How the scorecard works

The Guernsey Deputy Scorecard came first. We built it for the States of Deliberation and launched it in August 2026; this site is the same method applied to a second Assembly, because the case for a public record of promises against votes does not stop at one island. Guernsey’s methodology page remains the older and, in places, the fuller of the two.

The scoring core is shared, but this is an adaptation rather than a copy, and the differences are not cosmetic. Jersey grades two Assemblies at once, a closed 2022–2026 record and the live 2026–2030 term. Its members are Senators, Deputies and Connétables, and a member’s seat can change between Assemblies, so the record hangs off a member-and-term pair rather than a person. Attendance is computed from the Assembly’s own recorded reasons instead of the workaround Guernsey needs. Several Guernsey-specific weightings were deliberately not carried over, because they have no Jersey basis: there is no special treatment of any single tax or policy here. And this site has no speaking or debate layer, because Jersey’s chamber transcripts are not ingested.

• Votes and ballots— the States Assembly’s own recorded-vote and ballot records, one row per member per division.
• Manifestos— each candidate’s own page on vote.je for the election that returned them. Every promise links to the page it was read out of.
• Members, seats and offices— the Assembly’s published register at statesassembly.je, including each member’s dated Record of Service, plus the States Assembly post-holder dataset on opendata.gov.je.
• Island data— Statistics Jersey and opendata.gov.je, cited by dataset on each topic page.
• Media coverage— a fixed allow-list of Jersey outlets, linked to the original article in every case.

Vote outcomes, ballots and attendance involve no model at all. They are read from the official record and counted in code.

A model is used for four things: reading forward-looking promises out of a manifesto, matching a promise to the votes that bear on it, judging whether a vote that went against a promise is clearly a breach of it, and drafting the biography on each profile. In every case the code decides what is published, not the model. Biographies are grounded only in the Assembly’s register and the member’s own manifesto, and are checked programmatically before publication — a draft claiming an office the member’s own Record of Service does not contain is rejected rather than published.

Each promise is set against the recorded votes that bear on it. For each such vote we record which way of voting would have delivered the promise, and how the member actually voted. A promise is then kept where the member voted to deliver it, broken where they voted against delivering it, and mixed where they did both across different votes.

Where no recorded vote bears on a promise, it is recorded as outstanding and is not scored either way. Weakly connected votes appear on the page as context and are deliberately left unscored.

The evidence bar is asymmetric, and this is the most important rule on the site. A vote counts in favour of a promise on the terms above. A vote counts againsta 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. Every adverse vote is adjudicated separately on that question. Where calling it a breach would need an inference, an assumption about the member’s motives, or a judgement about whether one particular measure was the right way to honour a general commitment, the vote is shown on the page and set aside, with the reason given, and it does not grade the promise. Doubt is resolved in the member’s favour.

In practice that means most budget and amendment votes do not count against a promise. Voting against one funding amendment is not a repudiation of a general commitment the member has otherwise supported: they may have preferred a different route, judged the measure unaffordable or premature, or pursued the aim by other means. The same applies to procedural votes, and to votes on timing, funding levels, reporting, board membership, or a different mechanism within the same policy area.

The reason for the asymmetry is that the two kinds of error are not equivalent. 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, which is a different order of harm. A smaller number of well-founded adverse findings is worth more than a large number of arguable ones.

The rankings order members by an accountability score built from their graded promises and their attendance. Kept promises count for the member; broken and mixed promises count against, with a broken promise weighted more heavily than a mixed one; and a promise the member neither supported nor opposed when it came to a vote carries the heaviest weight of all, because declining to take a side on your own manifesto commitment is the least defensible outcome on the record.

A member appears in the ranking once at least 3of their promises have been graded, or once every promise they made has been tested. Below that the sample is too small to rank fairly, and they are listed without a position. Each Assembly is ranked separately — the term switcher at the top of the rankings page moves between them — because the two terms have entirely different amounts of voting record behind them.

Attendance is the share of recorded votes a member either voted on or abstained on, out of those they were expected to cast. It comes straight from the Assembly’s own ballot record.

The States Assembly records a reason against every non-vote, which lets this site do something Guernsey’s cannot: tell being ill, off-island, on parental leave, presiding over the sitting, or barred by a declared interest apart from simply not turning up. Absences with an accepted reason are removed from both sidesof the rate rather than counted against the member — declaring a conflict of interest must never read as an absence.

A member is flagged for low participation below 85%, and no rate is published at all until they have been expected to cast at least 10 votes, so a handful of divisions cannot produce a headline percentage.

• The live term has barely voted.The 2026–2030 Assembly first sat in June 2026 and has held only a handful of recorded votes, so almost every promise on it is tracked and untested, and no delivery verdicts have been generated against it. The graded record on a returning member’s page is their 2022–2026 term.
• Only the public record is measured. Private negotiation, Scrutiny-level influence and work that never produces a public document are invisible here. That is a real limitation, not a claim that such work does not matter.
• No speech layer.Unlike Guernsey’s scorecard, this site does not cover what members said in the chamber, because those transcripts are not ingested.
• Manifesto wording is interpretive. A commitment written broadly can plausibly 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.
• Members who left in 2026no longer have a page on the Assembly’s register, so their profiles carry fewer sources than a sitting member’s.

Votes and coverage are ingested on a schedule; promise matching re-runs as new votes land. Editorial corrections, suppressions and breach adjudications are pinned in the data and carried across each re-run, so a correction cannot be quietly undone by the next job.

Every change we make is listed on the corrections page. Any individual ruling can be challenged from the promise itself, and any member can report a problem from their own profile.

The questions States Members actually ask — who is paying for this, why Jersey, what the record cannot see, and what happens when we get something wrong — are answered on the FAQs page.