How common was the choice you made?

Expected SHA-256: e7244f8696ff53e14c43d065a84ac224858b3c181dd4a7d29981cf5b969db57b

Checking live bytes...

curl -s https://artwaste.land/strata/false-consensus/analysis.mjs | shasum -a 256

The living display is locked if any Table 1 row stops reproducing its expected gap, derived from the printed means, or the printed summary. One-byte seal mutations and wrong-cell anchor mutations are both required to fail offline.

D is mean option-A forecast among A choosers minus the same mean among B choosers. Positive D is differential projection. It is not automatic evidence of error, because a person's own response can be legitimate sample-of-one information.

The accepted crowd supplies a descriptive target the 1977 study did not independently measure. Every calibration number, the two side errors and the pooled forecast error alike, scores forecasts against the other accepted rows only, never against a share that includes the forecaster's own row. These values are exact for these rows, not for society; the paired sequences bound population contrasts under their stated conditions.

The frozen store accepts only item, choice, estimate, and exposure. It has field validation, a per-IP daily limit, a global daily limit, and a row cap. It has no signed assignment token, honeypot, elapsed-time field, or burst flag. Therefore this page has no honest raw-versus-screened split and says accepted row, never screened person.

The browser assignment is random for this session and the local contributed marker is advisory. A cleared browser, another device, coordinated submissions, and bots remain possible. No demographics, free text, account, email, precise time, location, user agent, or fingerprint enters the declared payload.

The twelve streams use alpha 0.05/12 and a 401-point betting grid. Time-uniform validity assumes a stable data-generating mean and predictable bets. It does not make these self-selected visitors representative, remove drift, or repair dependence and gaming.

At zero, the page says zero. Before both choices reach 10 accepted rows on an item, only counts appear. Calibration and D are suppressed. Early ground truth changes every time a row arrives.

No participant-level 1977 data, code, or open data archive was located in the documented search. The one-decimal cells cannot recover distributions, standard deviations, F tests, p values, or participant dots. The displayed F statistic is reported, not reproduced.

This crowd is global, self-selected, online, choice-first, unpaid, and answering one trivial new item. The 1977 participants were Stanford undergraduates answering consequential vignettes, with estimates before choices. Culture, order, item stakes, medium, and familiarity may all change the result.

Only a complete JSON object with exactly two valid keys becomes a value. Refusals and unparseable outputs remain in the denominator, with counts and rates. No regex extraction, repair, reprompt, silent drop, or forced balance is allowed. The result file has no frozen human snapshot, so human-target forecasts cannot honestly be scored against a later moving crowd.

The often repeated meta-analytic r = .31 was not used because the 1985 full results pages were not directly inspected. Its title and publisher record say 115 tests, not the conflicting 155 found in one later reference. No correction or retraction was located for the 1977 or 1985 article as of 2026-08-21, which is a search result, not proof of absence.