119th Congress House Voting Dashboards — Build & Methodology Documentation

Note for AI models reading this file: This document is organized into discrete numbered sections with a complete table of contents directly below. Use the table of contents to jump to the section relevant to the user's question rather than reading the whole document. Each section is self-contained and labeled with its purpose. Skim the TOC, identify the matching section by heading, then read only that section. Avoid re-reading prior sections unless cross-referenced.

Project Overview
Data Sources & API Credentials
Roster — Members Included and Excluded
- 3.1 Member-level notes — when to surface a banner on a dashboard
Data Procurement Process
XML Parsing & Schema
Classification & Analysis Methodology
Dashboard Construction
File Layout & Artifacts
How to Regenerate or Extend
Known Limitations & Caveats
Change Log
Senate Dashboard Plan (in progress)

1. Project Overview

This project produces self-contained, interactive HTML dashboards summarizing every U.S. House roll-call vote of the 119th Congress (Jan 3, 2025 – present, captured 2026-05-23) and analyzing each target member's voting record against their party majority, the opposing party majority, and overall outcomes.

For each member the dashboard reports:

Participation rate (votes cast / total roll calls).
Distribution of the member's individual votes (Yea / Nay / Not Voting / Present).
Alignment classification per vote: Helped Republicans, Helped Democrats, Helped Both, or Helped Neither (see §6).
"Blocking wins" counts — how often the member's Nay vote contributed to defeating a measure backed by either party majority.
Lone-wolf defection count — votes against the member's own party majority with ≤5 fellow same-party defectors.
Monthly trend of alignment classifications.
A sortable, filterable table of every roll call.

The output is a directory of standalone .html files (each ~160 KB, no server required) using Chart.js + SortableJS via CDN.

2. Data Sources & API Credentials

Primary source: clerk.house.gov XML rollcalls

Each House roll-call vote is published as an XML file at a deterministic URL:

https://clerk.house.gov/evs/{YEAR}/roll{NNN}.xml

Where {NNN} is a 3-digit, 1-indexed roll number resetting each calendar year. This is the authoritative source (Clerk of the House) and includes per-member vote records, party totals, bill identifier, question, result, date, and time. No API key is required. No authentication, no rate-limit headers documented. Self-imposed throttling of 350 ms between fetches was used (see §4).

Year ranges & index discovery

The Clerk publishes year-index pages at https://clerk.house.gov/evs/{YEAR}/ROLL_{XXX}.asp (grouped in blocks of 100). These were scraped once to determine the highest roll number per year:

2025: rolls 001–362 (362 votes)
2026: rolls 001–191 (191 votes)
Total: 553 roll-call votes

Supplemental source: Congress.gov API (used for Phase 0.5 roster enrichment)

A Congress.gov API key is required for the complete-roster enrichment step performed by enrich_roster.py (Phase 0.5 of the project plan). The key is loaded from .env at runtime:

CONGRESS_GOV_API_KEY=<see .env (gitignored); rotate via https://api.congress.gov/sign-up/ if exposed>

Endpoint: https://api.congress.gov/v3/...?api_key=<key>

The clerk.house.gov XML provides every field required for the core roll-call analysis (vote breakdown, member vote, bill ID, question, result, date), so the original vote-tally pipeline does not call this API. The key is used by the roster enrichment pass to fill in members who never appear in a roll-call vote during the analysis window, and remains available for extensions needing bill subject codes, cosponsor lists, or roll-call cross-references that the Clerk XML omits.

Security note: The API key has been moved out of this document and into .env (gitignored). To rotate: (1) sign up for a replacement key at https://api.congress.gov/sign-up/, (2) drop the new key into .env as CONGRESS_GOV_API_KEY=..., (3) delete the old key from the Congress.gov dashboard. If the previously-exposed key is found in git history or redistributed copies of this doc, rotate immediately.

