# CAN elections pack handoff

Written 2026-09-03 to `/workspace/elections-ingest/can/`.

## Files / counts

| File | Rows (excl. header) | Path |
|---|---|---|
| elections.csv | 2 | /workspace/elections-ingest/can/elections.csv |
| people.csv | 338 | /workspace/elections-ingest/can/people.csv |
| money-edges.csv | 218699 | /workspace/elections-ingest/can/money-edges.csv |
| polls.csv | 62 | /workspace/elections-ingest/can/polls.csv |
| results.csv | 1959 | /workspace/elections-ingest/can/results.csv |
| SOURCES.md | — | /workspace/elections-ingest/can/SOURCES.md |
| STATS.json | — | /workspace/elections-ingest/can/STATS.json |
| HANDOFF.md | — | /workspace/elections-ingest/can/HANDOFF.md |

## Cycle status (verified against Elections Canada / the statute)

- `can-hoc-2025` **sitting** — 2025-04-28, `confirmed_official`. 45th general election; Elections Canada
  published the official voting results (raw data page last modified 2025-11-12). 45th Parliament sitting.
- `can-hoc-next` **sitting** — 2029-10-15, `date_confidence=constitutional_due`. Not called. The date is
  **computed from the statutory formula**, not announced: Canada Elections Act s.56.1(2) — "the third
  Monday of October in the fourth calendar year following polling day for the last general election" —
  which after 2025-04-28 is Monday 2029-10-15. s.56.1(1) expressly preserves the Governor General's power
  to dissolve Parliament earlier, s.56.2 allows an alternate day, and the 45th Parliament is a **minority**,
  so treat this as a ceiling, not a fixture.
- No officially announced general-election date contradicts the brief. One live-content note: the Elections
  Canada "Elections" landing page still contains the sentence "a federal general election has been called"
  **inside an HTML comment** — a 2025 leftover, not current. The visible content on 2026-09-03 announces
  only the 2026-08-31 by-elections (Chicoutimi–Le Fjord, Beaches–East York, North Vancouver–Capilano).

## Join keys

- `people.csv.person_id` = **House of Commons "Person ID"** — the numeric id in
  `ourcommons.ca/members/en/{id}` (e.g. `89156` = Ziad Aboultaif). Never a name slug.
- `people.csv.election_id` → `can-hoc-2025` for all 338 rows.
- `money-edges.csv.election_id` → `can-hoc-2025` on every row (join anchor only — see money caveats).
- `results.csv.constituency_id` = **Elections Canada electoral-district number** (e.g. `10001` Avalon);
  343 districts. `result_id` = `can-ec-2025-{district}-{CandidateName}`.
- `polls.csv.election_id` → `can-hoc-next`.
- **There is no id-level bridge between people, money and results.** See below; do not build one by name.

## Licences

- Elections Canada material (elections.csv sources, money-edges.csv, results.csv): the `license` column
  says only **"Elections Canada open data (elections.ca)"**. The Elections Canada Important Notices and
  Open Data terms pages both returned **HTTP 500** from this box on 2026-09-03, so no licence text could be
  machine-verified; confirm the exact terms before commercial redistribution.
- people.csv: **"House of Commons of Canada (ourcommons.ca) - see site Important Notices"** — same
  caveat, the House publishes its own terms of use.
- polls.csv: each row carries the pollster's own copyright string; all figures are the pollsters' published
  toplines.

## Bars

- Sitting MPs: **338** of 343 seats (5 vacancies — see caveats). `person_id` = ourcommons Person ID, 338/338 present.
- Polls: **62 option-rows** from **11 surveys** across 5 organisations (Abacus Data, Angus Reid Institute,
  Ipsos, Léger, Nanos Research) — bar was ≥40.
- Money: **218,699** Elections Canada contribution edges, **CAD as filed**, non-empty.
- Results: official Elections Canada poll-by-poll returns, reconciled to EC Table 12 (no CLEA fallback needed).

## People

Incumbents only (`role=incumbent`, `incumbent=true`); there is no next-election candidate list because the
election is not called.

