Model Changelog

12 March 2026

Fix Reform swing calibration for councils with 2024 local results

Councils where Reform stood in the 2024 local elections were applying an undampened national swing to their local baselines, over-projecting Reform in metropolitan boroughs. The model now applies the same dampening framework used for other data sources, calibrated for local-to-local conversion. This most affects metropolitan boroughs like Sunderland and Barnsley where the undampened swing was producing high Reform projections. Sunderland Reform falls from 49 to 33 seats.

12 March 2026

Fix By-election data now overrides existing baselines

Previously, by-election results could only set Reform's baseline in wards where Reform had not stood before. If a ward already had a Reform result from the 2024 local elections (even at just 5-10%), the by-election data was ignored - even when it showed dramatically different results. This was most visible in Sunderland's Hetton ward, where the November 2025 by-election showed Reform at 45% but the model was stuck at the 2024 baseline of 11%. By-election results now blend directly into the ward baseline (70% by-election, 30% existing) regardless of prior Reform presence, and receive zero additional swing since they already reflect current political conditions.

12 March 2026

Data 43 by-elections from October-December 2024 added

Added 43 by-election results from October to December 2024 to the model's Tier 1 evidence base. These include Reform gains in Bilston North (Wolverhampton, 35%), Blackbrook (St Helens, 41%), and strong Reform performances in Woodhouse (Sheffield, 33%), Brockmoor and Pensnett (Dudley, 30%), and Featherstone (Wakefield, 22%). Where by-election ward names match 2026 ward boundaries, the results are blended into the baseline as the most reliable local evidence available.

12 March 2026

Data 2025 county council results blended into 28 district councils

District councils that sit inside county council areas have no recent Reform election data of their own - Reform did not stand in any of these districts in 2022, 2023 or 2024. Until now, these councils relied entirely on the model's entry estimate, which can miss local variation. We have now used actual 2025 county council division results to set ward-level baselines for all parties in 28 district councils across 10 counties. Each district ward is mapped to the county division it falls within, and the full county party shares are blended into the ward baseline. The 28 councils span Staffordshire, Lancashire, Kent, Warwickshire, Worcestershire, Hertfordshire, Lincolnshire, Cambridgeshire, Oxfordshire, Gloucestershire, and Devon.

12 March 2026

Fix Reform projection ceiling from county evidence

In county-blended district councils, the combination of county evidence and national swing adjustments was pushing some party projections above the actual county division results they were anchored to. A ceiling now prevents any ward's projection for Reform, Green, or Liberal Democrats from exceeding the county division actual result by more than 3 percentage points after all adjustments. The small headroom allows for genuine within-division variation. This prevents runaway single-party dominance in blended wards while preserving the county signal.

12 March 2026

Fix Ward maps rendering correctly

Ward maps on individual council pages were showing hatched (unmapped) areas in several councils. Two issues were causing this: ward boundary codes were not being passed through to the map data, and councils with minor ward name changes (renames or thirds-rotation names) were not building the name bridge needed to match wards to their map polygons. Both issues are now fixed, and maps render correctly across all councils including those with recent boundary changes.

11 March 2026

Fix Improved boundary change handling across 25+ councils

Comprehensive review of how the model handles ward boundary changes across all councils with LGBCE reviews. Where new ward boundaries cross old ward boundaries, the model needs to estimate the political character of the new ward by blending results from the old wards it overlaps. We rebuilt overlap matrices from scratch using GIS boundary data for 12 metropolitan and unitary boroughs (Barnsley, Bradford, Coventry, Gateshead, Kirklees, Newcastle upon Tyne, Sandwell, Sefton, Solihull, South Tyneside, Sunderland and Swindon), verified and corrected the Milton Keynes matrix manually, and fixed Calderdale's manual remap where Halifax Town ward was being treated as a simple rename rather than a new creation drawing from several old wards. A further 9 councils with minor boundary changes (1-3 ward renames or splits) had their name mappings corrected. Combined with the county GIS rebuilds from 5-6 March, boundary handling has now been reviewed for every council with an LGBCE boundary change.

11 March 2026

Fix Milton Keynes ward sources corrected

Several Milton Keynes wards were drawing their baseline from the wrong predecessor wards. The new Walton ward was missing its connection to old Monkston - where the Liberal Democrats polled 44% - making it appear as a safe Conservative ward rather than the genuine three-way contest it should be. Great Linford was similarly drawing from only one predecessor instead of four.

11 March 2026

Fix Swindon ward merge using incomplete data

Two new Swindon wards (Chiseldon & Ridgeway and Old Town & Lawn) were being constructed from only one of their two predecessor wards. The missing predecessor had been abolished in the 2024 boundary change and was not being loaded from historical data, making Chiseldon & Ridgeway appear as a safe Conservative ward when it should be a Lab/Con marginal.