3. Roster — Members Included and Excluded

Included (House members of the 119th Congress)

Display Name	Bioguide	Party	Chamber	Dashboard file
Thomas Massie (KY-4)	M001184	R	House	`ThomasMassie119.html`
Ro Khanna (CA-17)	K000389	D	House	`RoKhanna119.html`
Alexandria Ocasio-Cortez (NY-14)	O000172	D	House	`AlexandriaOcasioCortez119.html`
Ilhan Omar (MN-5)	O000173	D	House	`IlhanOmar119.html`
Marjorie Taylor Greene (GA-14)	G000596	R	House	`MarjorieTaylorGreene119.html`
Jim Jordan (OH-4)	J000289	R	House	`JimJordan119.html`
Byron Donalds (FL-19)	D000032	R	House	`ByronDonalds119.html`

The three Trump-loyal Republican House members selected (Greene, Jordan, Donalds) were chosen as among the most frequently identified by political-press coverage as among Donald Trump's most consistent House allies and surrogates during the 119th Congress.

Explicitly excluded (with reason)

Requested Name	Reason for exclusion
Lindsey Graham	U.S. Senator (R-SC), not a House member. clerk.house.gov data does not cover the Senate. Senate roll-call XML lives at `senate.gov/legislative/LIS/roll_call_votes/...` with a different schema. Would require a separate fetcher; out of scope for this build.
Ron Paul	Not a member of the 119th Congress. Last served TX-14 in the 112th Congress (ended Jan 3, 2013). No 119th roll-call record exists.

3.1 Member-level notes — when to surface a banner on a dashboard

Some members have circumstances that materially affect how their voting record should be read. To prevent misinterpretation, the builder supports an optional member note — a short banner rendered directly under the dashboard header, above the KPI grid, styled as a yellow-bordered callout. The note is also preserved in source so it can be discovered when reading the roster.

When you MUST add a note

Add a member note whenever the raw numbers would mislead a reader who does not already know the member's circumstances. Trigger conditions:

Mid-Congress entry or exit — resignation, death, expulsion, appointment to executive branch, or seat-flip via special election. Any of these truncates the member's voting window relative to the full Congress. Example: Marjorie Taylor Greene's voluntary resignation effective 2026-01-05; her dashboard shows 325 of 553 votes cast and the rest as "Not Voting," which without context looks like extreme absenteeism.
Extended leave of absence — publicly reported illness, family leave, campaign for higher office, or military deployment lasting weeks+.
Party switch or change in caucus affiliation mid-Congress. The alignment/blocking metrics compute against R/D majorities and the member's own party — a switch changes what "lone wolf" means.
Speakership or leadership role that conventionally affects voting patterns (e.g., the Speaker traditionally votes only to break ties).
Vacancies, contested seatings, swearing-in delays beyond the first few days of the Congress.

When notes are optional but recommended

Member voluntarily abstains from a class of votes for stated reasons (e.g., conflict of interest, recusal).
Member is well below median participation (<80%) for reasons not otherwise documented.
Member's classification is heavily skewed by a known confounder (e.g., the chair of a committee voting in lockstep on procedural motions specific to that committee).

When you should NOT add a note

Don't add notes that editorialize policy positions ("a controversial conservative voice"). Notes are factual circumstance only.
Don't paraphrase the same caveats documented in §10 — those apply to every member and live in the methodology section, not on the dashboard.
Don't restate numbers already visible on the dashboard.

Format

Notes are passed as the optional 5th element of a ROSTER tuple inside build_member.py (House) or build_senator.py (Senate):

("G000596", "Marjorie Taylor Greene", "R", "MarjorieTaylorGreene119.html",
 "Rep. Greene publicly announced her resignation from the House in late 2025, "
 "effective January 5, 2026. This explains her substantially lower "
 "participation count (325 of 553) versus other members analyzed. Votes "
 "after her departure date are necessarily recorded as Not Voting in "
 "clerk.house.gov data."),

The text appears verbatim inside a <div class="member-note"> banner. Keep it under ~3 sentences. Lead with the fact; follow with the implication for the metrics on this dashboard.

Single-member rebuild after adding a note

python3 build_member.py G000596 "Marjorie Taylor Greene" R MarjorieTaylorGreene119.html
# (Single-member CLI mode does not accept a note — re-run the full ROSTER
#  loop with `python3 build_member.py` to pick up note changes.)

4. Data Procurement Process

The fetcher (fetch_votes.py) executed once and cached every XML locally so subsequent member-by-member analysis re-parses from disk (no re-fetch).

