocol and one distribution run the whole bench, and the panel tags every number as measured, reported, modelled or read.

hit = rank 1 of 4 ⇒ p₀ = 25% · series: X ~ Binomial(n, p) · z = (X − Np₀)/√(Np₀q₀) · claimed: 32.2% (106/329) · ES 0.2 ⇒ 33.7%

The baseline is arithmetic. The judge must rank exactly four candidates, so a first-place match by pure guessing is exactly one in four. Nothing about remote viewing needs to be believed or doubted to accept this number; it is the construction of the experiment. Every z the bench prints is measured against it.

The worlds are explicit. A world is nothing but a true hit probability: 25% for pure chance, 32.2% for the claimed autoganzfeld rate played back as if true, or any rate you set on the dial. The bench's point is epistemic: the dial is invisible from inside the data. All anyone ever sees is the pile, and the game is inferring the dial from it.

The statistics are live, not scripted. Trials are real Bernoulli draws; histograms accumulate real runs; z = (X − Np₀)/√(Np₀q₀) and the one-sided p are recomputed from the actual tallies as you watch. Occasionally a pure-chance run will clear three sigma in front of you. The bench does not prevent this; it is the honest price of sampling, and the verdict line will tell you what you just saw.

The leaks are mechanisms, not slanders. Each switch implements a critique that exists in print: cues and randomization from Hyman's 1985 appraisal and the 1986 joint communiqué, judging concentration from his 1995 review, optional stopping and the file drawer from the general methodological literature. Their magnitudes here are illustrative round numbers, tagged MODELLED, because the historical sizes are exactly what nobody can measure anymore.

And the verdicts are quoted, not summarised. The two AIR conclusions appear as written, at equal visual weight, with their agreements preserved: both signatories accepted that the departures were too large for statistical flukes. Where they part, on whether that establishes psychic functioning, the bench stops with them.

The baseline is arithmetic, the claimed margins are quoted from their publications, the machinery is live probability you can audit, the leak sizes are disclosed illustrations, and the verdicts are two real signatures on one real document. What this bench refuses to do is add a third signature. The program's legacy is a perfect specimen of the hardest question in any anomaly file: not "are the numbers strange" but "what earned the strangeness". Twenty-three years bought seven extra hits per hundred and two readings of them. Both readings are here, running.

Three worlds and five leak switches, each a published position, not a mood.

• Chance, 25%. The null world. Run it long and the pile becomes the binomial bell around the mint line: the shape of pure luck, including the outlier sessions that would each make a headline alone. This is the default, because the default should be the null.
• Claimed, 32.2%. The autoganzfeld record, 106 hits in 329 sessions, played back as if it were the truth of the world. The pile shifts right by barely one bin, and z climbs anyway as N grows: the whole Utts argument in one animation.
• The free dial. Set any true rate from 15 to 45 percent and discover how many trials separate "invisible" from "undeniable". The dial nobody in the real story could see is the one you get to hold.
• The five leaks. Sensory cues, sloppy randomization, an unblind judge, optional stopping, the file drawer. Individually worth a few points each; together they carry a zero-psi lab to the claimed line. The switches demonstrate capability, and the panel is explicit that capability is not history.
• Series length. Eight to sixty trials per filed series. Shorter series make wilder piles and better outlier anecdotes; longer series tame them. A quiet lesson in why anecdote and aggregate disagree so reliably.

Five claims about the program, with who argues each and where it lands. One is settled and closed the program; three are live and share the statistics; one is a reading the record never paid for.

