Tactical voting for the next General Election: How to block Labour

Method

Poll-of-polls: choose a 15- or 30-day rolling window and the arithmetic mean or median. Each party is aggregated across polls reporting that party, then shares are normalised to 100%. Commissioner checkboxes remove that client’s polls from the entire series before cleaning. Mean is the default. The outlier detection rule always uses medians. The detection median combines older accepted polls within the selected window with all polls from the latest seven calendar days, including previously rejected polls and the complete current-day batch. Each poll is counted once. This lets recent changes influence the median without automatically accepting every recent poll. Polls are still screened using the ±15% Labour/Conservative/Reform rule, and accepted polls form the cleaned average. Uncleaned uses every poll. No double dipping uses only the newest accepted poll per pollster for its average and older median inputs; all polls from the latest week still contribute to its detection median. When Restore is not reported separately, the reported Other figure is split 50/50 between Restore and Other.

Election projection: each 2024 constituency party vote is multiplied by the national proportional-swing factor. Completed Westminster by-elections since the 2024 general election are used as a third local observation: for parties contesting both elections, the local 2024→by-election movement is divided by the same period's cleaned 30-day national polling movement, then carried forward as a constituency-specific velocity correction. Parties absent from a by-election keep their 2024 baseline and by-election novelty/Other candidates are not extrapolated. Tactical and pact calculations are then performed server-side using the existing 30% transfer and historical-route heuristics. The optional Restore pact redirects Restore support to Reform nationally, retaining up to 0.7 percentage points as a scenario assumption, and allows Reform-only transfers to Restore in the listed target seats.