Steps performed:

Year-index discovery — fetched https://clerk.house.gov/evs/{YEAR}/ROLL_{000,100,200,300}.asp for each calendar year, grep'd rollnumber=NNN parameters, took the max.
Per-vote download — looped year ∈ {2025, 2026} × roll ∈ [1..max_roll]:
- URL: https://clerk.house.gov/evs/{year}/roll{roll:03d}.xml
- Cache path: vote_cache/{year}_{roll:03d}.xml
- Skipped if cached file already exists and exceeds 200 bytes.
- Sent a User-Agent: Mozilla/5.0 (research; polisci-analysis) header.
- Throttle: time.sleep(0.35) between successful network fetches (≈2.9 req/s), well under any conservative rate-limit threshold.
- Failures logged to stderr, loop continues (no votes were lost; all 553 fetched).
Total payload: 553 XML files, ~14 MB on disk under vote_cache/.

Total wall-clock time for the cold fetch: ≈3.5 minutes (553 × 0.35 s + transfer).

5. XML Parsing & Schema

Each roll{NNN}.xml follows DTD vote v1.0 20031119. Key extracted fields:

<rollcall-vote>
  <vote-metadata>
    <majority>            R or D
    <rollcall-num>        e.g. 47
    <legis-num>           e.g. "H R 1234"
    <vote-question>       e.g. "On Passage"
    <vote-result>         e.g. "Passed" / "Failed" / "Agreed to" / "Rejected"
    <action-date>         e.g. "3-Jan-2025"
    <vote-desc>           short bill title
    <vote-totals>
      <totals-by-party>   { party, yea-total, nay-total, present-total, not-voting-total }  × R/D/I
  <vote-data>
    <recorded-vote>
      <legislator name-id="M001184" party="R" state="KY" ...>Massie</legislator>
      <vote>Yea | Nay | Aye | No | Present | Not Voting</vote>
    ... (one per legislator)

Aye/No are emitted for procedural questions; Yea/Nay for ordinary passage. Both are normalized to Yea/Nay for analysis purposes.

A member's vote is looked up by exact match on legislator/@name-id against the target Bioguide ID. If the member did not vote, the field is absent entirely and we record None (rendered as "absent" in classification).

6. Classification & Analysis Methodology

For every vote, the analyzer determines each party's majority position:

party_position = Yea  if  yea > nay
               = Nay  if  nay > yea
               = Split otherwise (tie or zero)

6.1 Alignment classification

Per vote, the member's normalized vote (Yea/Nay) is compared to each party's majority position:

"Helped Both" arises on bipartisan votes where both party majorities aligned (common on naming-a-post-office bills, suspension-calendar items, etc.). "Helped Neither" arises when member is on the losing side relative to both party leaderships — usually a small protest/defector cluster.

6.2 Blocking analysis

A "blocking win" is recorded when:

The member voted Nay, AND
The measure failed (result matches fail, reject, not agreed, not passed), AND
The other party's majority was on the opposite side (i.e., it was a partisan vote, not a bipartisan defeat).

blocked = "Democrat" if Dem majority was Yea and Rep majority was not Yea (i.e., Dems backed it, member's Nay helped sink it).

blocked = "Republican" if Rep majority was Yea and Dem majority was not Yea (i.e., GOP backed it, member's Nay helped sink it).

This metric attributes a single "share" of credit to the member for the defeat, regardless of margin. Members with many such tallies are disproportionately blocking their own caucus's agenda (notable for Massie).

6.3 Voted-with / voted-against by party majority

Across votes where each party's majority took a definite position (not Split), count how many times the member's normalized vote matched (with) or differed (against) that party's majority. Reported as KPI cards plus a stacked bar chart with raw counts and percentages.

6.4 Lone-wolf defection

A vote is a lone-wolf defection if all of the following hold:

The member's own party (R or D, per roster) had a definite majority position.
The member's normalized vote opposed that majority.
≤5 fellow same-party members also defected (i.e., the member was part of a very small dissenting bloc within their own caucus).

This identifies the "stubborn outliers" within a caucus.

6.5 Monthly trend

Each vote's action-date is parsed (DD-Mon-YYYY) and bucketed by YYYY-MM. The four primary alignment classes are summed per month and rendered as a multi-series line chart.

7. Dashboard Construction

7.1 Template strategy

build_member.py holds a single HTML template string with __PLACEHOLDER__ tokens. The Python builder fills in the data, embeds the full per-vote JSON payload (typically ~150 KB) inline, and writes the result to results/<Member>119.html. The output has no runtime dependencies on local files — Chart.js and SortableJS load from public CDNs.

