WESTMINSTER RADAR
Menu
Back to feed

Methodology

Evidence before inference.

Westminster Radar makes published declarations easier to inspect. It does not score influence, allege wrongdoing or claim causation.

Sourced fact

Directly reported in the cited public record.

Explanatory context

Process or background that helps interpret the record.

Analytical inference

A labelled calculation or match, never presented as source wording.

How we use automation and AI tools

Most of Westminster Radar is built automatically from official data: Parliament, IPSA, the Electoral Commission and the ONS. Our software checks for new records every day. You can see the records once we have checked and added them.

We use AI tools to help draft guides and the plain-English summaries of Commons votes. Every summary is checked against the official motion or explanatory papers before it's published, and each one links to its source. AI tools don't decide what records are included or how an MP is described.

If you find a mistake, please use the corrections form.

Topic voting summaries

These summaries use our own explicit division mappings, separate from the keyword topics used to browse activity. Each approved package has three substantive decisions: a reasoned amendment refusing Second Reading (No counts as for), Second Reading (Aye counts as for), and Third Reading (Aye counts as for). Procedural divisions, incidental amendments, enabling money or ways-and-means motions and non-binding Opposition day motions are excluded.

Three divisions on one Bill, including two on the same day, are not three independent policy decisions. Each division has equal weight; the Bill can change between stages. A package summary does not establish the MP’s stance on an entire policy field or every Bill provision.

Calculation and bands

For share is recorded votes on the mapped for side divided by recorded Aye/No votes in the topic. No recorded votes are excluded; unavailable lists are not treated as absences. Named tellers count once in topic summaries and a proxy counts once for the represented MP. This differs from the separate against-party and two-MP agreement measures, which exclude tellers.

  • 100%: Consistently voted for.
  • 75% to below 100%: Mostly voted for.
  • Above 25% to below 75%: Voted a mixture of for and against.
  • Above 0% to 25%: Mostly voted against.
  • 0%: Consistently voted against.
  • Fewer than two recorded divisions: Not enough votes to say.

Use unrounded shares. With only three divisions, a 2–1 split is a mixture and neither mostly band is possible.

Party comparator

For each included division, calculate the fraction of that party’s named recorded votes on the mapped for side, then average these fractions equally across the same complete division set. Use the party recorded on the day and include the subject MP and named tellers once. This mean is distinct from the Commons medians on activity dashboards. If a division has no party votes, withhold its like-for-like comparator. MPs who changed party receive separate labelled comparators with the dates of their recorded party membership. Independents have no party comparator.

Evidence and publication gate

Every named list must reconcile with its official Aye/No totals after excluding tellers from those totals; member IDs must be unique, and the official title and date must match the approved mapping. Proxy votes are already in named records and are never added again. A missing or quarantined list blocks the whole topic, its summaries, links and sitemap entries. We never invent missing voters or parties. Data withheld for quality is not labelled “Not enough votes to say”. Member topic pages require at least two recorded votes.

Approved division mappings

Establishing Great British Energy — Bill package

Passage of the Great British Energy Bill as presented at each recorded stage; this is not a score for climate policy, energy prices or every provision of the Bill.

Official debate confirming the reasoned refusal

Employment law changes — Employment Rights Bill package

Passage of the Employment Rights Bill as presented at each recorded stage; this does not establish support for each individual employment measure or later changes.

Official debate confirming the reasoned refusal

Employer National Insurance — 2024–25 Bill package

Passage of the National Insurance Contributions (Secondary Class 1 Contributions) Bill; rates, thresholds and the Employment Allowance are a package, not a general view on taxation.

Official debate confirming the reasoned refusal

Transfer of apprenticeship and technical education functions

Passage of the Bill transferring the Institute for Apprenticeships and Technical Education functions; opposition is not opposition to apprenticeships or technical education.

Official debate confirming the reasoned refusal

Product regulation and metrology powers — Bill package

Passage of the Product Regulation and Metrology Bill; this does not assign a general position on European relations or consumer protection.

