How the scorecard works

Sources

A full, per-section listing of every dataset used across this site is on the References page.

• Promises - each candidate’s manifesto on election2025.gg for the 18 June 2025 general election, plus the equivalent by-election candidate page for any seat contested since.
• Party pledges - Forward Guernsey’s three members (Gavin St Pier, Rhona Humphreys, Tom Rylatt) stood on the Future Guernsey policy movement’s manifesto, adopted “in its entirety”, so its pledges are scored alongside each member’s own promises and marked with a “Forward Guernsey pledge” badge. The original manifesto document is no longer available online, but a near-final draft was sourced directly from Forward Guernsey, so pledges are drawn from it across all six policy areas: tax, housing, the economy, climate, education and health.
• Votes - the States of Deliberation’s Simultaneous Electronic Voting record (statesvoting-records.gov.gg), which publishes how every deputy voted (Pour / Contre / Ne vote pas) on each recorded proposition. That record is not quite complete: when the electronic system is unavailable the States’ Greffier takes the division by manual name call, and those divisions are printed in the Official Report but never reach the electronic record. On 16 July 2026 the chamber wi-fi failed and five divisions were taken this way. We read those name tables out of Hansard and file them alongside the rest, and they are labelled as Hansard-sourced wherever they appear.
• Amendments laid - taken from the Official Report, which records who moved each amendment and what became of it. Shown on each deputy’s page and never scored: see Limitations.
• Delivery verdicts - where no recorded vote bears on a promise, an AI assistant researches the public record (gov.gg, parliament.gg, States Resolutions, the Guernsey Press, BBC, ITV) and cites its evidence. Every AI-generated verdict is checked by a human editor before publication.

How we do and don’t use AI

AI (Claude, from Anthropic) touches three stages of the pipeline. It never touches a fourth: the arithmetic that turns a verdict into a score is fixed code, not a model call.

• Extracting promises- each deputy’s manifesto page is fetched from election2025.gg and Claude reads it to pull out forward-looking commitments as discrete, quotable promises. It is not asked to judge, rank or paraphrase them into something stronger or weaker than what was written.
• Linking promises to votes- for each deputy, Claude is shown every one of their promises alongside every recorded vote of the term and asked which votes genuinely bear on which promise, and which way of voting would deliver it. This is the “editorial judgement” referred to under Limitations below - in practice, that first pass is AI-drafted, against a written standard (the vote must materially bear on the promise, judged from the actual motion text, not just a shared proposition title).
• Delivery verdicts - covered under Sources above: where no vote bears on a promise at all, Claude researches the public record and writes a sourced verdict (delivered / partial / in progress / outstanding / reversed / unverifiable), citing what it found.

What AI does not do:

• It does not decide how a deputy voted. That comes only from the States’ own published voting record (or Hansard, where the electronic system failed) - a mechanical read of a public document, not a judgement call.
• It does not compute any score, rank or badge. Every point value on this page (+1 for a kept promise, −8 for a broken GST promise, and so on) is fixed, published, and applied by ordinary code - the same arithmetic runs whether a verdict came from a vote or from an AI delivery verdict.
• It does not have the final word on any individual link or verdict. Once a promise-vote link or a delivery verdict has been corrected by a Bridgehead editor, that correction is permanent: it is flagged as editorially settled and is never regenerated or overwritten by a later automated run, even one that re-processes everything else.
• It is not used to characterise a deputy, write commentary, or decide what counts as good or bad conduct - the scoring rules below do that, and they are the same numbers for every deputy.

Nothing here is retrospectively hidden: a promise-vote link and a delivery verdict both carry their rationale and cited evidence on the deputy’s own page, so you can see what the model was shown and what it concluded, not just the resulting label. If you think an AI-drafted link or verdict is wrong, use the “Report a profile issue” button on the deputy’s page - that is what routes a correction to a human editor and locks it in going forward.