11 March 2026

Fix Thurrock boundary remap corrections

Three Thurrock wards (Little Thurrock Rectory, Stifford, Tilbury Riverside) were inheriting vote share baselines from the wrong predecessor wards after the 2024 boundary review. The worst case - Tilbury Riverside - was showing as a safe Labour ward when the actual predecessor area had a strong independent and Reform presence. Source mappings have been corrected using the LGBCE Final Recommendations report.

11 March 2026

Fix Walsall ward name matching

One Walsall ward (Rushall-Shelfield) was being silently dropped due to a hyphen in the ward name not being normalised. The ward is now correctly included, bringing Walsall from 19 to 20 wards.

11 March 2026

Fix County council blend capped and scoped

The blend of recent district council results into county baselines is now capped at 5 percentage points per division, preventing small systematic shifts from compounding across a whole county. Hampshire and West Sussex are excluded from the blend entirely - their 2021 baselines were structurally low for Labour due to weak campaigns that year, making the district data a poor correction signal.

11 March 2026

Fix Identical ward predictions after boundary splits

When a ward is split into two new wards under boundary changes, both inherited identical vote shares from the parent. This affected 10 pairs of wards across Milton Keynes, Essex, Suffolk, Swindon, Thurrock, West Oxfordshire and Winchester. Each pair now has differentiated baselines using LGBCE electorate data and blending with adjacent wards. The largest correction was in Milton Keynes, where Great Linford and New Bradwell now differ by nearly 6 percentage points.

10 March 2026

Fix Independent councillor swing protection

Independent councillors with a meaningful baseline (above 5%) now have the national Others swing zeroed out. Independents are local brands whose support is driven by personal vote, not national trends - applying a uniform swing was systematically eroding their projected vote share. This partially addresses the model's systematic under-prediction of Others seats, recovering around 30 seats nationally.

10 March 2026

Fix Reform under-predicted in high-Leave outer London

In outer London boroughs with high Leave vote (above 55%), the Reform entry model was constrained by mismatched draw proportions. The model allocates Reform's vote share by drawing from other parties, but in wards dominated by a single party (e.g. Labour at 48%, Conservatives at 2%), the fixed draw proportions left much of the allocation stranded. A targeted second pass now redistributes unused allocation from parties with headroom, raising Reform in boroughs like Barking and Dagenham to levels consistent with local polling and by-election evidence.

10 March 2026

Fix Multi-member ward baseline normalisation

In wards electing two or three councillors, baseline vote shares were not fully adjusted for the number of candidates each party fielded. Parties running a full slate appeared stronger in the baseline than their per-voter support warranted. Baselines are now normalised per candidate before computing shares, improving accuracy in London boroughs and metropolitan boroughs with multi-member wards.

10 March 2026

Fix County council calibration

Reform's county council baselines were slightly over-calibrated due to overlapping adjustments in the swing formula. The calibration has been simplified and re-tested against 2025 county results. Conservative projections in eastern counties remain low, reflecting the scale of national Conservative decline and Reform's strong local entry - consistent with what we observed in comparable councils in 2025.

10 March 2026

Fix Multi-member seat allocation

Seat allocation in multi-member wards now splits seats more readily in tight two-party contests, preventing cases where a narrow lead translated into a clean sweep of all seats in the ward.

10 March 2026

New Candidacy flags for Green projections

Councils where the Green projection relies on candidacy assumptions are now flagged on their projection page. In some boroughs the Greens are projected to gain seats from a low baseline, and these projections will be revised when April nomination data confirms which wards have Green candidates.

7 - 8 March 2026

Fix Reform over-estimated in boundary-change councils

Wards with new boundaries (splits, renames, merges) were missing their GE2024 Reform result because the lookup used the new ward name instead of the old one. This caused them to fall back to a council-wide MRP average - roughly double the correct ward-level target. A name bridge now maps new ward names back to their pre-boundary predecessors for GE2024 data. Reform drops around 21 contested seats nationally as a result, with Labour and Conservatives gaining.

7 March 2026

Fix Ward holder now based on seats won

In multi-member wards, the incumbent party was previously determined by vote share. This meant a party could top the vote count but win fewer seats - and the wrong party would receive the incumbency bonus and swing resistance. The model now counts seats won (using the DCLEAPIL elected column) and assigns the holder accordingly. This affects around 113 wards where the seat winner differs from the vote leader.

7 March 2026

Fix All-out to by-thirds baseline normalisation

Metropolitan boroughs that held all-out elections in 2024 but return to by-thirds in 2026 needed their baselines adjusted. In an all-out election, voters cast multiple votes and full-slate parties accumulate more raw vote share. A slate-weighted blend now normalises these baselines: full-slate parties are averaged per candidate, while single-candidate parties are blended between raw and averaged shares proportional to their slate completeness. This prevents single-candidate opposition parties from being artificially inflated.

7 March 2026

Fix Kirklees ward map corrected

Colne Valley East and West were mapped to the wrong polygons on the Kirklees map. The boundary remap had the old wards assigned the wrong way round - confirmed against the LGBCE final recommendation shapefiles. Both the map and the underlying data assignment are now correct.