Official debate confirming the reasoned refusal

Border security, asylum and immigration — 2025 Bill package

Passage of the Border Security, Asylum and Immigration Bill; this is not a general measure of attitudes towards immigration, asylum seekers or border security.

Official debate confirming the reasoned refusal

Football governance — independent regulation Bill package

Passage of the Football Governance Bill; this does not claim that every component of the regulatory framework was supported.

Official debate confirming the reasoned refusal

English devolution and community powers — Bill package

Passage of the English Devolution and Community Empowerment Bill as presented at each stage; a vote against the package does not establish opposition to devolution in principle.

Official debate confirming the reasoned refusal

Sentencing changes — 2025 Bill package

Passage of the Sentencing Bill as presented at each stage; this is not a general score for criminal justice policy or sentence severity.

Official debate confirming the reasoned refusal

Diego Garcia and the Chagos Archipelago — treaty implementation

Passage of the Diego Garcia Military Base and British Indian Ocean Territory Bill. British Indian Ocean Territory and Chagos Archipelago are both named; the summary covers this treaty-implementation package, not a general position on sovereignty or defence.

Official debate confirming the reasoned refusal

Change log

— initial ten-package selection and calculation rules approved. No prior public topic-band mapping existed. Future revisions record their date, old and new mappings, official evidence and reason. Editorial approval does not override the source-reconciliation gate.

Comparing two MPs

Agreement covers divisions from 4 Jul 2024 where both MPs have a recorded Aye or No in available named lists. Tellers, no recorded votes and unavailable lists are excluded. Source-confirmed free votes are included and labelled. The denominator is shared recorded divisions, not all Commons divisions; no shared evidence produces no percentage.

Business costs use the same complete IPSA financial year with staffing, office, accommodation, travel and other categories. Different service periods are shown without annualising costs. These are business costs, not personal income. Role notes cover the displayed period and link to official career records.

Ordinary comparisons use two ordered MP slugs as GET parameters, remain noindex, follow and canonicalise to the first MP’s profile. Existing four-MP links remain supported. Only a verified current MP/immediate predecessor comparison for the same official constituency identity can have an indexable clean constituency comparison path, and only where common complete annual costs or shared recorded divisions provide useful evidence. Otherwise the path returns 404 and is absent from links and sitemaps. Query variants remain noindex with a clean canonical.

— two-MP comparison and conditional constituency indexing rules approved.

Richer views: periods, topics and identities

MP vote bars cover the current comparison year beginning on 4 Jul. Party comparisons use the party recorded for the voter in that division and exclude tellers from both sides of the comparison. Confirmed free votes, unavailable lists, independent MPs and tied party votes have no party-majority position. The dedicated against-party page covers stored divisions since 4 Jul 2024 and exists only when at least one qualifying vote is recorded.

Question charts deduplicate by UIN and tabled date and group by the official answering department. Speech and vote topics use the existing published keyword taxonomy; one record can have up to three topic matches, so topic counts need not add to the total. Unclassified records remain visible. Charts state their period; GET filters select the source lists without creating new indexed pages.

Directory House medians and ranges include all current MPs, including zero observations. First-elected filters use the earliest Commons membership date in the Members API snapshot; unknown dates are not assigned to a cohort. Role notes suppress affected comparison positions.

Register timelines compare complete stored editions, grouping main entries by register category. Updated and removed dates are publication dates when the change was observed. Missing declared values are not zero, and different reporting periods and non-cash benefits make declared-value sums unsuitable as income estimates.

Donation recipient matching reuses the rules on the matching report. Donor groups use valid supplied company identifiers, or exact normalised source names plus donor type labelled as unverified identity groups. Ambiguous or anonymous identities remain separate. Local or central accounting units appear only as supplied by the Electoral Commission, with no inferred onward transfer to an MP. The concentration statement is the share of the three largest reported donor groups in the stated period, not a ranking of recipients.

Constituency representation history comes from Members API constituency identities. Changed boundaries can create a new identity; earlier same-name seats are not silently merged.

