How the research works

Blinding

A viewer never sees the target pool before sealing their response. Target images are stored outside the public web root and served only through a gated request, so there is no address to guess. When a session is unblinded for any reason — including by the administrator — the system records it and that session is excluded from clean results rather than quietly kept.

Present and future conditions

In the present condition the target is generated when your session opens: it exists while you are perceiving it. In the future condition the target is not generated or selected until after your response is sealed — so at the moment you respond, there is no target yet to read.

These are different claims about how perception works, and pooling them would hide the more interesting result. They are tracked and reported separately, always.

Scoring

Multiple-choice sessions are scored against a chance baseline of 25%. Pick-your-own sessions are compared against the expected number of matches for that game, which is shown alongside every result. Sample sizes in the tens mean nothing; these figures only become meaningful in the hundreds, and the interface says so rather than flattering an early streak.

Targets

Numbers are drawn using a cryptographic random source rather than a standard pseudo-random generator. For a platform whose entire output is statistical, a weak generator would put a structural artefact in every result we ever report.

What we will publish

Aggregate and de-identified results, including negative ones. A study that only reports its successes is advertising.