Every test run on past elections

each written-up test and check, by its own title

Every test run on these past elections

92 written-up tests and checks of the forecast models and their data, most of them registered before they were run, scored on the same past elections as the record above. That reuse is why the record is development evidence, not proof: November 2026, scored from forecasts published before it, is the first clean test.

• Result · Does presidential approval explain the polling miss? (approval_poll_error_result.md)
• Result: the model and statistical findings of the 2026-10-26 audit (M-14, M-18, 7.1, 7.2), 2026-10-27 (audit7_result.md)
• Audit M03 and M04: the district voter profiles (audit_m03_m04_result.md)
• Audit M05 and M06: the turnout experiments, rebuilt on a sound construction (audit_m05_m06_result.md)
• Audit M07: one target for the statewide forecasts, and the undecideds (audit_m07_result.md)
• Audit M08: overlapping waves of one firm counted as independent polls (audit_m08_result.md)
• Audit M09: statewide chance from the band's own distribution (2026-10-10) (audit_m09_result.md)
• Audit M10: chamber errors normalised (2026-10-10) (audit_m10_result.md)
• Audit M11: the scale law's intercept in a midterm year (audit_m11_result.md)
• Audit M12: the House national term at one horizon (2026-10-10) (audit_m12_result.md)
• Audit M13: the statewide record on real poll release dates (2026-10-10) (audit_m13_result.md)
• Audit M16: what the statewide record can and cannot show (audit_m16_result.md)
• Challenger batch 01 · results (batch01_results.md)
• Result: blending on information, not on total variance (2026-10-24) (blend_shared_variance_result.md)
• Certifying results against MIT Election Lab (certified_results_crosscheck.md)
• Result: CES bottom-up district estimate - rule met, gain indistinguishable from noise; not adopted (2026-10-01) (ces_bottom_up_result.md)
• Result: Senate and governor chamber gate (2026-10-03) (chamber_gate_result.md)
• Result: chamber simulations that keep each race's own forecast (2026-10-25): ADOPTED (chamber_marginals_result.md)
• Result: the deciding round (deciding_round_result.md)
• Result: the measured design effect in place of 1.6 (2026-10-25): ADOPTED (design_effect_measured_result.md)
• Result: the de-tiding fix, A/B as registered (run 2026-10-26) (detide_prior_result.md)
• Gate result: ground-up district model (run 2026-10-01, after the pre-registration at 02:24) (district_gate_result.md)
• Result: a shared district-poll term in the House model (TODO H86), 2026-10-28 (district_poll_shared_result.md)
• The district-poll arms (v2 and v4), scored on 2026 (district_polls_v4_2026_result.md)
• Result: the House error model's state variance (M-14), 2026-10-27 (district_state_var_result.md)
• Result: the bottom-up voter profile does not improve the forecast (2026-10-01) (electorate_backtest_result.md)
• Result: the measurement-excess term compared like with like (run 2026-10-26) (excess_measurement_adjusted_result.md)
• Result: the 2012/2014 outside money and the ND 2004 / MS governor records (TODO H128 (4), (10)), 2026-10-28 (h128_money_and_history_result.md)
• Historical poll corrections of 2026-10-23: result (historical_poll_corrections_2026_10_23_result.md)
• Result: restoring the shares of historical polls (registered in historical_poll_shares_preregistration.md) (historical_poll_shares_result.md)
• Result: a lone partisan poll's limit (TODO H90 part 4), 2026-10-28 (house_lone_partisan_result.md)
• Result: the House poll update's variance split (2026-10-25): NOT ADOPTED (house_poll_variance_result.md)
• District record (structure + polls, no raters), 1562 contested districts 2018-2024 (house_primary_polls_result.md)
• Result: House leans blended toward the raters (2026-10-08) (house_rater_blend_result.md)
• Result: the House district error scale (2026-10-12) (house_spread_result.md)
• Result: stretching the House model's district leans (2026-10-08) (house_stretch_result.md)
• Result: the tiered rater blend (2026-10-20) — FAILED, not adopted (house_tiered_blend_result.md)
• Result: independents' prior from a contest of the same kind (2026-10-24) (independent_prior_result.md)
• Result: likely-voter adjustment of registered-voter and adult polls — FAIL (2026-10-12) (lv_rv_adjustment_result.md)
• Result · three checks of the national model against the forecasting literature (2026-10-15) (national_literature_checks_result.md)
• Result: the November round in the champion's start prior (2026-10-24): ADOPTED (november_round_prior_result.md)
• Result: one round, November, for the statewide panel's nominee runoffs (2026-10-24) (november_round_result.md)
• Result T1: off-year reversion by place — FAIL (evaluated once on 2025, 2026-10-05) (offyear_reversion_result.md)
• Result: national-origin swing tests (T2a, T2b, T2c) (origin_swings_result.md)
• Result: a measured spread for the House seats forecast from their own polls (TODO H122), 2026-10-28 (own_polls_band_result.md)