Dashboards, context and drill-downs

Every highlighted metric links to its report or source rows. Clean MP report pages with published source records are indexable; empty reports and ad-hoc filtered views remain noindex. Native “What this shows and doesn't show” disclosures and chart lists work without JavaScript. A count is not a score of effort, quality, influence or effectiveness.

Comparison strips and role badges

The overview strips show each current MP’s count over the same displayed dates, with party and Commons medians, including zero counts. Overlapping dots remain on the same line; positions represent counts, never ranks. Party labels use the retained current Members API directory. We recalculate medians from stored member counts rather than reusing legacy cached means.

Minister and Whip badges match dated Members API government, opposition and other posts overlapping the displayed period. Minister matches minister, secretary, chancellor, attorney or solicitor general, Lord President, Lord Privy Seal and Leader of the House titles. Whip matches whip, Lord Commissioner of Treasury and Household whip-office titles. They suppress debate and written-question comparisons. Speaker and Deputy Speaker (including Chairman of Ways and Means titles) suppress vote comparisons. Sinn Féin’s abstentionist badge suppresses votes, debates and questions. Badges include the source and explanation; unavailable biographies never result in an inferred role.

A proxy badge requires an official division’s named proxy field within the displayed period, and suppresses vote positioning. It proves a recorded proxy on that date, not a continuing appointment. We do not infer future or unrecorded proxy periods.

Motion type and party-majority comparison

We match the official motion title, case-insensitively, in this order: “second reading”; “third reading”; “amendment”, “amendments” or “amended”; “opposition day”; then “programme”, “business of the House”, “closure”, “adjournment”, “standing order” or “allocation of time” for Procedural. Otherwise the label is Other. This is a title-based aid, not an authoritative legal classification.

“Voted against their party majority” compares named Aye and No votes using the party recorded in that division. Tellers, tied parties, independent MPs, unavailable lists and source-confirmed free votes are excluded. The party vote pattern does not prove a whip instruction. A known free vote is labelled “Free vote — no party position applied”.

No recorded vote means no named vote in the retained list for a member whose retained membership dates overlap the division. It does not establish absence or abstention, or explain the member’s circumstances. The roster combines stored membership dates with a dated official Members API directory snapshot and retained earlier membership periods. Party labels for these non-vote rows are the retained directory affiliation and may differ historically. Quarantined or unavailable lists never generate non-vote counts.

Money and register changes

Canonical main-entry register pages are indexable when they belong to the latest complete edition, have an official source, and retain declared fields or at least 80 characters of summary. Child detail and thin or superseded record views stay noindex to avoid duplicating their parent entries. Sitemap inventory uses the same eligibility rule; query-filtered views always use a clean canonical and noindex, follow.

Money at a glance uses the latest published annual IPSA breakdown with all five categories available, keeping annual totals separate from individual claims. Staffing/payroll, accommodation, travel/subsistence and office wording identify claim categories, in that order; remaining rows are Other. Annual travel totals remain source rows where individual journeys are not published. Filtered cost pages use query parameters, a clean canonical and noindex, follow.

New, changed and removed interests compare entry IDs and retained fields between the latest two complete register editions. Removed means absent from the next edition; reasons are not inferred. Category counts count main entries, not linked child details. Every comparison retains links to the change record and Parliament’s register.

Current coverage

The processor imports every top-level declaration and linked child record in each available House of Commons register archive. Coverage and any incomplete register are shown on the Data page.

Sources

Declaration data comes from the UK Parliament Register of Interests API. An individual interest page requests the current source record and shows every returned field, including fields marked as not supplied; if that request fails, the page falls back to the latest stored copy containing that interest. Member details come from the UK Parliament Members API. Commons division results and individual member votes come from Parliament’s Commons Votes API. Committee roles and events come from Parliament’s Committees API. Staffing and business-cost records come from IPSA’s official CSV publications. Where Parliament supplies a Companies House identifier, the interest page and organisation dossier also read the corresponding public Companies House record. Reported donations and loans come from the Electoral Commission’s Political Finance Online CSV exports.