How a promise is scored

For each promise we identify the recorded votes that genuinely test it, and for each we determine which way of voting would deliver it. We then read off how the deputy actually voted:

A promise is kept when the deputy voted to deliver it. Whether the motion passed is not part of it: a deputy who votes for an amendment that is then defeated has still kept the promise, and most of the votes scored on this site are on motions that lost. Nor does a promise have to end up in law.

Procedural divisions are not scored. A Rule 24(6) motion asks whether an amendment goes beyond the propositions before the Assembly, and a Rule 26(1) motion asks whether debate should stop - neither asks whether the underlying policy is any good. Scoring them would record deputies as having taken positions they never took. They still count towards participation and are still shown on a deputy’s page.

• Kept - voted to deliver the promise on every bearing vote.
• Broken - voted against delivering it on every bearing vote.
• Mixed - voted to deliver on some, against on others.
• Abstained - there were bearing votes, and the deputy was present for every one of them but explicitly declined to take a side (recorded as “Ne vote pas”).
• No side taken - there were bearing votes, but the deputy never took a side on any of them for some other reason - absent, excused, or otherwise unable to vote. Scored the same as Abstained (see below); shown separately because being absent and deliberately declining to vote are different facts on the record.
• No recorded vote - nothing in the recorded voting record bears on the promise (often the case for executive or committee-level commitments). These fall back to the AI delivery verdict.

How deputies are ranked

The default (“Overall”) ranking has two parts, added together: a promise-quality score and an attendance penalty. Every point below is a fixed, published number - there is no hidden weighting.

The promise-quality score is the deputy’s points from the list below, averaged per tested promise - not summed. That matters: a deputy who kept every one of 3 promises and a deputy who kept every one of 30 both score exactly the same (the full +1 below), because keeping every promise you made is equally good regardless of how many you made. Before 2026-08-19 this was a raw sum, which let a deputy with a large volume of kept promises always outrank a smaller, equally spotless record - rewarding volume rather than reliability. There is no way to separately weight a promise by how significant it was (nothing in the underlying data tags that), so this is the fair measure available: rate, not raw count. Points:

• +1 for each promise kept.
• −2 for each promise broken on a topic other than GST - broken promises count doubly against a deputy’s score, on top of dragging down their rate, rather than being treated as merely the absence of a kept promise.
• −8 for each promise broken on GST - four times the standard broken-promise penalty, and above the abstention penalty below. GST has been the single most contentious manifesto topic this term, so a reversal there is treated as the single most costly thing a deputy can do to their record.
• −1 for each promise landing mixed (delivered on some bearing votes, broken on others) on a topic other than GST - half the ordinary broken-promise penalty, since a mixed record is a real but partial reversal rather than an outright one.
• −4 for each promise landing mixed on GST - four times the ordinary mixed-promise penalty (the same GST severity multiplier used for broken promises above), but still below the full GST-broken penalty: a partial GST reversal is judged more seriously than an ordinary mixed verdict, without being treated as equivalent to an outright GST break.
• −3 for each promise landing Abstained or No side taken - more than an ordinary broken promise, since a broken promise at least reflects a decision taken and defeated, whereas neither of these reflects a side ever being taken at all. Weighted identically: whether a deputy was absent or deliberately abstained, the promise itself was equally untested.

The attendance penalty is added on top of that average as an absolute cost, not divided by promise count - attendance reflects term-wide conduct, not a promise outcome, so it shouldn’t be diluted for a deputy who happened to make a lot of promises.

