What. 252 questions were drafted. 27 forecast-framed items (holiday Over/Under wording) were withheld before publication and rewritten as respondent questions for Issue 05. The published book is 225. The withhold came after seeing predictions. Editorial note, September 9, 2026: Issue numbering changed after publication. These 27 questions remain withheld for a future Issue and are not part of Issue 05.
Panel. n=1,000 synthetic respondents per question. National items use a U.S. adult draw. State items use that state's census panel. The District of Columbia is excluded from the state map, as pre-registered.
Aggregation. Option shares from the synthetic panel. No post-hoc swap after outcomes.
Scoring, clarified October 17. The registered headline is winner-call rate: eligible hits divided by hits plus misses, using each question's frozen call rule. Option-share mean absolute error is supplementary where the options map one-to-one. Tie handling, population and instrument matching, missing results, and discontinued series must be checked under the registered rules. Not all 225 items may yield eligible grades. Unscorable items stay visible and will be reported with reasons; grades remain pending until reviewed. The final scorecard date is January 8, 2027.
Comparison arm. A second archived run will be scored at resolution. It is not the headline.
Hashes. c73a440e commits the registered canonical question and metadata fields, excluding response shares. 801ed3f3 adds the frozen option shares. These commitments cover those original fields; they do not hash every line of scoring prose or the bytes of this public display extract. Hash verification alone does not establish release or interview timing for each row.