Refresh and retention

The Data page can start a refresh and shows the last recorded scheduled run, if any. The refresh covers Parliament’s member directory and register catalogue, recent Commons divisions, speeches, written questions, committee memberships, IPSA business costs and the latest 18 months of Electoral Commission donation and loan records. The scheduled handler retries a failed pass once when it is invoked. The refresh also imports one complete historical finance year per successful pass, working backwards from the previous calendar year to 2000. Historical imports do not generate retrospective watchlist alerts.

Provenance, duplicates and identity checks

Every published timeline event records the source organisation, official URL, source publication date, retrieval date and stable source record identifier. Imports upsert on stable keys, so rerunning a source updates the existing event instead of creating a duplicate. If the same stable source record later resolves to a different Parliament member ID, the candidate record is withheld in an identity quarantine for review rather than reassigned automatically. Current source failures, date coverage, record counts and quarantine totals appear on the Data page.

Commons voting records

The Votes page reproduces Parliament’s division title, date and official teller counts. Parliament says the teller headcount can differ from the number of names recorded; each affected division displays both figures and links to the official record, without inventing missing names. Individual Aye, No, teller and proxy records with valid Parliament member IDs appear on MP timelines. A member appearing on both sides is withheld as a conflicting record; other members in that division remain visible. The site does not infer why an MP voted a particular way, and an absent name is not labelled as an abstention.

Rebellions

A rebellion is an Aye or No vote in the opposite lobby to the majority of the MP’s own party who voted in that division. We use the party recorded with the vote, exclude tellers from both the MP count and party majority, and do not count ties, independent MPs or divisions with unavailable voting lists. Known free votes are excluded only when supported by a source linked on the vote. Parliament’s division data does not identify every free vote or whip instruction, so this is a comparison of published votes, not proof of defying a whip. It does not imply motive or wrongdoing.

Snapshot comparisons use the median for current Commons MPs and the MP’s current party over the same dates, including zero counts. They describe published activity and do not measure constituency work, effort or effectiveness. “Above” and “below” compare with the median, not a rank or a percentile. The median is the middle sorted value (the mean of the two central values for an even count). Office cost means include only MPs with published amounts for the financial year; an unpublished amount is not treated as zero. Office costs and declared interest values are kept separate from personal income.

Committees

Current committee roles are linked by Parliament member ID. Recent events are shown as activity by a committee on which the MP serves. They are not labelled as personal attendance unless the official event record identifies that MP as an attendee.

Staffing and business costs

IPSA records are linked using the Parliamentary ID supplied in its downloads. Individual costs retain their reporting year, date, category, cost type, status and paid, unpaid or repaid amounts. Annual totals keep office, staffing, accommodation, travel and other published spending separate. If IPSA rejects a live server request, a verified official release snapshot can be loaded through the owner-only import route; the Data page identifies that source and continues to surface the live-feed failure. These are costs of carrying out parliamentary duties and are not presented as personal income. Security and disability-related information may be withheld or aggregated under IPSA’s publication policy.

Topic classification

Westminster Radar publishes a fixed set of public-policy topics and the keyword rules used for each one. Classification reads the title, summary and retained source fields of an indexed activity. A record can receive up to three topics when several rule sets match. A record that matches no rule remains visible as Unclassified. Topic counts describe only the public sources and period shown; they are not a measure of an MP’s priorities, effort, expertise, support or opposition. Signed-in users can report a questionable label from an issue journey for review.

Browse the taxonomy and issue journeys.

Plain-language explanations

“What happened” text states the recorded event, date and source type. “Why this may matter” explains the general parliamentary or procedural role of that kind of record. The two are visually separated. Context does not allege motive, intent, influence or causation, and every supported record keeps its official evidence link.

