# swe elections pack handoff

Written 2026-09-03 for Cartevoir.

Sweden has 2 elections: swe-riksdag-2026 (2026-09-13, confirmed_official, imminent) and swe-riksdag-2022 (2022-09-11, confirmed_official, sitting). people.person_id is the official Riksdag intressent_id prefixed swe-. The roster contains 349 current Riksdag members; bios are blank rather than authored.

money-edges.csv is header-only: Kammarkollegiet disclosures were not obtained in machine-readable form; no amounts were invented. polls.csv contains the latest Novus published topline (8 option rows; field dates 2026-08-03 through 2026-08-16; sample 5617; MoE not printed in cited table). results.csv is CLEA Release 18, 14,978 rows, constituency/candidate grain.

Join results on election_id and constituency_id; join people only on official person_id. CLEA does not provide Riksdag intressent IDs, so result person_id is blank. Source licenses/terms require downstream review before republication; CLEA citation is Kollman et al., Release 18, ICPSR 39687.

## 2026-09-03 repair: polls.csv enrichment

Appended 72 option-rows to `polls.csv` (preserving the original 8 Novus rows), bringing it to 80 rows across 10 distinct survey waves from 7 institutes: Novus, Demoskop, Verian, Ipsos, Indikator Opinion, SCB (Statistics Sweden, Partisympatiundersökningen), and Sentio. Field dates run from early August through the final pre-election weeks; most recent field_end is 2026-09-01, twelve days before the 2026-09-13 vote. All rows use `swe-riksdag-2026`, which exists verbatim in `elections.csv`. Sifo/Kantar Public was searched but no cleanly citable 2026 topline table was found in the time available, so it was not included — no data was invented to fill the gap. `moe` and `sample_n` are blank wherever the publishing outlet did not print them (e.g. some Demoskop/DN tables report only party shares without a stated margin of error).

`results.csv` was re-evaluated for a val.se upgrade. Valmyndigheten's "rådata" 2022 packages are aggregate/party or personal-vote-count (`personröster`) files, not a party×constituency×candidate file carrying total votes, vote share and seats comparable to CLEA's grain, and none of them carry a Riksdag `intressent_id` to join against `people.csv` — building that join would require fuzzy name-matching ~7,000 candidate rows against 349 members, which is not a "cheap" upgrade. Per the repair brief, CLEA Release 18 (14,978 rows, unchanged) stays in place with its existing `CLEA (Kollman et al., Release 18, ICPSR 39687)` citation.

`STATS.json` regenerated from the files as they stand.

Data-quality note (found, not altered): 10 of the 14,978 CLEA rows in `results.csv` (e.g. `swe-109917`, `swe-111884`, `swe-111889`, `swe-116331`, `swe-116338`) are short one column — independent/minor-party candidates (e.g. "Panos Alepliotis,SKP,", "Per-Olof Jönsson,SKP,") whose `party` field appears to have been dropped rather than left empty during the original CLEA-to-CSV conversion, shifting `votes`/`vote_share`/`seats`/`source`/`source_url` left by one and leaving `license` blank for just those rows. This predates this repair session (results.csv was not rewritten) and is left as-is rather than guessed at; flagging here since it's a first-class layer per SCHEMA.md.