{
 "name": "Lives fitted to records reproduce how a population splits on five fresh 2024 questions (party group x age group)",
 "registered": "2026-10-10",
 "why": "The gun-permit test (protocol-gunlaw.json) missed: the records regression was 3.22 points off with every age gap inside its interval, the voice 16.4. That stays published. This new study keeps its rule per question (six-cell error at most 3.0 points AND each party group's age gap inside its 95% interval) and the battery rule of protocol-battery.json (at least 4 of 5 questions), on five 2024 questions no run here has read.",
 "honesty": "For the five questions and their sibling questions, only the number of valid 2024 answers was read (abnomore 1522, abpoor 1531, absingle 1525, colrac 2177, spkrac 1128); no 2024 answer distribution was read. The 2024 answers sit in gss_2024_SEALED.parquet, opened only by records_battery.py run. Development read 2006-2022 only.",
 "data": {
  "source": "General Social Survey cumulative file 1972-2024, Release 3a (NORC, July 2026), public; GSS_stata.zip downloaded 2026-10-10 from gss.norc.org, sha256 d4fd4e5057d53d6023e55d5114685430321c3ab00ef2bc11099c2358efc65dfe",
  "inputs": "prepare_records.py at this commit; gss_dev.parquet d0703d89812784eabfcddaa6bd54de414ae404caa98efa32553c7aa16c789889, gss_2024_facts.parquet ef043b38fbd9f88215d100b14f735046675118a9eb1fe3d6b6af46e2c4bb9ac0, gss_2024_SEALED.parquet 9b06a35cbcdd2516d7d62d721cfd939aa92f3cfdd0e7dcb7349950a3198794cd",
  "weights": "WTSSPS",
  "train": "2018, 2021, 2022",
  "test": "2024 only"
 },
 "questions": {
  "abnomore": "Should it be possible for a pregnant woman to obtain a legal abortion if she is married and does not want any more children? share yes",
  "abpoor": "... if the family has a very low income and cannot afford any more children? share yes",
  "absingle": "... if she is not married and does not want to marry the man? share yes",
  "colrac": "Should a person who believes that Blacks are genetically inferior be allowed to teach in a college or university? share allowed",
  "spkrac": "If such a person wanted to make a speech in your community claiming that Blacks are inferior, should he be allowed to speak? share yes"
 },
 "selection": "Fixed in development before any 2024 answer was read: among the binary questions of four GSS families (abortion 7, civil liberties 9, sexual morality 6, gender roles 4; devsib2.sh), the ones on which this method met the rule in all five development waves (2012, 2014, 2016, 2018, 2022, each predicted from the three waves before it: devsib2-results.txt). Five qualified, all from two families; all five are registered.",
 "population": "every 2024 respondent with a valid answer to the question and valid facts (the facts of protocol-gunlaw.json)",
 "method": "records_battery.py at this commit. Per question: logistic regression of the answer on the person's facts (age, sex, race, degree, party identification, political views, region), party group x age interactions, the person's own 2024 answers to the other questions of the same family (never the question itself; a missing sibling answer has its own indicator), and their answers to questions already read and published here (gun permits, death penalty, marijuana, the spending questions, confidence in the Supreme Court and the executive, immigrants, police striking a citizen); fitted on 2018, 2021 and 2022 with weights halving every four years back from 2022 (L2 0.003); predictions averaged with WTSSPS over each cell. This is the imputation of a split questionnaire design (Raghunathan and Grizzle 1995, JASA 90:54-63): a question one population was not asked, filled from what the same people answered.",
 "outcomes": {
  "primary": "per question: six-cell MAE in percentage points (party group: Democrat partyid 0-2, Independent 3, Republican 4-6; age 18-44 / 45+), with 95% intervals from 2,000 bootstrap resamples of 2024 respondents (seed 20261005), and the protocol-gunlaw.json rule: MAE <= 3.0 AND each party group's predicted age gap (18-44 minus 45+) inside the measured gap's 95% bootstrap interval",
  "battery": "the claim 'lives fitted to records reproduce a measured split' is met if the method matches on at least 4 of the 5 questions (protocol-battery.json's rule)",
  "also": "carry-forward (2022 cell share) per question, reported for comparison only"
 },
 "development": "devrec.py, devsib.py, devsib2.sh; summary devrec-summary.txt, per question and wave devsib2-results.txt. Facts-only records (the registered gun-permit model) met the rule on 3 to 6 of 30-35 questions in 2022 and on 13 to 15 of 35 in 2016; adding sibling answers brought the sibling families to 11 of 25 in 2022 and 19 of 28 in 2018. With the frozen code in check mode (2022 predicted from 2016, 2018 and 2021): abnomore 1.93, abpoor 2.08, absingle 2.16, colrac 2.07, spkrac 2.61, every gap inside (check-2022.json). Without a 2021 wave (2022 from 2014-2018) colrac and spkrac missed (4.11, 4.65) and absingle missed (3.32).",
 "limits": "The siblings are answered by the same people in the same wave, so a sample that leans one way leans the same way in its predictions; this is the use the method is for (filling a question from a person's other answers), not forecasting a question nobody in the wave was asked. The voice is not part of this study.",
 "order": "after this commit: records_battery.py run on [host name redacted] as one capped unit; results-records-battery.json committed whatever it says",
 "seed": 20261005
}