• −3 flat, once a deputy’s States-chamber attendance crosses the low-attendance threshold below (see Attendance) - a base cost for being a persistent non-participant, on top of the per-vote penalties below.
• On top of that flat penalty: −0.2 for each chamber-wide vote where the deputy was present but recorded “ne vote pas” (an abstention), and −0.5 - two and a half times as much - for each vote where they were absent with no documented excuse on file. Not showing up at all, without even the transparency of a recorded “no”, is judged the more serious failure.
• That States-chamber attendance penalty (the flat cost plus the two per-vote costs above - never the promise-quality score) is then reduced by up to 30% for a deputy with a strong committee attendance record, in proportion to it: a deputy who attended every committee meeting has that one penalty cut by up to three-tenths, a deputy with no committee data or a poor committee record gets no reduction at all. Committee attendance still can’t be used to shrug off chamber attendance to any great degree - showing up to committees is not a substitute for voting - and it never touches the promise-quality score. The two rooms stay separately visible everywhere else on the site (see Attendance below); this is the one place they interact.
• Separately: −6 flat once a deputy’s committee attendance crosses that same 85% threshold - double the equivalent States-side flat penalty, since missing committee meetings is judged more seriously than missing chamber votes - plus −0.4 for each percentage point missed below 100% once flagged. Unlike the offset above, this is a genuine, standalone cost for poor committee attendance in its own right - not just a softener on the States-side penalty.

The combined score, then kept-rate, then the raw number of promises kept, break any remaining ties. Only promises with a direct or clear (high- or medium-relevance) bearing vote count towards the score - votes that touch a promise only partially or indirectly are shown on the deputy’s page as “related votes” for context, but never move the verdict. To be ranked, a deputy needs at least three scored promises, unless every promise they’ve made has already come to a vote - a smaller record is still a complete one once nothing is left untested, so it ranks rather than waiting on a floor built for partial records. Deputies who clear neither bar are listed separately as an insufficient record rather than placed in the table.

The rankings page also offers a “build your own list” menu to re-rank by any combination of promises kept, promises broken (a plain count there - every broken promise weighs the same whatever its topic, unlike the Overall score above, which still weights a GST break more heavily), committee attendance (the real committee-meeting figure, not the chamber one), and States attendance. Selected on its own, “promises broken” is a worst-offenders-first list; selected alongside another value, it switches to a penalty, so a strong kept record and a bad broken record can’t both push a deputy to the top of a combined list.

Most other topics - including environment and energy - don’t have enough scored promises per deputy for a fair ranked table (the most any deputy has is two), so they get a plain editorial record instead of a rank: see Energy & environment on the rankings page for who promised what on offshore wind, tidal power and climate, and what happened when it came to a vote.

Disagree with our point values? A “Choose your own penalties” panel on the rankings page exposes every number above as a slider and re-ranks the whole table by your own weighting instead of ours. Your choice never changes our own published figures, and is shareable via the page’s URL so you can send someone your version of the table.

Attendance: two separate measures

Deputies work in two rooms, and we count them separately. One measure covers the States chamber, the other their own committees. They are never blended into a single attendance figure - a deputy’s card, and the committee-attendance ranking lens, always show the two rooms on their own terms, and a deputy can be diligent in one and absent from the other. The one exception is the Overall score above, where a strong committee record can soften (not blend away) the chamber-attendance penalty specifically. Both rooms use the same bar for the badges below: missing more than 15% of that room’s meetings. Where a deputy is flagged, their card says which room.

A strong record is credited on the same terms. Attend 95% or more of a room’s meetings, and the card carries a positive badge for it. Because the two rooms are scored independently, a deputy can be credited for one and flagged for the other on the same card - that is the split doing its job, not a contradiction. Between the two bars, a record is neither notably good nor notably poor and carries no attendance badge at all. We set the credit at 95% rather than a perfect 100% because committee attendance counts a deputy present whether they stayed for the whole meeting or only part of it, so a perfect score is a softer claim than it first appears.

1. The States chamber

Separately from kept-rate, we track how often a deputy actually takes a side - Pour or Contre - across every recorded vote since they took their seat, not just the ones that bear on a manifesto promise. A deputy who is frequently absent or votes ne vote pas is flagged on their card and their page once they have at least ten recorded votes to judge from. This measures showing up and taking a position, which is a different thing from whether the position taken matched a promise - a deputy can have a perfect kept-rate on the few votes they attend and still be flagged here for missing most others.

