01Overview — what is in the box
Eight sections built from shared keys, all regenerable from open-data CSVs. Every number traces to a report + CSV in the source repo.
02Money flows vs lobbying intensity
Atlantic Canada year by year. Lobbying moves on the registry; the money is what actually arrived. Toggle normalization to compare shapes, not scales.
The full overlay lead_lag.py · series_annual_atlantic.csv
Grant money tends to follow lobbying spikes by about two years — the numbers move together strongly (r=0.81; r=0.60 even after stripping the shared upward trend, so it is not just "everything grew"). It is a pattern, not proof: behind every grant sits an ordinary paperwork story on its face.
Where the Atlantic money sits
Federal dollars actually delivered, by channel.
Grants: $84.8B handed out (2005–2026) · Contracts: $33.65B awarded to Atlantic companies (21,003 deals) · University research: $2.92B · of which ACOA: $8.2B.
Lobbyist donors — who receives the money
Who ends up with the money: the party's central fund, the local riding association, or a candidate personally. This shows where donations landed — not what donors think.
$53.0M from 10,264 registered-lobbyist keys, 2004–2026 ·
rebuilt from the correct column 2026-10-31 (fixes in
analysis/corrections.md).
Giving around the exit date — controlled exit_lag.py · exit_lag_bins.csv
243 officials-turned-lobbyists with exits 2007–2023, equal 3-month bins and placebo exit dates. Giving ramps up before leaving, drops after, then partly recovers.
The old "15–18-month spike" was a binning and truncation artifact: with equal bins and exits inside the donation window, the post-exit rebound only returns giving to its pre-exit level. It is a dip, not a reward spike. Name keys remain unverified; a pattern, not a charge.
03The revolving door
From "met" to "registered" to "on the ethics books" — each step a registry fact, none a charge.
1 · Government gets lobbied
Across 353,940 logged lobbying contacts, 25,582 government officials are named by name. 24% of registered lobbyists self-declare that they used to be a government official — mostly ex-House-of- Commons staff.
2 · Officials become lobbyists
1,235 officials later sign up as lobbyists; 624 did so within two years. Some pivots are astonishingly fast — one staffing-agency official had 164 lobbying contacts, then registered as a lobbyist 3 days after the last one.
3 · The circle closes
For 91 of those people, the organization they met while they were in office later shows up on their own lobby file — and 381 are double-checked against government paperwork.
Cooling-off exposure — appointments vs first registration oic_cooling.md
Out of 23,365 official appointment records, 844 of the appointees later turn up as registered lobbyists. Checking 431 of those handoffs closely: 183 started lobbying again within a year. All of it legal — this is the width of the lane, not a violation of any rule.
The speed of the door dpoh_lobbied.py · dl_office_verified.csvregistry-published facts
Each dot is one official who later registered as a lobbyist (381 in total). The scale spreads out small numbers so the fastest pivots are visible. Red band = registered within a month of leaving. Orange dots also contacted their old office after registering.
All of this is public in the lobbying registry — the chart only shows what the registry itself discloses. Hover any dot for the person, office, contact count and the clients they filed for.
The door as a share of new registrants revolving_2026.py · revolving_years.csvself-declared flag
Every named lobbyist row asks whether the person is a former designated public office holder (DPOH). The ex-DPOH share of newly appearing lobbyist ids roughly tripled from 2.1% in 2017 to 7.0% in 2025.
Full years only: the flag is effectively populated from 2017 (84% blank for 2015 first appearances), and 2026 is partial to 4 May. The flag is self-declared and id-level, not an adjudicated finding. Hover a point for the underlying counts.
04New Brunswick deep-dive
The home province through every layer: 158 registered clients, $15.64B in grants, $5.59B in contracts, $13.8M of riding donations.
NB officials under the ethics registry ethics_crawl.py · eth_nb.csvgroup B = office-linked, not residents
Group A = the 10 actual NB Members of Parliament (matched by riding). Group B = staff tied to ACOA / New Brunswick offices — they are linked, not necessarily residents. "Disc." = how many disclosure documents they filed (assets, debts like mortgages, sponsored travel, gifts). The ✓ columns flag same-name matches only — they are not identity confirmations.
NB donors by receiving party (top name keys) name-level
From the 6.27M-row Elections Canada file, NB-region keys with collision tiers (nb_deep.py).
NB's most-lobbied named officials nb_targets_dpoh.csv
Named government targets inside NB clients' contact logs.
05Case files: the two funding giants
NB-linked organizations whose money moves through channels the lobbying registry barely sees — with honest measure notes.
Medavie — the quiet contractor
$1.76B of federal contracts across just 13 references (2009–2025) while the lobbying footprint is one registration, zero communications, zero declared funding. A payment channel, not an influence channel.
