Transparency

Pre-registration & audit log

This page is for forecasters, journalists, and researchers who want to verify or replicate any of our predictions. Every forecast file is sha256-committed before publication, every assumption is machine-readable, and every input source is timestamped.

Latest forecast manifest

General election, next-GE-as-if-held-this-week

Each row is a downloadable JSON file with a sha256 commitment to its contents.

FileSnapshot IDGenerated at (UTC)
data/predictions/ge-next/constituencies.json ge-next-2026-05-27 2026-05-27 04:30:42
data/predictions/ge-next/summary.json ge-next-2026-05-27 2026-05-27 04:30:42
data/predictions/ge-next/assumptions.json ge-next-assumptions-2026-05-27 2026-05-27 04:30:42
data/identity/pcons-ge-next.json pcons-ge-next-2026-05-27 2026-05-27 04:30:38
data/polling/override.json auto-refreshed daily see file retrieved_at

Input sources

What feeds the forecast

A plain-English version of the same list lives at Sources.

Assumptions

Every parameter, machine-readable

{
  "use_strong_transition_model": true,
  "ge_dampening": 1,
  "bes_prior_weight": 0.15,
  "tactical_voting_competitiveness_gap": 0.1,
  "tactical_voting_floor": 0.05,
  "tactical_voting_transfer_rate": 0.3,
  "by_election_overlay_weight": 0.3,
  "incumbency_personal_vote_pp": {
    ">=20yr": 4,
    ">=10yr": 3,
    ">=5yr": 2,
    "<5yr": 1
  },
  "retirement_drag_pp": {
    ">=20yr": 3,
    ">=10yr": 2.5,
    "<10yr": 2
  },
  "defection_open_seat_drag_pp": 1.5
}

Download the raw JSON: assumptions.json

Coverage

Identity-table coverage

Audit standards

What "audit-grade" means

  1. Pre-registration. Every forecast file's sha256 is computed and recorded before publication; subsequent edits produce a new snapshot ID.
  2. Source manifest. Every input file lists path, retrieved-at timestamp, and licence.
  3. Machine-readable assumptions. Every weight, dampening, ceiling, and threshold lives in assumptions.json so any third party can replicate or stress-test by tweaking one value at a time.
  4. Open data. All prediction JSON files downloadable directly from the site under CC-BY.
  5. Open methodology. Every prediction page shows the full methodology trace; the model code is public at github.com/tompickup23/ukelections.

Targets: Royal Statistical Society Code of Practice for forecasters · British Polling Council membership criteria.