- `bio_text` is **empty for all 338 rows** — ourcommons.ca member pages publish roles, committee
  memberships and contact details, not a biography paragraph. Nothing was authored.
- `photo_url` is the official House of Commons 45th-Parliament MP photo, read off each member's profile
  page (338/338 found). These filenames could **not** be derived from name + party: the filename encodes
  the party *at photo time* (e.g. `DEntremontChris_CPC.jpg` for a member now sitting as a Liberal), so
  guessing them would have produced wrong or dead links.
- `party_code` is **blank on every row** — the House member feed publishes the full caucus name only
  ("Conservative", "Bloc Québécois"), with no official abbreviation column. Not invented.
- Party mix (feed strings): Liberal 173, Conservative 138, Bloc Québécois 21, NDP 5, Green Party 1.
- Provinces: ON 121, QC 75, BC 43, AB 37, MB 14, SK 13, NS 11, NB 10, NL 7, PE 4, NT/YT/NU 1 each.

## Money

Elections Canada **Political Financing → Open Data**, "Contributions — as submitted" bulk download
(206 MB zip → 4.4 GB CSV, **12,655,377** contribution rows across all events and fiscal years). Streamed
and filtered on this box; **CAD exactly as filed** (no conversion, no rounding). Three streams, two `cycle` labels:

- `cycle=2025-ge45` — **6,656** edges, all `Candidates` rows for the *45th general election*
  (`Candidate's Electoral Campaign Return (2019)`). Sum: **CAD 4,853,010.87** across **846** distinct candidate
  recipients.
- `cycle=2025-annual-writ-period` — **212,043** edges: registered **party** annual (fiscal 2025)
  returns, **176,050** edges / CAD 37,847,539.63 across 10 parties, plus registered **electoral district
  association** annual (fiscal 2025) returns, **35,993** edges / CAD 16,957,544.72 across 779 EDAs — in both
  cases restricted to contributions *received* in the writ period **2025-03-23 … 2025-04-28**.
- Whole-file sum: **CAD 59,658,095.22**. 89 edges have an as-filed amount of **0.00** (kept as filed).
- `edge_type` is what the regulator published: **`donation` 218,230** (Monetary amount) and
  **`donation_non_monetary` 469** (Non-Monetary amount). Nothing is labelled PAC or IE — Canada has no
  US-style PACs, and this file contains contributions only, not third-party advertising.
- One filed contribution row that carries **both** a monetary and a non-monetary amount becomes **two**
  edges (one per amount type). That is why 218,682 filed rows produce 218,699 edges. `as_filed_note`
  preserves the source row's `form=` (EC form id), `ret=` (stream), `rid=` (EC recipient id),
  `rparty=`, `dist=`, `ctype=` (contributor type), `loc=` and `recd=` (received date).
- Recipient party mix of the edges: Liberal 102,747 · Conservative 101,429 · Green 4,962 · NDP 4,024 ·
  Bloc Québécois 2,950 · PPC 1,067 · CHP 499 · Independent 397 · others <400 each.
- `contributor type` is **Individuals for 100%** of rows — Canada bans corporate and union political
  contributions, so there is no corporate donor layer to display.

**Every money row has a blank `person_id`.** Elections Canada keys recipients by its own **Recipient ID**
(kept in `as_filed_note` as `rid=`), which is a different registry from the House of Commons Person ID in
`people.csv`. **Do not name-match money to the roster** — "Duchesne, Larry" in a candidate return is not
evidence about any sitting member, and the candidate stream includes the many candidates who lost.