• Result: the House t(4) shape for the seats forecast from their own polls (TODO H122, second test), 2026-10-28 (own_polls_shape_result.md)
• PAC money in open seats: result (2026-10-25): NOT ADOPTED (pac_open_seat_result.md)
• Result: leaving out partisan-sponsored polls, scored on past elections (2026-10-23) (partisan_polls_backtest_2026_10_23_result.md)
• Result: the late trend in party money does not predict House results beyond the forecast (party_money_trend_result.md)
• Result · Does anything besides approval explain the polling miss? A screen (poll_miss_screen_result.md)
• Press-read result: can public news reproduce what the raters know? (run 2026-10-26) (press_read_result.md)
• Result: the statewide prior on the all-vote scale (2026-10-24): ADOPTED (prior_all_vote_result.md)
• Result: the statewide prior's spread by class of predicted margin (2026-10-24): ADOPTED (prior_sd_by_margin_result.md)
• Result: the statewide prior's width without the national swing (2026-10-24): ADOPTED (prior_width_net_result.md)
• Result: retrying refused fits in production (2026-10-07) (production_retry_result.md)
• Result: the raters alone against the model alone (2026-10-20) (raters_vs_model_result.md)
• Promotion case: ranked choice with the race model's tail (ELAB_RCV_STUDENT_T), ranked_choice_final_round 1.1.0 -> 1.2.0 (rcv_student_t_promotion_result.md)
• Result: backtests with real poll release dates (audit F09, 2026-10-04) (release_dates_result.md)
• Result: one religion-census vintage throughout (TODO H128 item 2), 2026-10-28 (religion_vintage_result.md)
• The statewide scandal rule's poll limit: blend 0.8.7 on the 2026 record (FM-255) (scandal_polled_limit_result.md)
• Result: a Senate seat's previous result from its own seat class (2026-10-24) (senate_seat_class_prior_result.md)
• The persistent scandal shift, scored on 2026 (shock_shift_prospective_result.md)
• Result: the shock centre-shift rule (shock_shift_result.md)
• Result: the shock uncertainty rule (shock_uncertainty_result.md)
• Result: a sparse statewide race counts its prior once (2026-10-28): PASSED ITS RULE, not yet adopted (sparse_prior_once_result.md)
• Result: sparse races and the second prior (2026-10-24) (sparse_second_prior_result.md)
• Result: the sparse blend counts shared polling error once (2026-10-25): ADOPTED BY THE OWNER'S AMENDMENT (sparse_shared_once_result.md)
• Result: pooling combined sponsor/firm pollster records (2026-10-10) (sponsor_pooling_result.md)
• Result · `stateleg_baseline/0.1` gate: FAILED (2026-10-13) (stateleg_baseline_result.md)
• Result · formal tests of the three added components (2026-10-15) (stateleg_components_result.md)
• Result · third candidates in multi-member districts (`ELAB_MM_THIRD`, H114) (stateleg_mm_third_result.md)
• Result · H41: state legislative districts move about 0.65 points per point of national change (2026-10-22) (stateleg_national_response_result.md)
• Result · `stateleg_swing/0.10` (2026-10-15) · published as the initial version (stateleg_swing010_result.md)
• Result · `stateleg_swing/0.11` (pre-registered 2026-10-15; `stateleg_swing011_preregistration.md`) (stateleg_swing011_result.md)
• Result · `stateleg_swing/0.3` gate: FAILED on condition 6 (2026-10-13) (stateleg_swing03_result.md)
• Result · `stateleg_swing/0.4` gate on 2020: FAILED on conditions 2 and 3 (2026-10-13) (stateleg_swing04_result.md)
• Result · `stateleg_swing/0.5`: FAILED conditions 2 and 4 (2026-10-13) (stateleg_swing05_result.md)
• Result · `stateleg_swing/0.6`: FAILED conditions 2 and 4 (2026-10-14) (stateleg_swing06_result.md)
• Result · `stateleg_swing/0.7`: FAILED conditions 2 and 4 (2026-10-14) (stateleg_swing07_result.md)
• Result · `stateleg_swing/0.8` (2026-10-15) (stateleg_swing08_result.md)
• Result · `stateleg_swing/0.9` (2026-10-15) (stateleg_swing09_result.md)
• Result · `stateleg_swing/0.2` gate: FAILED (2026-10-13) (stateleg_swing_result.md)
• Result: the statewide champion on the races its gate never saw (2026-10-02) (statewide_backfill_check_result.md)
• Statewide gate result and adversarial review (2026-10-01) (statewide_gate_result.md)
• Result: the statewide spread terms, second run on corrected data (2026-10-23) (statewide_spread_rerun_2026_10_23_result.md)
• Result: the statewide spread terms on today's data, and the drift law (2026-10-12) (statewide_spread_result.md)
• Result: the blend weight and state poll misses (2026-10-08) (statewide_weights_result.md)
• Result: the poll likelihood's StudentT scale (2026-10-24): NOT ADOPTED (student_t_scale_result.md)
• Result: governors and state legislatures on one national draw (trifectas) (trifecta_joint_result.md)
• Result: the races that cannot be backtested (unbacktestable_races_result.md)
• Result: the uncertainty-terms refit, A/B as registered (run 2026-10-26) (uncertainty_terms_refit_result.md)
• Result: volatility gaps need eight distinct states (M-19), 2026-10-27 (volatility_min_states_result.md)