7.2 Libraries used (CDN)

Chart.js v4.4.0 — doughnut, bar, stacked bar, horizontal bar, line charts.
SortableJS v1.15.2 — drag-and-drop reordering of KPI cards. Card order is persisted to localStorage under a per-bioguide key, so each member's dashboard remembers its own layout.

7.3 Layout

Header (member name + party pill + bioguide + date range)
Drag hint + reset-order button
KPI grid (6 reorderable cards): Roll-Calls/Participation, Voted-Against-GOP, Voted-Against-Dem, Blocked-Dem-backed, Blocked-GOP-backed, Lone-Wolf Defections — all with raw counts + percentages.
Chart row 1: Alignment doughnut · Vote-distribution bar (with inline labels).
Chart row 2: Voted-with-vs-against stacked bar · Blocking horizontal bar.
Chart row 3: Monthly alignment trend line.
Vote table: sortable, filterable (free-text search + dropdowns for alignment / blocking / member's vote).
Footer with source attribution and methodology pointer.

7.4 Defensive fixes applied

Each <canvas> is wrapped in a position:relative; height:300px container to prevent Chart.js's responsive-resize loop from infinitely growing the page.
.card { min-width: 0 } so CSS grid columns can shrink properly with long content.

8. File Layout & Artifacts

The tree is split into a working /data/ area (raw caches, intermediate JSONL, per-member metric files) that the pipeline reads and writes, and a /results/ area that is the actual embeddable artifact shipped to hosts. Everything under /results/<C>/ is self-contained — no external network calls at runtime, all vendored — while /data/ retains the upstream caches and build metadata needed to re-derive results from scratch.

polisci/
├── DOCUMENTATION.md            # this file
├── NOTES.md                    # deferred concerns; see file
├── PROJECT_SCOPE.md            # PM-owned scope record (created/updated by PM agent only)
├── CLAUDE.md                   # Claude Code project context
├── Methodology.md              # end-user methodology doc (copied into results/<C>/)
├── .env                        # gitignored — CONGRESS_GOV_API_KEY=...
├── .gitignore
├── .claude/skills/bulk-update/SKILL.md  # project skill: post-change PM+docs+commit orchestration
├── fetch.py                    # idempotent network fetch
├── parse.py                    # XML → votes.jsonl + roster.json (+ merge w/ Congress.gov directory)
├── analyze.py                  # pure analytics; classify_vote + aggregate
├── enrich_roster.py            # Congress.gov roster pull + LIS↔bioguide crosswalk
├── build_members.py            # parallel per-member JSON build
├── build_app.py                # template → results/<C>/ embeddable artifact
├── build_all.py                # one-command orchestration
├── tests/                      # pytest unit tests for analyze.py + parity_check.py
│   ├── fixtures/*.xml
│   ├── test_analyze.py
│   └── parity_check.py
├── template/
│   ├── app.html, app.css, app.js
│   ├── compare.html, compare.js
│   ├── ranking.html, ranking.js
│   └── vendor/
│       ├── chart.umd.min.js    # Chart.js 4.4.0
│       └── sortable.min.js     # SortableJS 1.15.2
├── data/119/
│   ├── house/{cache/, votes.jsonl, roster.json}
│   ├── senate/{cache/, votes.jsonl, roster.json}
│   ├── members/<id>.json       # per-member metrics
│   ├── manifest.json           # member index for picker
│   ├── members_directory.json  # Congress.gov roster (~551 members)
│   ├── lis_to_bioguide.json    # Senate ID crosswalk
│   ├── api_cache/              # cached Congress.gov responses
│   └── build_report.json
├── results/119/                # embeddable artifact — what ships
│   ├── app.html, compare.html, ranking.html, app.js, compare.js, ranking.js, app.css
│   ├── vendor/{chart…, sortable…}
│   ├── data/{manifest.json, members/<id>.json}
│   ├── Methodology.md          # copied at build time by build_app.py
│   └── README.md
└── legacy/                     # archived pre-pivot single-member dashboards

9. How to Regenerate or Extend

Full pipeline (step by step)

python3 fetch.py          --congress 119      # idempotent; near-zero work if cache populated
python3 parse.py          --congress 119      # XML → votes.jsonl + roster.json
python3 enrich_roster.py  --congress 119      # Congress.gov API → members_directory.json + lis_to_bioguide.json
python3 parse.py          --congress 119      # re-run: essential — re-does directory-merge now that members_directory.json + lis_to_bioguide.json are populated
pytest tests/                                 # gate: classifier behavior frozen
python3 build_members.py  --congress 119      # parallel: writes per-member JSON + manifest + build_report
python3 build_app.py      --congress 119      # template/ + vendor + data → results/<C>/

