How the scorecard works

The Guernsey Deputy Scorecard came first. We built it for the States of Deliberation and launched it in August 2026, and the Jersey States Member Scorecard applies the same method to a second Assembly. This site applies it to the House of Keys. Guernsey’s and Jersey’s methodology pages describe the original.

This is an adaptation, not a copy, because the Keys publishes less than the Channel Island Assemblies do. It prints the names of the members who voted For and the members who voted Against, and nothing else about a division. So there is no attendance figure here, because absences are not published. Most members are independents, so there is no party whip to measure. And there is no speaking or debate layer.

The site holds two terms: a closed 2021–2026 record and the 2026–2031 House, elected on 24 September 2026. Its first sitting, on 29 September 2026, elected the Speaker. The House has 24 seats in 12 two-member constituencies, and the 2026 election returned 22 independents and 2 Manx Labour members.

• Votes and Proceedings and Hansard, the official record of each sitting, published as PDFs on tynwald.org.im. Divisions are read from these.
• Members, constituencies and offices from the Tynwald member pages, cross-checked against the 2026 results on iomelections.com and Wikipedia’s page on the 2026 Manx general election.
• Manifestos from iomelections.com and from party and candidate websites. Each one was copied to the Internet Archive’s Wayback Machine so that a promise can still be checked if a page is later changed or taken down.
• Manx Radio’s candidate statements, used to find manifesto links. A short statement that is not a manifesto is not read for promises.

The references page lists each source with its address.

For each division the Keys prints the names of the members who voted For and the names of those who voted Against, in Hansard and in Votes and Proceedings. There is no official data feed. In June 2026 the Keys resolved to record every division electronically and publish the result by member, with no start date yet. Until that exists, this scorecard reads the PDFs.

Votes and Proceedings is the primary source. Hansard is a transcript, and it has been seen to differ. On 30 June 2026, on the Nicotine Products Bill, Hansard printed 23 For. The Speaker declared 22, and Votes and Proceedings printed 22. Where the two disagree, Votes and Proceedings and the Speaker’s declared tally decide.

A division counts only when two things are true. The names read from the document must equal the tally the Speaker declared, and every name must match exactly one member. If either fails, the division is held back. It is not shown as evidence of how anyone voted, and nobody’s vote is guessed to make it fit.

Names are printed as a title and surname, for example “Mr Ashford”. Each is matched to the members of that term. A name that matches nobody, or more than one member, holds the whole division back. Where a division was repeated, only the final one counts.

“The Speaker” and “The Deputy Speaker” are printed by role, not by name, so a vote under either label belongs to whoever held the role on that day. Juan Watterson was Speaker until 21 July 2026. The House then had no Speaker until David Ashford was elected on 29 September 2026. The sources disagree on when the Deputy Speaker changed in early 2024, so divisions between February and 6 March 2024 under that label are left out rather than attributed to the wrong member.

The record covers 111 sittings read from October 2021 to July 2026. In them we parsed 243 divisions, counted 241 and held back 2.

Divisions involve no model at all. They are read from the Tynwald documents and counted in code.

A language model, Anthropic’s Claude, is used for one thing: reading forward-looking promises out of a manifesto. The code decides what is published. A promise is kept only if its quoted wording appears verbatim in the manifesto, after whitespace and punctuation are normalised. A paraphrase presented as a quotation is dropped. A manifesto published only as scanned images has no readable text and is skipped, not transcribed by hand.

No kept or broken verdict is published yet, because the 2026–2031 House has not recorded a division.

When divisions arrive, the rule will mirror Jersey’s. This is the planned rule, not a live one. A vote will count in favour of a promise where the member voted to deliver it, and a promise with no vote bearing on it will stay outstanding and unscored.

The planned evidence bar is asymmetric. A vote will count against a promise only where it plainly contradicts it, where almost anyone reading the promise and the motion together would agree the vote runs against the promise. Where calling it a breach would need an inference or an assumption about the member’s motives, the vote will be shown and set aside, and doubt will go to the member. Votes on procedure will be kept out of the test altogether.

The reason is that the two kinds of error are different. Leaving a promise outstanding understates a member’s record, which is a shortcoming of this site. Calling a promise broken on a contestable inference is a false statement about a named person.

• A member who appears in neither list of a division may have been absent, may have abstained, may have been in the chair, or may have been out of the room. The printed record does not say which, so this site shows no attendance figure and never treats a missing name as an absence.
• Many divisions are on procedure, not policy. A vote on how the House conducts its business says little about whether a member backs a policy in a manifesto.
• Most members are independents. With no party whip behind a vote, a division reflects the member’s own choice, and a manifesto is a personal statement that varies widely in how specific it is.
• The 2026–2031 House has not yet recorded a division, so promises on it are tracked and untested.
• Some elected members published no manifesto. They have no promises on the scorecard, and that is a gap in what they published, not a finding about them.
• Only the public record is measured. Committee work, private negotiation and anything that never produces a public document are invisible here. That is a real limitation, not a claim that such work does not matter.
• Manifesto wording is interpretive. A commitment written broadly can bear on several votes, or none. Where a promise is open to more than one reading we would rather say so than claim a precision we do not have.

New sittings are added after Tynwald publishes its documents.

Every change we make is listed on the corrections page. Any member can ask for their record to be checked by writing to Bridgehead.

The questions members and journalists ask, such as why there is no attendance figure, why a division is held back and how to dispute a record, are answered on the FAQs page.