7 March 2026

Fix Historical vote shares corrected for multi-member wards

Ward explorer pages were displaying per-candidate percentages from multi-member elections rather than summed party totals. In wards like Wandle (Merton), this made the Lib Dems appear ahead of Labour when Labour actually received more votes overall. All cached ward history data has been regenerated with correct party-total vote shares.

5 - 6 March 2026

Fix County council boundary remapping

Six county councils have new ward boundaries for 2026. We built overlap matrices matching old wards to new ones, weighted by electorate size. This means the model correctly distributes historical vote shares across redrawn boundaries in Essex, Norfolk, Suffolk, Hampshire, East Sussex and West Sussex.

5 March 2026

Fix Multi-member ward seat allocation

Wards electing two or three councillors at once now allocate seats more realistically. Previously the model gave all seats to the winning party. Now, when two parties are close in vote share, seats are split between them - reflecting how multi-member wards typically behave in practice.

5 March 2026

Fix Green over-prediction in central London

The model was over-predicting Greens in heavily Remain-voting inner London boroughs. A calibration adjustment now moderates Green predictions in areas where their local election performance has historically fallen short of what demographics alone would suggest.

4 March 2026

New London inner/outer split

London boroughs are now modelled as two distinct groups: inner London (13 boroughs) and outer London (19 boroughs). These areas vote very differently - inner London is more pro-Labour and less favourable to Reform, while outer London behaves more like suburban England. Splitting them improves accuracy for both.

3 - 4 March 2026

Data Calibration against 2025 district council results

We incorporated 1,400 ward results from the May 2025 district council elections - the most recent local data available. This improved the model's swing calibration, particularly for Reform UK whose 2025 results provide a much better baseline than Westminster polling alone.

2 - 3 March 2026

Fix Reform double-counting removed

In wards where Reform had both a direct baseline (from a by-election or 2024 result) and an estimated baseline from MRP modelling, both were being applied. This inflated Reform's projected vote share in affected wards. Now the model uses the best single source for each ward and does not double up.

1 - 2 March 2026

Fix Candidacy thresholds for Greens and Lib Dems

In wards where the Greens or Lib Dems had very low baselines (below 3%), the model now treats the party as unlikely to be standing and reallocates their projected vote share to other parties. This prevents the model from predicting seats for parties that probably will not field a candidate.

28 February 2026

Fix UKIP baselines no longer treated as Reform

Some older ward results included a UKIP baseline from 2018 or earlier. These were being carried forward as Reform proxies, but UKIP's 2018 vote share is a poor guide to Reform's 2026 support. Wards where UKIP polled below 5% now use the standard Reform entry model instead.

27 February 2026

New Incumbency advantage

Sitting councillors tend to outperform their party's national swing. The model now gives a small bonus to the party that currently holds each ward, calibrated from historical results. The bonus is larger for Lib Dems and Greens, who tend to have stronger personal votes, and smaller for Reform, who have very few incumbents.

25 - 26 February 2026

New Ward-level model launched

The original model used a council-level cube law to convert vote shares into seats. This has been replaced by a ward-by-ward model that predicts individual ward winners and adds them up. This is a fundamentally more accurate approach, especially in councils with a mix of safe and marginal wards.

Expected April 2026

Upcoming Actual candidate lists

Democracy Club publishes candidate lists from the Statements of Persons Nominated in mid-April. This will tell us exactly which parties are standing in each ward, replacing our current estimates. Green and Lib Dem seat counts are likely to adjust downward in wards where they are not fielding a candidate.

Completed 11 March 2026

Done District council data for county predictions

Recent district council results have been blended into county baselines for Essex, Norfolk, Suffolk and East Sussex, giving fresher data than the 2021 county elections alone. Hampshire and West Sussex were excluded after testing showed the blend introduced more noise than signal in those counties.

Expected March 2026

Upcoming County blend for remaining district councils

12 further district councils could benefit from 2025 county council data blending, though these are in areas where Reform polled lower (18-26%) so the impact will be smaller. The remaining councils are: St Albans, Three Rivers, Watford, Welwyn Hatfield (Hertfordshire), Cambridge, Huntingdonshire, South Cambridgeshire (Cambridgeshire), Cherwell, Oxford, West Oxfordshire (Oxfordshire), Cheltenham (Gloucestershire) and Exeter (Devon). Districts inside Essex, Norfolk, Suffolk and East Sussex cannot be blended as their 2025 county elections were postponed due to local government reorganisation.

Expected March 2026

Upcoming Further metropolitan borough verification

The all-out to by-thirds transition has been addressed, but individual met boroughs are being spot-checked against local knowledge. Candidate data (expected April) will further refine predictions in these areas.

Late April 2026

Upcoming Final pre-election forecast

A final forecast incorporating complete candidacy data, late polling, and any remaining model improvements before election day on 7 May.