Changelog
Reverse-chronological log of model versions, methodology updates, bug fixes, and feature launches. Every entry links back to the responsible code commit when applicable. Builds trust the only way that scales: by being auditable.
Recent
2026-06-04 | model
lr-2026-06-04-A-complete published: open-seat encoding + candidate features
Publishable model rebuilt as the A spec: 14 features with explicit open-seat encoding (a retiring incumbent sets incumbent_party_d to 0.5 and zeroes the economy-by-incumbent interaction) plus a seat-level personal-vote spread. Fit on the 2020 and 2022 senate and governor races and tested out-of-sample on 2024: test_brier 0.1370 on that holdout against a 0.2500 coin-flip baseline. Per-cycle Brier is reported across the 2014-2024 backtest corpus. The candidate-level carryover spread (model B) was tested and benched: out-of-sample Brier gain was +0.0001 versus a +0.0020 inclusion gate, with a coefficient stable in sign across three folds. A benched-features registry now retests benched candidates automatically whenever a new race outcome resolves; promotion to the publishable selector stays manual.
2026-06-01 | model
v11d shipped: CivicTrust formula corrected, continuous state_lean
Registration-rate formula fixed to count voters as presumed-registered (per CPS skip pattern). Turnout source switched from CPS turnout_delta to MIT EL absolute senate turnout. state_lean continuous from 2020+2024 presidential margins (replaces binary +1/-1). Backtest corpus extended back to 2014. test_brier 0.135. Retired v11c (0.114) as broken-data ceiling and all pre-v11d variants.
2026-06-01 | feature
Albers USA SVG map replaces tile-grid choropleth
Real state shapes from us-atlas TopoJSON. AK / HI insets. Click-through to /race/:race_id on senate and governor maps.
2026-06-01 | feature
VoteROI.com Mood Forecast logo + shared site chrome
Final brand mark selected from three concepts. Header, footer, favicon, and OG cards unified.
2026-05-31 | model
v10 -> v11 series: Information_candidates clip and incumbent interaction audit
Information dimension candidate-quality sub-feature clipped for uncontested seats. Brier 0.137 -> 0.135 on 2024 holdout.
2026-05-30 | feature
Forecast page redesign + 10 inline graphics
Section structure: Hero National Index, Top 5 Presidential, Race Maps, Movers, Mood Rankings, State Composite Map, Dimension by Race Heatmap.
Model versions (live from database)
| Version | Test Brier | Train Brier | Features |
|---|---|---|---|
lr-house-2026-06-06-twospeed-v3 |
0.0583 | 0.0467 | 5 |
lr-house-2026-06-06-twospeed-v2 |
0.0986 | 0.0638 | 4 |
lr-house-2026-06-06-twospeed |
0.0628 | 0.0714 | 3 |
lr-2026-06-11-chal-serve |
0.1084 | 0.1086 | 17 |
lr-2026-06-11-chal-h2024 |
0.1140 | 0.1054 | 17 |
lr-2026-06-11-chal-h2022 |
0.1317 | 0.1460 | 17 |
lr-2026-06-11-chal-h2020 |
0.1381 | 0.1592 | 17 |
lr-2026-06-11-chal-h2018 |
0.1988 | 0.1387 | 17 |
lr-2026-06-11-chal-h2016 |
0.1557 | 0.1541 | 17 |
lr-2026-06-11-chal-h2014 |
0.1639 | 0.1719 | 17 |
lr-2026-06-10-chal-recal-twospeed |
0.1191 | - | 17 |
lr-2026-06-09-incfull-twospeed |
0.1519 | - | 15 |
lr-2026-06-09-incfull-serve |
0.1128 | 0.1115 | 15 |
lr-2026-06-09-incfull-selrecal-twospeed |
0.1205 | - | 15 |
lr-2026-06-09-incfull-recal-twospeed |
0.1189 | - | 15 |
lr-2026-06-09-incfull-h2024 |
0.1190 | 0.1079 | 15 |
lr-2026-06-09-incfull-h2022 |
0.1330 | 0.1514 | 15 |
lr-2026-06-09-incfull-h2020 |
0.1433 | 0.1675 | 15 |
lr-2026-06-09-incfull-h2018 |
0.2049 | 0.1443 | 15 |
lr-2026-06-09-incfull-h2016 |
0.1601 | 0.1600 | 15 |
lr-2026-06-09-incfull-h2014 |
0.1681 | 0.1783 | 15 |
lr-2026-06-09-house-recal |
0.0488 | - | 4 |
lr-2026-06-08-recal-w24-h2024 |
0.1381 | 0.1252 | 15 |
lr-2026-06-08-recal-w24-h2022 |
0.1576 | 0.1722 | 15 |
lr-2026-06-08-recal-w24-h2020 |
0.1754 | 0.1943 | 15 |
lr-2026-06-08-recal-w24-h2018 |
0.2368 | 0.1744 | 15 |
lr-2026-06-08-recal-w24-h2016 |
0.2066 | 0.2026 | 15 |
lr-2026-06-08-recal-w24-h2014 |
0.2072 | 0.2042 | 15 |
lr-2026-06-08-recal-twospeed |
0.1291 | - | 15 |
lr-2026-06-08-db1-serve |
0.1293 | 0.1301 | 15 |
lr-2026-06-08-db1-h2024 |
0.1397 | 0.1282 | 15 |
lr-2026-06-08-db1-h2022 |
0.1594 | 0.1670 | 15 |
lr-2026-06-08-db1-h2020 |
0.1818 | 0.1893 | 15 |
lr-2026-06-08-db1-h2018 |
0.2314 | 0.1791 | 15 |
lr-2026-06-08-db1-h2016 |
0.2031 | 0.1914 | 15 |
lr-2026-06-08-db1-h2014 |
0.2085 | 0.2075 | 15 |
lr-2026-06-08-db07-serve |
0.1286 | 0.1295 | 15 |
lr-2026-06-08-db07-h2024 |
0.1388 | 0.1273 | 15 |
lr-2026-06-08-db07-h2022 |
0.1586 | 0.1676 | 15 |
lr-2026-06-08-db07-h2020 |
0.1791 | 0.1903 | 15 |
lr-2026-06-08-db07-h2018 |
0.2328 | 0.1772 | 15 |
lr-2026-06-08-db07-h2016 |
0.2029 | 0.1914 | 15 |
lr-2026-06-08-db07-h2014 |
0.2060 | 0.2041 | 15 |
lr-2026-06-08-cq-serve |
0.1127 | 0.1133 | 17 |
lr-2026-06-08-cq-h2024 |
0.1190 | 0.1099 | 17 |
lr-2026-06-08-cq-h2022 |
0.1364 | 0.1509 | 17 |
lr-2026-06-08-cq-h2020 |
0.1406 | 0.1675 | 17 |
lr-2026-06-08-cq-h2018 |
0.2049 | 0.1440 | 17 |
lr-2026-06-08-cq-h2016 |
0.1702 | 0.1800 | 17 |
lr-2026-06-08-cq-h2014 |
0.2068 | 0.2039 | 17 |