Methodology

What happens to your sessions

Sessions are retained and analysed. Aggregate results across viewers are not currently public and no figures are published anywhere. You get your own data back in full on your sessions page.

Everything is counted

Every session that is opened enters the record, including the ones that expire without a ranking. They enter as explicit nulls and are never quietly dropped. A platform that logged only the sessions that finished would have a file drawer, and the figures coming out of it would not mean much.

The statistic

The primary statistic is the sum of the ranks you give the target, against the exact permutation distribution for five candidates. Direct-hit rate is secondary: it asks only whether the target came first, which throws away most of what a full ranking says. This was fixed before any data was collected and will not be changed after seeing results.

What the commitment does and does not prove

At session open the server fixes the target and publishes SHA-256 of the target, a random value and the session id. At reveal it publishes the target and the random value, so you can recompute the hash yourself. This proves the answer was not changed after seeing your work. It does not protect you from us: we run the server, and an operator can always cheat on their own machine. Saying so is more useful than implying a guarantee that does not exist.

Declared limitations

• Targets are still images only. The one direct comparison in the literature found dynamic targets did considerably better than static ones, so a null result here cannot distinguish "no effect" from "wrong kind of target". This is stated in advance so that it is a known limitation rather than a later excuse.
• Sessions here are solitary. The historical SRI protocol always included a second person who prompted and questioned the viewer; there is no monitor on this site. That is a deliberate product choice, and it makes this a solitary variant of the protocol rather than a reproduction of those conditions.
• You rank the candidates yourself. That is a standard free-response protocol and it produces valid statistics, but a forced choice cannot capture a session that described the target well and still ranked it wrong.

Why your own figures lag

A session you have ranked but not revealed is counted on your sessions page and deliberately left out of your statistic. If it were included you could recover its rank by arithmetic, which would hand you the answer before you looked. Your numbers therefore trail the underlying record by however many sessions you leave unrevealed.

What we hold about you

A nick, a password hash and a recovery-phrase hash. No email, no phone, no third-party analytics. Your IP address reaches our server and CDN, as it does with any website, and is never stored in plaintext: it is truncated, salted with a key that rotates every 24 hours, hashed, and used only for rate limiting. For real network anonymity, use Tor or a VPN.

Asking to be forgotten

Ask and your nick, your password and recovery hashes, and every sheet you photographed are deleted — the photographs from storage, permanently, with no backup to restore them from. The account can never be signed into or named again. Your sessions stay: rank, timestamps, target, pre-session state, and the hashes that prove each sheet was sealed before its reveal. Two reasons, and neither is convenience. Sessions that can be deleted let anyone quietly remove the ones that went badly, which biases the whole dataset in exactly the direction this site exists to avoid. And the retained hashes are what keep a sealed session verifiable after the photograph is gone — a hash of a picture nobody holds any more identifies nobody, and is the only remaining evidence that the record was not written afterwards.