# Methodology — "King of Hearts": the Westminster charm audit

## What was measured

Two metrics, for (a) every UK Prime Minister in the Hansard record and
(b) every member of Andy Burnham's cabinet, over each person's **whole
parliamentary career** (Commons and Lords):

1. **Laughs** — the number of "[Laughter]" stage directions Hansard
   transcribed inside the person's own spoken contributions (variants
   matched: "[Laughter.]", "(Laughter)", "loud/prolonged/renewed laughter"
   etc., by regex). Contributions merely containing the *word* laughter
   score zero. Hansard does not record whether the House laughed with or at
   the speaker; we note the most famous "at" cases in the piece.
2. **Romance (strict)** — occurrences of a 27-term affection vocabulary
   (full list below) where the usage is verified as genuinely
   affectionate/romantic. Everything else is retained, flagged, in the
   dataset: raw and cleaned counts are published.

Headline rates divide by years in Parliament (ParlParse service records,
dissolution gaps excluded).

## Sources and volume

- Hansard API (hansard-api.parliament.uk), spoken contributions only, swept
  22–23 July 2026: **48,329 unique (contribution, term) rows in 560
  requests, zero failures**, cached in SQLite (charm.db).
- Speaker resolution: Hansard MemberId → ParlParse for 2005+; parsed
  AttributedTo (name, initials, constituency, office wrappers) matched
  against ParlParse (memberships to 1761) for earlier records; Members API
  fallback for 82 modern IDs missing from ParlParse.
- **PM office attribution**: rows attributed to "The Prime Minister (…)"
  — the PMQs convention — are assigned to the office-holder on the sitting
  date, from the Wikidata list of PM tenures. During this step Wikidata was
  found to be listing *David, the fictional PM from Love Actually*
  (Q131426522), as an incumbent since 2003; he was removed (and footnoted,
  because some things are too good to delete).
- PM gender and TheyWorkForYou IDs from Wikidata (bulk SPARQL, ~4 queries).

## Romance term list and variants

romantic(ally); romance(s); honeymoon(s); "I love"/"we love"; fell/fall(en)
in love; in love with; my darling; sweetheart(s); love letter(s); love
story/stories; true love; wedding anniversary/ies; valentine('s);
kiss(es/ed/ing); cuddle/cuddly; wooing/wooed; courtship; love at first
sight; blind date(s); speed dating; Love Island; chat-up line(s);
flirt(ing/ed/ation); my wife; my husband.

## False-positive rules (all flagged, none deleted)

- **Political metaphors**: honeymoon period/is over; sweetheart deals;
  "kiss goodbye to", "kiss of death", "kissing hands" (taking office);
  wooing voters/investors; flirting with disaster/the idea/insolvency;
  romantic poets/movement; courtship displays (ornithology).
- **Names**: people called Valentine.
- **Quoted material**: occurrences inside quotation marks (curly or paired
  straight quotes) or immediately after reading-out markers ("wrote",
  "letter", "told me", "said:") — MPs constantly read constituents'
  letters mentioning "my husband/my wife"; those are the constituent's
  spouse, not the member's. (Discovered via Andy Burnham's rows, two of
  which quoted a contaminated-blood widow and a Leveson witness; also
  Edward Heath — famously unmarried — "mentioning his wife" while quoting
  an unemployed constituent.)
- **Bereavement and abuse contexts**: windows containing died/widow/
  funeral/grief etc., and grooming/exploitation contexts, are excluded
  from romance regardless of phrasing.
- Rule-flagged 'review' rows (5,308) are counted in the published
  "cleaned" column but NOT in the strict headline counts.

## Fairness caveats

1. **Per-year rates are suppressed** (shown †) where more than 20% of a
   person's rows fall outside their ParlParse-recorded service — chiefly
   PMs who spoke from the Lords before 1999 (e.g. Lord Salisbury: 48
   laughs, no fair denominator).
2. **Era bias in laughter transcription**: Victorian Hansard logged
   "[Laughter.]" enthusiastically and at multiple points per speech;
   the modern record is sparser. Cross-era comparisons flatter the 1890s;
   within-era comparisons (e.g. cabinet colleagues) are safe.
3. Pre-1909 Hansard is not verbatim; historic counts are floors.
4. Namesake risk in 19th-century name matching persists (three Sir Robert
   Peels); no top-20 result depends on an ambiguous era.
5. PMQs attributions ("The Prime Minister") begin appearing consistently
   in the modern record; historic PMs are mostly matched by name.
6. The gender split covers the 23 cabinet members only (12 women, 11 men
   — the first female-majority cabinet), normalised per 100 years of
   combined service.

## Reproducibility

Pipeline at /home/claude/kingofhearts: sweep_charm.py → classify_charm.py →
resolve_charm.py (+ mnis fallback) → analyze_charm.py → build_data.py.
All API responses cached; re-runnable as the cabinet changes. Outputs:
pm_leaderboard.csv, cabinet_charm.csv, gender_split.csv,
quote_candidates.md. Licences: Open Parliament Licence v3.0 (quotes),
CC BY 3.0 © UK Parliament (portraits), CC0 (Wikidata), ParlParse (public
data). Total external requests for the entire audit: ~700.