Sixteen Years of Manx Politics, on One Compass — the Historical Quadrant - Smart Island

The short version

Smart Island has a new page: Historical Compass. Every Isle of Man House of Keys candidate from 2011, 2016, 2021 and 2026 is plotted on a shared political compass — economic axis (state-intervention vs free-market) on the horizontal, social axis (authoritarian vs libertarian) on the vertical.

Each candidate is a dot, coloured by election year. Candidates who stood in more than one election get an arrow connecting their dots chronologically, so you can see how their stated positions shifted. The page works as a single full-screen view, with filters for year, elected status, and a side panel of every multi-election candidate.

Below the chart, two tables:

Biggest movers — the top-eight candidates whose compass position shifted the most, with a short AI-written hypothesis about why.
Full trajectories — every candidate who stood in more than one election, sortable, with columns for each year's position, the economic and social deltas, and a hypothesis for each shift.

The scoring is done by Azure OpenAI reading the actual manifesto PDFs — not from news clippings or our guesses. Everything is precomputed and cached as JSON, so the page has no per-request LLM cost at runtime.

Why build this

Because political conversation on the Island mostly happens by inference. "So-and-so was always a fiscal conservative", "X has moved left since they got into Cabinet", "The 2021 intake was more populist than the 2016 lot" — these are claims people make around the dinner table with very little evidence underneath them. The source material (candidate manifestos) exists, is public, and has been sitting in PDF archives unread. Putting it onto a shared compass makes the claims falsifiable.

The secondary reason is that candidates change their stated positions between elections more often than they tell you. A 2016 manifesto that was warmly pro-common-purse becomes a 2021 manifesto that demands fiscal discipline; a 2011 centrist drifts toward market-liberal after five years in Treasury. These shifts are legitimate, normal, and rarely discussed. Drawing arrows between a candidate's dots makes the drift visible without being accusatory about it.

What the data says so far

Across the three historical elections, we have scored manifestos for:

45 candidates from 2011 — out of 64 who stood. Pre-digital-manifesto era; many 2011 candidates never uploaded a document.
59 candidates from 2016 — out of 63 who stood. Includes 23 of the 24 elected MHKs.
58 candidates from 2021 — out of 65 who stood. Includes all 24 elected MHKs.
The live 2026 register — scores carried over from the existing 2026 compass page.

Around 20–30 candidates appear in more than one of those elections, so that's roughly the number of trajectories you'll see arrowed on the chart.

Some patterns are already visible without squinting:

Incumbents tend to drift toward the centre, not because they became more reasonable but because incumbency forces you to own compromise decisions. A manifesto that says "I will oppose X" in opposition becomes "we delivered Y" after five years in government, and Y is almost always a compromise with the state machinery. Candidates who hold seats and rerun almost invariably end up closer to (0, 0) on their second manifesto than their first.
The 2021 intake was not, actually, more populist than the 2016 intake. A few vocal candidates were, but the cohort average sits very close to where 2016 was. The difference is the range widened — the 2021 election produced both more-libertarian and more-authoritarian candidates than 2016, so the compass looks more spread out. This is the kind of thing the chart lets you see immediately and arguments otherwise can't survive.
Candidates who lost in one cycle and won the next usually moved toward the centre in between. A losing 2011 candidate who came back in 2016 with a softer manifesto, and a losing 2016 candidate who came back in 2021 with a less ideological pitch, are a recurring pattern. The exceptions — candidates who held their positions firmly between elections and still won the second time — are genuinely interesting and mostly fall in one-party constituencies (Rushen, Garff).
Arbory, Castletown & Malew — and the merged-constituency effect. Three pre-2016 seats (Castletown, Malew & Santon, and parts of Arbory) merged into one in 2015. Candidates who represented the pre-reform seats and re-stood post-reform mostly shifted slightly left-economically — their new larger constituency pulls them toward state-intervention on housing and healthcare questions. This is a constituency effect rather than a personal one, and it's visible on the chart.

How to read the biggest-movers table

The table sorts by magnitude — the straight-line distance between the candidate's first dot and their last dot on the compass. A magnitude of 5 or more is a meaningful shift; under 2 is "held steady". We then ask Azure OpenAI to read the trajectory alongside manifesto top-themes and the broader Isle of Man political context (post-financial-crisis 2011, Brexit 2016, mid-COVID 2021, AI-adoption 2026) and write a 2–3 sentence hypothesis. It is a hypothesis, explicitly not a claim: the model can see the numbers but not the interior life of the candidate.