1. Watch one full trial without touching anything. The randomizer picks, the viewer sketches, the transcript flies, the judge ranks, the beacon flashes. Then ask yourself the only question that matters at this scale: was that a psychic world or a chance world? You cannot know. That impossibility is the reason the rest of the bench exists.
2. Run 200 series of pure chance. In the Chance Machine, chance world, hit the burst button and read the pile: sessions at 36% and 40% are sitting there, earned by nobody. Any one of them, alone, is a convincing story. The pile is the antidote to the story.
3. Switch to the claimed world and find the difference. Same machine, 32.2%. The pile moves one bin to the right. Stand back and honestly ask whether your eye could tell these two piles apart. Then watch the z readout, which can.
4. Let sigma do its slow work. Stay in the claimed world at MAX speed. z passes 3, then 5; the one-sided p acquires exponents. Nothing about the margin changed, only N. That is Utts' entire case, and it is not wrong: at face value, chance is dead. The question that survives is the face value.
5. Buy the margin with procedure. In the Leaky Lab, turn on cues and the unblind judge, then add the file drawer. Watch the coral pile reach 32% in a world containing zero psi, and watch the drawer counter count what was never filed. This is Hyman's ledger as a physical object.
6. End in the File and read both cards aloud. Utts first, then Hyman, or the other way around. Notice they agree on the arithmetic and part on one sentence. Every anomaly file on this station eventually reaches this exact fork; this is the case where the fork got two federal signatures.

The honest line between what is measured, what the record reports, what is modelled here, and what is a reading:

In one line: the 25% baseline and the binomial machinery are arithmetic you can audit; the claimed margins, 32.2% autoganzfeld and effect size 0.2, are quoted from their publications and tagged as reported; the leak switches implement documented critique mechanisms at illustrative sizes and are tagged as modelled, as is the trial theatre that draws the sketch after the verdict; the operational legends are retellings and are shelved as readings; and the two 1995 verdicts are quoted verbatim, at equal weight, because two qualified reviewers read one dataset and signed opposite conclusions, which is precisely why this case belongs on this station. Run the machine and decide for yourself.

• Utts, J. "An Assessment of the Evidence for Psychic Functioning" (AIR, 1995). The pro verdict, quoted verbatim: "well established... far beyond what is expected by chance"; the ES ≈ 0.2 figures for the experienced viewers.
• Hyman, R. "Evaluation of the Program on Anomalous Mental Phenomena" (AIR, 1995). The counter verdict, quoted verbatim, plus the ledger this bench builds its leaky lab from: one lab, few viewers, concentrated judging, no independent replication.
• Mumford, M., Rose, A. & Goslin, D. "An Evaluation of Remote Viewing: Research and Applications" (American Institutes for Research, 29 Sept 1995). The programme review: vague, ambiguous, never used alone; the recommendation to terminate.
• Bem, D. & Honorton, C. "Does Psi Exist? Replicable Evidence for an Anomalous Process of Information Transfer", Psychological Bulletin 115:1 (1994). The autoganzfeld: 106 hits in 329 sessions, 32.2%, the bench's CLAIMED world.
• Hyman, R. "The Ganzfeld Psi Experiment: A Critical Appraisal", Journal of Parapsychology 49 (1985), and Hyman & Honorton, "A Joint Communiqué" (1986). The randomization and cue critiques, and the protocol standards they produced.
• Milton, J. & Wiseman, R. "Does Psi Exist? Lack of Replication of an Anomalous Process of Information Transfer", Psychological Bulletin 125:4 (1999); and Bem, Palmer & Broughton's 2001 reply. The replication fight after the program closed.
• Targ, R. & Puthoff, H. "Information transmission under conditions of sensory shielding", Nature 251 (1974). Where the SRI work went public; the paper the whole lineage cites.
• Puthoff, H. "CIA-Initiated Remote Viewing at Stanford Research Institute", Journal of Scientific Exploration 10:1 (1996). The founder's history: SCANATE, Price, the early coordinates work.
• The Star Gate collection, CIA FOIA reading room (released January 2017, ~89,000 pages): session transcripts, tasking memos, programme reviews, the code-name lineage.
• Smith, P. H. Reading the Enemy's Mind (2005), and McMoneagle, J. Memoirs of a Psychic Spy (2002/2006). The inside retellings of Fort Meade, the Typhoon claim, the Iran hostage sessions, read here as memoir.
• Signals from the Periphery: the Culture 13 review of Stranger Things, and the wiki cluster on the Stargate program, remote viewing, the AIR review, and the named viewers and cases.