49 seats for a majority of the 96-seat Senedd. Ranges are 5th–95th percentile across the simulations.
These entries needn’t total 100%. The model derives an “others” share from them and rescales everything to 100%, so figures in National vote share above won’t match these exactly: below the 2026 total of about 98% they come out a little higher, above it a little lower, and the further the total sits from 98% the wider the gap.
| Constituency | Vote share | Elected members (6) | Next seat |
|---|
Each pollster carries a score from their historic accuracy, shown 0–100 in the coloured circle beside their name. A grey circle (50) is a pollster with no track record yet, scored as an average 3-point error.
So a high-volume pollster can't dominate the average by sheer quantity, a firm publishing several times inside a 10-day window has its weight shared across those polls.
The trendline is a tricube-weighted local average over a 56-day window — wider than the GB page’s 35 days, because Senedd polling is sparse enough that a narrow window would let a single poll swing the average. It is anchored on the actual 7 May 2026 Senedd result, so the line starts from a known point rather than drifting in from nothing.
A two-pass robust fit: the first pass measures how far each poll sits from the trend, then re-weights it down — to zero for extreme outliers — before fitting again.