The States’ own voting record distinguishes two different ways of not participating in a vote, and we keep that distinction rather than merging it into one number:

• Abstained - recorded as “Ne vote pas”: the deputy was present for the vote but explicitly declined to take a side.
• Absent without reason - recorded as “Did not vote” or “Absent” on the source page, with no documented excuse on file: the deputy simply wasn’t there.

Where a deputy has a documented legitimate reason on file (for example, illness), that vote is dropped from the sample entirely - it counts neither for nor against them, rather than being lumped in as an unexplained absence.

2. Committee meetings

The second room. Deputy pages show committee meeting attendance where it has been reported. Nothing here feeds the chamber figure above, or the other way round. This is not an official States dataset - the requirement for committees to compile and publish members’ attendance was rescinded by States’ resolution on 31 March 2022 (P.2022/15), on the reasoning that simultaneous electronic voting already showed who was present for each vote. Committees still keep internal minuted attendance, and the Guernsey Press has twice obtained it on request this term and published full per-deputy, per-committee tables. The figures shown here are transcribed directly from those two articles, each dated and linked, and are never averaged across the two reported periods.

A deputy who misses more than 15% of their committee meetings is flagged on their card - the same bar as the chamber, applied independently to this room. That headline figure is weighted by how many meetings each committee held, and drawn from the most recent reported period only, so sitting on a busy committee counts for more than sitting on a quiet one. Two caveats matter when reading it. A deputy counts as attending whether they were there for the whole meeting or only part of it, so these figures are deliberately generous - we would rather understate absence than overstate it. And because the figure combines a deputy’s committees, one can clear the bar overall while still having a weak record on a single committee - the per-committee breakdown on each deputy’s page is the fuller picture, and any individual committee missed more than 15% of the time is shown in red there.

Limitations

• Not every States decision is a recorded vote - many pass on a show of hands. Only recorded (electronic) votes can be attributed to an individual deputy.
• Matching a promise to a vote, and judging which way delivers it, involves editorial judgement. Each link carries a short rationale; we welcome corrections.
• The scorecard measures how a deputy voted, not what they proposed or attempted. A deputy who drafts a policy, lays it as an amendment and is then prevented from having it debated casts no ballot on its substance, so nothing about that effort can reach their promise scores - the promise simply reads “no recorded vote”. Deputy David Goy put this to us directly in August 2026, after four of his Personal Income Tax amendments were discarded under Rule 24(6) without debate. He was right that the record was silent on work that had plainly been done, and each deputy’s page now carries an Amendments laid section recording what they put before the States and what became of it. It is shown beside the voting record, not folded into it.
• A lopsided division says more about the promise than about the vote. Where a motion carries without opposition, a deputy who voted for it has still delivered what they promised, and a deputy who never made that commitment gains nothing from the same division, having no promise for it to test. What the tally shows is how demanding the commitment was. A promise that nothing in the chamber opposed was an easier one to make than one carried against real opposition. Every scored vote carries its own division tally for that reason, so a promise kept 39 to 0 can be read differently from one kept 21 to 16.
• The current term began in July 2025, so the record is still short - many promises have no bearing vote yet, and AI verdicts often read “in progress”.
• This covers all 38 People’s Deputies elected on 18 June 2025. Where a seat has changed hands since (for example, following a by-election), the outgoing deputy’s record stays on the site for transparency but is excluded from the ranking.

Updates

New recorded votes are ingested daily; newly ingested votes are matched to promises weekly, and scores update automatically as ballots land on already-matched votes. Manifesto promises are fixed for the term. Delivery verdicts are refreshed on a rolling 30-day cycle.

Questions

Who funds this, whether any of it is new information, how AI is and is not used, why attendance is published at all, and what happens when a deputy says something is wrong, are all answered on the FAQs page.