One-command equivalent

python3 build_all.py --congress 119

Script reference

fetch.py — downloads House (clerk.house.gov) and Senate (senate.gov/LIS) roll-call XML into data/<C>/{house,senate}/cache/. Idempotent; skips files already on disk.
parse.py — parses cached XML into votes.jsonl and roster.json per chamber. If members_directory.json exists, merges it into the roster so members who never cast a vote are still listed.
enrich_roster.py — pulls the canonical Congress.gov member directory and emits members_directory.json plus the lis_to_bioguide.json Senate-ID crosswalk. Requires CONGRESS_GOV_API_KEY in .env.
analyze.py — pure functions (classify_vote, aggregate) shared by build_members.py and the parity tests. No I/O.
build_members.py — parallel per-member metric computation; writes data/<C>/members/<id>.json, manifest.json, and build_report.json.
build_app.py — copies template/ + vendored libraries + the built data into results/<C>/ — the self-contained, embeddable artifact.
build_all.py — orchestrates the seven steps above with a single --congress argument.
pytest tests/ — gate run between parse and build; freezes classifier behavior and includes the legacy 8-member KPI parity check.

Extending to a new Congress

python3 build_all.py --congress 120

No code changes required; the pipeline is parameterized end-to-end on --congress.

10. Known Limitations & Caveats

House only. Senate votes are not covered; senators (e.g., Lindsey Graham) would require a parallel fetcher targeting senate.gov/legislative/LIS/....
Roll-call votes only. Voice votes, unanimous-consent agreements, and motions adopted without a recorded vote are invisible to this analysis. A member's silence on a controversial measure that passed by voice vote cannot be detected here.
Bill-subject classification is not attempted. Alignment counts treat procedural votes (e.g., motions to recommit), naming bills, and major policy votes equivalently. Heavy weighting of procedural calendar votes can inflate "Helped Republicans" / "Helped Democrats" counts vs the substantive picture.
"Helped Both" interpretation. A bipartisan vote that passes overwhelmingly is genuinely the member helping both sides; it is not noise — but it can dilute the visual share of the more interesting partisan classes.
Blocking-wins attribution. Each blocking tally credits the member individually for a defeat that involved hundreds of other Nay votes; the metric is a count, not a marginal causal estimate.
Lone-wolf threshold (≤5) is a heuristic. Tighter (≤2) would isolate true singletons; looser (≤15) would capture organized defector groups. Adjust in aggregate() in build_member.py if needed.
Member resignations / mid-term entries are not flagged explicitly in data. A low participation count may reflect resignation, illness, or running for higher office — check the underlying date pattern in the votes table. (See §3 note on MTG's announced 2026 resignation.)
No correction for "Aye"/"Yea" semantic difference. Both are normalized to Yea for analysis; the distinction (passage vs. procedural) is preserved in the per-vote table column.
Data freshness. Snapshot taken 2026-05-23. Re-run fetch_votes.py and build_member.py to refresh.

11. Change Log

Date	Change
2026-05-23	Initial Massie dashboard built (553 votes, 6 KPIs, 5 charts, filterable table)
2026-05-23	Fixed Chart.js infinite-resize bug by wrapping canvases in fixed-height containers
2026-05-23	Added percentages to doughnut legend + tooltip
2026-05-23	Added inline count + % labels above each bar in vote-distribution chart
2026-05-23	Added "Voted Against GOP/Dem Majority" KPI cards + with/against stacked bar chart
2026-05-23	Added blocking-wins horizontal bar; percentages on all KPI cards
2026-05-23	Added SortableJS drag-and-drop card reordering with `localStorage` persistence
2026-05-23	Merged "Total Roll Calls" and "Massie Voted" into single Participation card
2026-05-23	Parameterized builder (`build_member.py`); generated dashboards for 6 additional House members
2026-05-23	Wrote DOCUMENTATION.md; moved Massie dashboard to `results/ThomasMassie119.html`
2026-05-24	Added §3.1 member-notes guidance; rendered MTG resignation banner on her dashboard
2026-05-24	Wrote Senate fetcher + builder; generated `LindseyGraham119.html` (see §12)