Three more money warnings:

1. **This is a filing snapshot, "as submitted".** Only 846 of the 1,959 GE45 candidacies appear, because
   candidate returns are published as they are filed and audited. Absence of a candidate is **not** evidence
   of zero contributions.
2. The annual party/EDA money is **not** election-period money in any legal sense; it is annual-return
   money whose *received date* happens to fall inside the writ period. Use the `cycle` column, not
   `election_id`, when describing it. Party and EDA returns can also each report the same individual's
   gift at different levels of the organisation — do not sum across streams and call it "total 2025 donations".
3. Deliberately **excluded** (present in the source, not donor→recipient edges for this cycle):
   quarterly party returns (111,725 rows, which would double-count the annual return), leadership-contestant
   returns (78,390 rows), nomination-contestant returns (536), and by-election candidate returns
   (Aug 2025 / Apr 2026, 313 rows). Contributions are also the only financial flow here — expenses and
   transfers were not ingested.

## Polls

`polls.csv` is long format, one row per response option, attached to `can-hoc-next`.

- **Abacus Data** — 4 surveys (9–16 Jul, 23–29 Jul, 7–12 Aug, 21–26 Aug 2026), full toplines with
  printed sample sizes and Abacus's stated comparable-probability MoE.
- **Angus Reid Institute** — 2 surveys. 4–10 Jun 2026 (n=4,237) prints only Lib/CPC/NDP, so only those
  three rows exist; 25–27 Jul 2026 (n=1,770) topline taken from ARI's own published data tables PDF
  (decided+leaning base 1,571).
- **Ipsos** for Global News — 2 surveys (25–26 Feb, 2–7 Apr 2026), n≈1,000, **credibility interval**
  ±3.8 recorded in `moe` with the term named in `notes` (Ipsos does not publish a classical MoE).
- **Léger** — 2 surveys (19–22 Jun, 1–3 Aug 2026), decided voters.
- **Nanos Research** — 1 survey, the weekly tracking wave ending 7 Aug 2026 (n=1,038, accurate ±3.1),
  read from the Nanos political package PDF. Nanos publishes a rolling 4-week average; the release index
  gives publication dates, which are **not** fieldwork dates, so only the wave with a printed field
  window was used.
- **Research Co. produced no rows** — its site index at build time carried issue polls (health care,
  policing, Alberta separation, justice) and no federal vote-intention table. Not invented.
- No Wikipedia anywhere. `moe` and `sample_n` are blank wherever the source did not print them.

## Last-cycle returns

`results.csv` — **1,959** rows, `election_id=can-hoc-2025`, `office=lower_chamber`, all from Elections
Canada's official 45th-general-election **poll-by-poll** files (343 district CSVs, all-Canada zip):

- `votes` = each candidate's votes summed over every polling division in the district.
- **Validated against Elections Canada Table 12** (official candidate votes and published percentages):
  district totals match for **all 343 districts** (0 mismatches) and the derived percentages match the
  published ones for **all 1,959** candidate rows.
- `vote_share` is **derived** (candidate votes ÷ district valid votes) and stored as a fraction, not a
  percentage; `seats` = 1 for the 343 elected candidates, else 0. Total valid votes: **19,641,663**.
- `person_id` is blank on every results row — Elections Canada publishes candidate names, not House
  Person IDs, and joining the two registries would be a guess.

## Known holes

- **338 ≠ 343.** The House feed carries 338 sitting members; five seats were vacant on 2026-09-03, three of
  them contested at the 2026-08-31 by-elections whose winners were not yet in the official roster. Do not
  invent the missing five.
- **Senators are not in this pack.** sencanada.ca renders its senator list client-side (no senator links,
  no embedded JSON, no API in the served HTML; headless Chrome returned an empty DOM) and Senate profile
  URLs are **name slugs**, not stable ids. Senators were omitted rather than keyed on a slug.
- The next-election date is a **statutory latest**, computed from s.56.1(2), and a minority Parliament can
  fall earlier. It is not an announced polling day.
- No riding-level *money* aggregation is provided: EDA and candidate money live in different streams with
  different recipient registries, and stitching them per riding would require assumptions the regulator
  did not publish.
- Rejected/spoiled ballots and elector counts exist in the poll-by-poll source but are **not** in
  `results.csv` (the header has no field for them); re-derive from the source zip if needed.
- Elections Canada licence text could not be fetched (HTTP 500 on both terms pages) — see Licences.