Following, briefings and alerts

Signed-in visitors can follow MPs and topics. Follows and notification preferences are stored against the authenticated ChatGPT user ID. The Following page matches recent indexed activity against those choices. The weekly briefing shows up to ten newest matches from the last seven days and explicitly reports when none were found. Activity alerts use stable event keys and unique per-user records to prevent duplicates; declaration amendments continue to use adjacent-register change keys. Pausing alerts stops new activity-alert creation but retains existing history.

Sharing and contact drafts

Share links open a stable Westminster Radar evidence page containing the event date, official source, retrieval details and a neutrality notice. The contact-your-MP tool creates an editable draft from one selected evidence record. It does not send messages, and it asks the user to check facts, personalise the wording and confirm the intended MP before using their own email application.

Fact, context and inference labels

Sourced fact reproduces or directly describes a cited public record. Explanatory context describes the relevant parliamentary process. Analytical inference labels a calculation, topic match, comparison or candidate relationship produced by Westminster Radar. Inference is never presented as official source wording, and uncertain links remain visibly unreviewed.

Search and comparisons

Advanced search combines indexed activity with filters for source type, date, MP, party, constituency, topic and organisation. Topic filters apply the published keyword taxonomy. MP comparisons show raw measures for a selected period and do not produce a composite score. Constituency peers are explicitly user-selected because the current sources do not contain a verified demographic similarity model. Interest context uses active MPs in the latest complete register, excludes child records and displays the median and middle 50% range.

Known gaps

Public parliamentary records do not capture constituency casework, private correspondence, party work or every ministerial duty. Source publication delays and outages can also create temporary gaps. The absence of a record is not evidence that an MP did no work, and a published record does not prove wrongdoing, influence, motive or causation.

Corrections, privacy and measurement

Correction reports receive a stable reference and retain a public audit note when status changes. Signed-in preferences can be deleted from the Privacy page. Product analytics stores only a random first-party visitor identifier, event type, page path and time; it excludes postcodes, search phrases, message content and correction details and is intended for aggregate service improvement.

Change detection

Adjacent complete register snapshots are compared by Parliament interest ID. An ID appearing for the first time is labelled new; a shared ID with different canonical source fields is amended; and an ID absent from the next register is removed. For the latest pair, the site compares summary, category, dates, member name and every retained declared field to show readable before-and-after values. These labels describe changes to the published register, not the reason for them.

Organisation mentions

Organisation names are extracted only from relevant declared fields. Automatic grouping is limited to a supplied company identifier or an exact name after Unicode, case and spacing normalisation. Placeholder values such as “confidential” and “private individual” are excluded from the directory. Fuzzy aliases are not merged automatically, every dossier carries a matching-confidence label, and all groups remain marked unreviewed.

Companies House enrichment

A company record is linked only when a company number is explicitly present in the parliamentary declaration. Older numeric identifiers are normalised to the eight-digit format used by Companies House. The interest page and dossier can show the registered name, status, type, incorporation date and registered office read from the public company page, with direct links to the official overview and filing history. A shared company number may connect separately declared names, but it does not prove that every unnumbered mention refers to the same legal entity. Companies House states that it does not check the accuracy of information filed.

Political finance enrichment

The Finance page reproduces reported donations and loans, including the participant, regulated entity, amount, dates, status and direct official-record link. Organisation dossiers match finance participants only by an exact normalised name or a company number explicitly present in both sources. MP profiles match only when the regulated-entity name exactly matches the member’s current Commons-register name. These conservative links can miss valid matches; they are evidence pointers, not assertions of identity, influence or wrongdoing.

Member profiles

Current party, constituency, official portrait, posts, committee memberships and policy focus are requested from Parliament’s Members API. Biography and focus sections are omitted when Parliament does not supply them; the declarations remain available from the locally verified register index.

Interpretation

A declaration records information supplied under Parliament’s rules. Appearance in this service is not evidence of improper conduct. Monetary fields are not combined unless their meaning and units are directly comparable.

Remaining methodological work

Organisation review decisions need an auditable approval workflow before any broader alias matching is introduced. The Electoral Commission historical backfill is deliberately staged and its progress is shown on the Data page. Monetary values remain contextual fields rather than a single combined score.