2026-05-24 — v1.0.0 Interactive SPA rewrite (`build_*`, `template/`, `results/`)

Replaced the 8 standalone dashboards in legacy/ with a parameterized pipeline producing a single interactive SPA covering every 119th-Congress member (552 total: 449 House + 103 Senate). Single-page member picker with searchable typeahead + sidebar filters; comparison view overlays up to 6 members across 5 charts; URL-deep-linkable; framework-free and embeddable into third-party hosts via standalone, iframe, or inline modes (see results/119/README.md). Roster completeness now sourced from the Congress.gov API (enrich_roster.py); the Congress.gov API key moved from §2 of this document into .env (gitignored). Phase 3 KPI-parity gate confirmed 8/8 legacy members reproduce exactly.

Generalizes to future Congresses via --congress N.

Known limitations carried forward to v1.1: see NOTES.md.

2026-05-24 — v1.1.0 Rankings page, structural banners, file:// support

Three rounds of post-launch enhancements:

Rankings & UX (Round A) — New ranking.html / ranking.js page sorts House or Senate members by any of 14 metrics with chamber/party filters and shareable URL state. Manifest now carries per-member KPI dict (k) so the page needs a single fetch. build_app.py inlines the full manifest into both HTML heads, restoring file:// support for the picker (per-member JSON still requires HTTP). Fixed a Chart.js infinite-growth feedback loop by giving .chart-frame a fixed height and wrapping each canvas in a .chart-canvas-wrap with position: relative + flex: 1 1 auto. Two new KPI tiles ("Voted With GOP" / "Voted With Dem", now 8 total). Page footers now link to the underlying clerk.house.gov / senate.gov XML and to a new Methodology.md (end-user methodology doc, also copied into the shipping artifact). New CLAUDE.md (Claude Code project context).

Delegate banner (Round B) — Detected the 6 territorial House delegates (AS, DC, GU, MP, PR, VI) by USPS code; parse.py overrides the vote XMLs' "XX" state code with the directory's real code. Per-member JSON gets an is_delegate flag; app.js renders a yellow banner explaining that delegates may vote in committee and on Committee-of-the-Whole amendments but not on House final passage — their low participation rate is structural, not absenteeism.

Structural banners for member-elect / replaced / died (Round C) — enrich_roster.py now does three extra passes beyond the original bulk fetch: (1) Rescue — individual /v3/member/{bg} lookups for any vote-derived House bioguide missing from the per-Congress bulk listing (recovers Matt Gaetz, FL-1, never-seated member-elect, plus his full name and term history); (2) Replacement-linking — pairs predecessor↔successor by (state, district) within the Congress window, emitting replaces / replaced_by bioguide refs; (3) Detail-enrichment — individual lookups for every member on a replacement chain to get accurate per-Congress congress_term (startYear/endYear/district) and death_year (the bulk listing only carries chamber + startYear). parse.py and build_members.py propagate these into the per-member JSON; app.js renderNote() branches by status with priority delegate > unseated > died > replaced_by > replaces

served_partial. Predecessor and successor names render as in-app links via manifestById lookup. 8 House replacement pairs auto-linked in the 119th (Gaetz→Patronis FL-1, Waltz→Fine FL-6, Grijalva R.→Grijalva A. AZ-7, Turner→Menefee TX-18, Connolly→Walkinshaw VA, Greene→Fuller GA, Green→Van Epps TN, Sherrill→Mejia NJ).

Developer tooling (Round D) — .claude/skills/bulk-update/SKILL.md defines a project-level Claude Code skill that orchestrates the full post-change cycle: PM updates PROJECT_SCOPE.md, parallel programmer agents sync CLAUDE.md and DOCUMENTATION.md, then git add/commit/push.

12. Senate Dashboard Plan (in progress)

This section captures the design for extending the House pipeline to U.S. Senators, plus the implementation choices for the first build (Lindsey Graham, R-SC).

12.1 Data source

Senate roll-call XML lives on senate.gov, not clerk.house.gov:

Index per session: https://www.senate.gov/legislative/LIS/roll_call_lists/vote_menu_119_{S}.xml where {S} is 1 (2025) or 2 (2026).
Per-vote XML: https://www.senate.gov/legislative/LIS/roll_call_votes/vote119{S}/vote_119_{S}_{NNNNN}.xml (5-digit zero-padded vote number — note: different padding from House rolls, which use 3 digits).

