Sleight Is Live: The Instrument Is Pointed at You

Vaynerov Project Sleight runs four real experimental paradigms across one session of about nine minutes, and it writes every moment of that session — every onset, every choice, every answer, every time you look away — into an append-only journal before doing anything else with it. Sleight opens today, and the finale, The Reveal, is derived exclusively from that journal: nothing you are shown at the end has any other source. What you are handed are not lessons about the mind in general. They are receipts from yours, each stamped T+minutes:seconds from the moment you confirmed your session. We built a machine that misleads you for nine minutes and can then prove everything it did. Most software works the other way around.

Why an agency runs experiments on you

The fair question first: why does a web studio build psychology instruments at all? Because the Lab is how this studio argues. Anyone can claim craft; the claim costs nothing, and every portfolio page on the internet says the same words. So instead of adjectives, the Lab publishes instruments — Tremor listens to real seismometers, Backbone draws the real internet from the route collectors, the Orrery propagates real satellites overhead — and holds each of them to the standard the paid work is held to. Instruments, not demos, is the house law, and it doubles as an invitation: we show you our instruments so you can judge our work. Sleight is the sixth of them and the first pointed inward, because once you have drawn the planet's seismic pulse and the internet's routing table, the most interesting live signal left in a browser session is the person holding it.

There is an obvious way to build a psychology page, and Sleight was designed against it: a quiz that tells you about cognitive biases — ten slides, a score, a share button, the mind explained like a museum placard. A page like that teaches you a fact; an instrument that runs the experiment on you, then shows you what happened, second by second, teaches you something no placard can. So the page is honest at the level of the frame — the subtitle warns, up front, that the instrument “will not always be straight with you” — and covert everywhere below it. In-session, nothing hints at which moments are the experiment. That is not a trick for its own sake; it is the operating condition of the science. An instrument that tells you where the trick is measures nothing.

Nine minutes, measured

The four paradigms are the canon of experimental psychology, run as published rather than paraphrased: false memory in the Deese–Roediger–McDermott tradition (Roediger & McDermott, 1995), change blindness (Simons & Levin, 1998), choice blindness (Johansson, Hall, Sikström & Olsson, 2005), and anchoring in judgment under uncertainty (Tversky & Kahneman, 1974). Where each one hides inside the nine minutes, this dispatch will not say. The instrument will be run by people who read this first, and the literature's quiet joke is that forewarning tends not to save you — which is exactly what the replay button is for.

From the visitor's side, the session is deliberately calm. A parlor registers you — no name, no account, just a session number that sits quietly in the corner of the stage. You hold a run of words. You tend a small instrument bench, settling tones and matching glyphs while the instrument watches how you watch. You choose between abstract seals, and later defend your choices in a sentence of your own typing. Then comes the audit — and under every answer sits the real measuring device: a confidence slider from guessing to certain, because what Sleight measures is not whether you are right. It is how sure you are when you are wrong. There are no visible timers anywhere — time pressure contaminates reaction times — so the only chronometer on screen is one hairline progress filament. And somewhere in those minutes there is a shutter, and nothing behind it ever moves while you can see it.

The journal law

The instrument cannot exaggerate its own trick — if it isn't in the journal, it didn't happen.

That sentence is not copy; it is the architecture. The journal is append-only: events are written once, in order, at the millisecond they occur, and no later phase of the session may edit, reorder, or delete one. The Reveal never reads the interface's own state — it reads the journal, and only the journal. Look away and the schedulers freeze and an interruption is logged, along with how long you were gone. Refresh and the session restarts from nothing, by design: mid-session persistence would be a spoiler channel one devtools tab away, so the only things Sleight keeps in your browser between visits are whether you have finished before, your consent choice, and your fidelity grade. Even chance is on the record — everything random about a session, every assignment, every order, every generated mark, expands deterministically from a single seed drawn at the start. One seed, and the whole session is a mathematical consequence.

That determinism is what pays for the receipts. Every receipt is a reconstruction, not a screenshot: nothing was captured, nothing was saved as an image. When the reveal shows you a before and an after, it re-renders both from the journal — same seed, same code path, pixel for pixel — next to what you said about them. A reconstruction is a stronger kind of evidence than a picture: a picture asserts, a re-derivation proves, and a receipt that stretched the truth would simply fail to re-render. In the reveal the session itself becomes a small constellation: every journal moment a steel star threaded on a helix of time, with the fabricated moments burning magenta among them, and the tour walks that sky one receipt at a time.

• Every receipt carries its timestamp. “At T+04:12” is a different kind of argument than “studies show”.
• One receipt re-renders a before and an after from the journal — same seed, same pixels, one difference.
• One quotes a sentence you typed back to you, verbatim, next to what was actually on the table.
• One sets your estimate against the record's two histograms and lets the gap speak for itself.
• The fidelity report closes the tour: how sure you were, held against how right you were.

Six languages, six memories

Shipping Sleight in six languages surfaced the strangest engineering problem in the whole build, and it has nothing to do with rendering. One of the four paradigms runs on word lists whose entire power is associative: every word in a list pulls, in a native speaker's memory, toward one word that is never shown. That pull is not in the words — it is in the reader, accumulated over a lifetime inside one language. Translate such a list and the surface survives while the pull dies: the words come out correct, the experiment comes out dead, and no reviewer can tell the difference by reading. It is the one place on this site where a perfect translation is a broken build.

So the lists were never translated. They were rebuilt, language by language, from each language's own published free-association norms — the Russian Associative Dictionary for Russian, Alonso and colleagues' norms for Spanish, Steffens and Mecklenbräuker's for German, Corson and Verrier's for French, Albuquerque's norms for Portuguese. Each locale's stimuli live in typed code modules the translation tooling cannot reach, because an admin correcting a “typo” in a stimulus mid-experiment would be quietly rewriting the science. What everyday territory each list circles, this dispatch — under its own law — will not say, in any of the six. And a unit test patrols the border: no critical word may appear anywhere in the site's translation catalogs, in any language. The words this dispatch is not allowed to contain are checked by a machine before it ships.

The honest floor

The record holds itself to the same honesty as the receipts. Sleight shows no “X% of visitors” line until the aggregate record holds at least fifty sessions; until then the copy says plainly that the record is young, and every number on screen comes from the published literature, labeled as literature. The estimate histograms wait for twelve sessions per condition before they draw at all. The record also excludes the knowing: run the instrument a second time and your session carries a replay flag — welcome on screen, quoted nowhere, because a visitor who has already seen the seams is no longer a naive subject and would quietly corrupt the numbers the receipts stand on. And the moment of maximum vulnerability — right after the instrument has shown you your own fabrications — asks you for nothing: no email field, no sign-up, no personality type to share. A young instrument that manufactured social proof from an empty table would be lying in precisely the way it exists to expose.

Sleight ships in six languages and runs on phones — low-tier devices trade the 3D constellation for a cinematic 2D reveal with the same receipts and the same timestamps. Like everything else in the Lab, it is free and needs no account: vaynerov.com/sleight. Sit for it once, cold. Then run it again — knowing — and see how much that knowledge buys you.