VAC = Veterans Affairs treatment-benefit health-claims processing · IFHP = Interim Federal Health (IRCC) · DND = base paramedics via Medavie Health Services / EMT Inc.
Irving — the procurement whale
$15.76B of contracts across 220 awards with Irving-named suppliers — $14.7B of it shipbuilding. The comparison numbers are different measures, not a discrepancy: the registry reports advocacy; the contracts report the dollars.
Declared G&C (registry filings, received+expected, deduped) vs actual G&C receipts ARE comparable — both grants channel. The $15.76B belongs to procurement and must not be set against them (corrections.md).
06People & boards
CRA's charity-return open files list board members for all 83,727 registered charities (568,739 board seats, 2024 snapshot). We matched member names against lobbying/officials names — purely name-level, capped to rare names before anything is quotable.
Same-name matches only — and how careful we are with them common names collide nationally
A board seat is not evidence of lobbying. Common names like John Smith exist by the hundreds nationally — only rare-name matches are quotable; the NB table below shows exactly that rare subset.
NB charities connected by board members — the quotable subset charity_boards.py · cb_overlap_nb.csv
Only 162 of NB's 669 board-name matches are rare enough nationally to be quotable (2 or fewer people in the whole country share that name). Even then, verify with a second source before naming anyone.
Where research money lands (Atlantic top 8) university_money.md
From the three federal research agencies; Dal $1.35B leads the region, Memorial $0.51B, UNB $0.31B.
Bills that pull the most lobby registrations bills_lobby.py
The proposed laws that pull in the most lobbying paperwork.
07Networks — the connection graphs
Three node-link views, all fully interactive: drag nodes, scroll to zoom, drag background to pan, click a node to isolate its edges, type to search. Everything is name/registry-level data — hover for provenance.
Exploreforce layout, hand-rolled physics
08The current registry, 2022–26
The core atlas above uses the 2022 extract. This section uses the 2026 cumulative dump — communications to 16 October 2026, registrations to 4 May 2026 — and reads it through the same Atlantic lens. It is the same registry, keyed consistently enough to stitch.
A record year, then a higher floor post2022.py · post2022_fy.csv
Communications counted in the extract, by April–March fiscal year. The 2025–26 bar is the record; the office's own report verifies 39,879 against 40,113 in the extract (0.6% gap).
2026–27 is omitted because it is a partial year. The report attributes the 2025–26 rise to the April 2025 election and the threshold cut from 32 to 8 hours over four weeks, in force 2026-01-19.
Who registered after the cut threshold_mix.py · threshold_new_profiles.csv
New registration profiles in the 3.5 months after the cut, against the same calendar window a year earlier. Consultants account for most of the absolute rise, but every category grows.
In-house organizations 50 → 150; corporations 42 → 110; grant/benefit work rises from 33.5% to 40.5% of new profiles. The post-cut wave names 1,007 clients and 232 firms. Atlantic new-profile share rises from 0.9% to 2.1%, even as Atlantic communications share falls from 2.4% to 2.0%.
Atlantic Canada's share of logged lobbying prov_universe.py · prov_comms_years.csv
The four Atlantic provinces together, as a share of national communications. The dashed line is the region's 6.5% share of the population.
The gap is structural, not a one-year artifact: Atlantic clients were 2.5% of national comms in 2024 and 2.3% in 2020. Ontario alone is about 59%. 2026 is January–October; hover for the Ontario comparison and the raw count.
Appointees who also donate name keys only
A name-key join between 23,365 OIC appointment orders and the Elections Canada individual-contribution roll-up. This is a candidate list, not a finding.
Nationwide: 5,190 matched keys, 538 unique/tight, 619
also registered lobbyists. A name key can merge two people; loose
common-name matches are excluded from the quotable set. No appointment
date is compared with a donation date here, so this does not show that
donations bought appointments. Full table: gic_donors.csv.
Who Atlantic clients target in Ottawa post2022_targets.py · post2022_targets_inst.csv
Share of the 3,456 Atlantic communications naming each institution, 2022–26, against the national mix. One communication can name several targets, so shares sum above 100%.
ACOA is 15.1× its national share, DFO 4.6×, DND 2.6×, PSPC 2.3×, Infrastructure 2.6×. The House of Commons is still the largest single target (23.8%) but under-indexes at 0.8×. New Brunswick is DND/PSPC-heavy. Filing channel: 57.5% of Atlantic comms are consultant-filed vs 35.7% nationally; in-house organizations are 11.7% vs 42.0%. Hover a dot for counts; small denominators make these descriptive, not statistical tests.
Scope note: registrations lag communications in this dump, so the post-cut registration effect is measured only to 4 May 2026, while communications run to 16 October. Provincial figures are floors: about 5% of communications still have no usable client address, and new profiles often lack one.