No API key, no authentication. Same 350 ms throttle policy applies.

12.2 Vote counts (snapshot 2026-05-24)

Session 1 (2025): 659 votes
Session 2 (2026): 130 votes
Total: 789 Senate roll-call votes (≈43% more than the House count for the same Congress, driven by Senate's heavy nominations calendar).

12.3 Schema differences vs. House

Field	House (`clerk.house.gov`)	Senate (`senate.gov`)
Root element	`rollcall-vote`	`roll_call_vote`
Per-party totals	`vote-totals/totals-by-party`	Not present — must aggregate from per-member records
Member ID	Bioguide (`name-id="M001184"`)	LIS (`lis_member_id`, e.g. `S293`)
Vote element	`<vote>Yea</vote>`	`<vote_cast>Yea</vote_cast>`
Member party	attribute on `<legislator>`	child `<party>` element
Date format	`3-Jan-2025`	`January 9, 2025, 02:54 PM`
Vote question	`<vote-question>` + `<vote-desc>`	`<question>` + `<vote_title>` + `<vote_document_text>`
Result	`<vote-result>` (e.g. "Failed")	`<vote_result>` (e.g. "Cloture on the Motion to Proceed Agreed to")

12.4 Architectural plan

Two new scripts, mirroring the House pair:

fetch_senate.py — fetches vote_menu_119_{1,2}.xml, discovers max vote number per session, then loops to download every per-vote XML to senate_vote_cache/{S}_{NNNNN}.xml. Idempotent like the House fetcher.
build_senator.py — parses cached XML, aggregates per-party totals from member records, runs the same classification logic as the House builder (§6) so the output is methodologically comparable across chambers. Emits HTML to results/<Senator>119.html using the same template (substituting "House" → "Senate" in the header and source attribution).

12.5 Key implementation choices

Party totals from members. For each vote, walk <members>/<member> and increment {R,D,I} × {yea,nay,present,not_voting} based on <party> and <vote_cast>. Independents are tallied separately but classified by which caucus they conference with (Sanders, King → D for majority-position computation, since they reliably caucus with Democrats).
Vote vocabulary. Senate uses Yea/Nay/Present/Not Voting; no Aye/No distinction. No normalization needed.
Bill identifier. Compose from <document>/<document_type> + <document_number> (e.g. "S. 5"). For nominations, fall back to <vote_title> (which contains "Motion to Invoke Cloture: ...").
Blocking analysis for nominations. The Senate's heavy nominations load means many "blocking" results would be against confirming a presidential nominee. The same classification rules (§6.2) apply; readers should remember a "blocked Dem-backed" Senate measure during a Republican trifecta is rare by construction, while "blocked GOP-backed" is more common with a slim majority and a 60-vote cloture threshold.
Member identification. Roster entries use LIS IDs, not bioguide. Map of LIS → bioguide can be added later if cross-chamber joining is needed. For Graham: S293 (LIS), G000359 (bioguide).

12.6 Initial build target

Display Name	LIS ID	Party	Chamber	Dashboard file
Lindsey Graham (SC)	S293	R	Senate	`LindseyGraham119.html`

12.7 Future Senate roster candidates

Suggested next senators (not built yet):

Bernie Sanders (I-VT, caucuses D) — independent benchmark
John Fetterman (D-PA) — heterodox Democrat, useful comparison to Massie
Susan Collins (R-ME) — most-defection-prone Republican
Rand Paul (R-KY) — Senate counterpart to Massie's libertarian-leaning record
Ted Cruz (R-TX) — Trump-aligned Senate Republican

Add via ROSTER in build_senator.py once written.

12.8 Reuse of House template

The HTML template, CSS, and JS in build_senator.py are intentionally identical to build_member.py so dashboards are directly comparable. The only structural changes:

Header subtitle: "Senate roll-call votes" instead of "House roll-call votes."
Source attribution: senate.gov instead of clerk.house.gov.
"Voted Against GOP/Dem Majority" denominators use party tallies derived from member-level aggregation (functionally identical, computed differently).

This is intentional: the comparative value of these dashboards depends on consistent visual + methodological treatment across chambers.

DOCUMENTATION.md 35 KB História Raw