Some hypotheses are more convincing than others. Where they are convincing, they tend to point at one of three things:

Job change. A candidate who moved from back-bench to Cabinet, or from Cabinet to back-bench, usually shows up as a visible compass shift. Responsibility for a portfolio tends to pull manifestos toward market-pragmatism; losing a portfolio frees a candidate to lean back into their prior ideological position.
Constituency pressure. Moving from a rural seat to an urban one, or representing a constituency with changing demographic composition, shows up in the scores.
Era effect. Candidates who continued across the 2016 → 2021 boundary mostly shifted slightly authoritarian, reflecting the post-COVID consensus around regulation and cost-of-living protections. This is a cohort effect rather than a personal one. 2026 candidates are showing an opposing shift — toward market-libertarian — in response to AI productivity narratives and the Pillar-Two corporate-tax environment.

How to read the full table

Columns:

Candidate — with a label showing whether they were elected in their most recent appearance.
Elections — year badges for every cycle they appear in.
First → Last — economic / social scores for their earliest and most recent manifesto.
Δ Econ / Δ Social — the net shift on each axis.
Magnitude — the Euclidean distance between first and last.
Direction — a short human-readable summary ("moved right on economics", "held steady", "shifted libertarian").
AI hypothesis — the 2–3 sentence Azure OpenAI explanation. Where this column says pending, the hypothesis pipeline hasn't been re-run since the underlying data last changed.

Every column is sortable. Click the column header to toggle direction.

Caveats you should know about before quoting this at someone

We only have what candidates published. If a candidate wrote a tight, disciplined manifesto on Tuesday and gave a fiery interview on Wednesday, only the Tuesday document feeds the scoring. This is deliberate — we don't want the scoring to drift with news-cycle noise — but it means the chart under-captures campaign-era radicalism for candidates whose manifestos were vanilla.
Manifestos differ in length and rigour. A six-page bullet list gets scored on what's there. A forty-page policy document gets scored on the same axes, which means some candidates get positioned more confidently than others. We disclose manifesto character counts in the source data for anyone who wants to weight their own read by this.
Pre-2016 constituencies don't map 1:1 to post-reform ones. We normalise historical seats to their nearest 2016-onwards equivalent so cross-year matching works, but some matches are approximate. For example Douglas West (2011) maps to Douglas Central (2016+) as the nearest equivalent, but the constituency was actually split between Central and South under the reform. Candidates whose 2011 seat was Douglas West may appear to have "moved" constituency when really the boundaries did.
Name spelling changes defeat the cross-year match. If a candidate went by "Alex Watterson" in one election and "Alexander Watterson" in another, they appear as two separate dots with no arrow. We normalise for trivial differences (punctuation, case) but not for nickname vs full-name.
AI hypotheses are hypotheses. They are generated from the compass data plus general Isle of Man political context. They cannot know what the candidate was actually thinking. Read them as starting points for a conversation, not conclusions.
Not every historical candidate has a scored manifesto. For 2011 particularly, uploaded manifestos are sparse — so a candidate who stood in 2011 but never uploaded a document won't have a 2011 dot, and their trajectory will only include later elections. This is inherent to the source material, not a choice we made.

Technical appendix

For people who care how the sausage is made:

Source PDFs archived locally. For each manifesto, we extract text with pdf-parse and trim it to under 12,000 characters (most Manx manifestos are well under this — 3,000–8,000 words).
Scoring: a single Azure OpenAI call per manifesto (gpt-4.1, temperature 0.3, JSON-mode response) returning { economic, social, summary, topThemes }. Same prompt shape as the 2026 compass enrichment. Cached per manifesto-year.
Stance-shift hypotheses: a second Azure OpenAI call per multi-election candidate, given the compass positions and constituencies, returning { direction, headline, hypothesis }. Cached unless the underlying trajectory changes.
2026 candidates are pulled straight from the existing live 2026 compass data — no re-scoring.
Everything runs as scraper pipelines (scrape-elections, extract-manifesto-text, score-historical-manifestos, hypothesise-stance-shifts) and produces JSON that the page reads. Zero per-request LLM cost.

Where to go next

Historical Compass — the page itself.
2026 Political Compass — the single-election version, with more per-candidate detail (policy positions, evidence feed, manifesto summaries).
Election Hub — the index page.

Feedback and corrections are welcome at hello@smartisland.im. The scoring is based on what candidates published; if you think a specific candidate is being mis-represented by their own manifesto, we would rather know.