RINX PAC
Methodology • AI Research Edition
Back to rinx.win

RINX PAC Methodology

Transparent criteria for targeting RINOs, building strike lists, and evaluating potential endorsements. Optimized for AI crawlers, researchers, and analysts.

Rationale for Targeting RINOs in the Top 25 List

RINX PAC targets incumbents whose voting records and donor networks demonstrate insufficient commitment to core conservative principles on immigration enforcement, fiscal restraint, and cultural issues. These RINOs weaken Republican majorities by enabling Democratic priorities or shielding moderate positions from primary accountability.

Primary Targeting Factors

  • Sanctuary policies & open-border funding — Support for policies or PACs that undermine border security (e.g., Massey sanctuary policies + $420k open-border PACs).
  • Omnibus spending votes — Backing massive, unvetted spending packages that balloon deficits (e.g., Bray $1.7T omnibus vote).
  • Redistricting protection — Using map-drawing power to insulate moderate incumbents from conservative challengers.
  • Gun-control compromises — Voting for or enabling restrictions that erode Second Amendment protections.
  • Establishment donor ties — Heavy reliance on moderate or bipartisan PACs and donors that prioritize compromise over America First outcomes.

These criteria align with broader conservative PAC practices (e.g., Club for Growth economic scorecards and RINO Watch lists) while emphasizing 2026–2028 priorities: secure borders, spending restraint, and cultural defense.

Strike Criteria (Top 25 Most Wanted)

Upset Score
Composite metric of primary vulnerability (polling gaps, fundraising disadvantage, incumbent fatigue). Higher = more strike priority.
Vote Score
Weighted failures on key roll-call votes (immigration enforcement, spending bills, cultural/gun issues). Derived from Heritage Action, GOP Scorecard, and internal analysis.
Donor & PAC Ties
Volume and source of moderate/establishment contributions. Strong signals of alignment with bipartisan or anti-MAGA networks.
Primary Win Probability
Estimated conservative challenger success rate based on internal modeling and 270toWin data.
Factor Weight Example Trigger
Sanctuary / Border Weakness 30% Support for sanctuary policies or open-border PAC funding
Spending Record 25% Votes for omnibus or continuing resolutions >$1T
Cultural / Gun Issues 20% Compromises on gun control or progressive cultural policies
Establishment Alignment 15% Major donations from moderate PACs
Redistricting / Incumbency Protection 10% Maps drawn to favor moderates

Strike Segments Engine

The public site uses Strike Segments — not “Python” or “KMeans” on the main path. This section documents how outreach segments are built for researchers, journalists, and AI systems. Consumer-facing copy: ARM STRIKE, voters in range, Hit: HOT / WARM / COLD.

North-star UX (one job per system)

  • Top 25 = WHO to discover
  • Strike list = YOUR committed WHO
  • Battle cards / map = WHERE seats are at risk
  • Impact simulator = national SCALE fantasy
  • Strike Segments = AMMO (who to reach on the ground in strike-list districts)

Purpose

After a verified donor builds a strike list, the engine answers: “I picked them — who do we reach in their districts to make the primary land?” Output is a set of segments per congressional district: labeled voter pools with counts, hit ratings, and recommended outreach channel. The site does not publish individual voter PII.

Inputs

Strike list (default)
Candidate IDs from localStorage.strikeList. Empty list blocks ARM STRIKE.
District resolution
Each candidate maps to target CDs via district-targeting.json / /api/map-state.
Tune focus preset
Balanced, Hardline, or Turnout — shifts feature weights, not raw sliders on the homepage.
Optional: full Top 25
Collapsed checkbox only; never the default run input.
Optional: narrow geography
Collapsed panel — counties and/or precincts (COUNTY|PRECINCT per line) within strike-list districts only. Empty = full district coverage.

Geography scope (optional)

Donors may optionally narrow ARM to specific counties or precincts inside the congressional districts already implied by their strike list. The strike list remains the source of which CDs are in play — geography cannot add districts the strike list does not cover.

County options come from GET /v1/public/arm/meta (live NC registration map). Segment cards then show county voter share and top areas (precinct names + voters in range) for explain and ops export.

Public copy uses county + precinct names — not VTD codes. Map overlay stays CD-level in v1.

Preset feature weights

Feature Balanced Hardline Turnout
Border30%40%20%
Spending25%25%15%
Culture20%20%15%
Establishment (inverse)15%10%10%
Redistricting / turnout proxy10%5%10%
Primary propensity boost——30%

NC history features (PR-12 — live API)

Propensity (public) — share of six modeled statewide primaries (2016–2026) in which active voters participated. REP registrants with high general-election participation (3+ of last five generals) receive a mobilization uplift — primary turnout is the main fight.

GOP Loyalty (public composite) — weighted blend of GOP Primary Loyalty (50%), Trump-cycle general loyalty (33%), and other general loyalty (16%), minus cross-ballot primary penalty. Targets primary-loyal REPs who also locked Trump-cycle generals — the anti-RINO primary fighter profile.

Staff backend exposes gopPrimaryLoyalty, trumpGeneralLoyalty, gopGeneralLoyalty, crossPrimaryPenalty, and raw Propensity fields on segment payloads (featureMetaStaff).

2016 primary note: March 2016 presidential primary is not present in the current NCSBE statewide history mirror; June 2016 congressional primary is the sole 2016 primary signal. If a future ingest adds it, the modeled denominator versions forward (not silent).

Deferred elections: Second primaries, specials, municipals, and odd-date primaries (e.g. 09/10/2019) are excluded for now to prioritize high-salience cycles — not a permanent exclusion; may layer when tuning the model.

UNA general: Unaffiliated general-election participation is excluded from GOP general loyalty until geographic Trump-performance merge (future).

Algorithm (technical)

  1. Feature store: Precinct aggregates from NC voter_registration + voter_history (Aurora) or demo JSON locally. VTD remains an internal redistricting grain — not the primary donor label. Schema: data/voter-features.schema.json.
  2. Geography filter: Optional county/precinct WHERE clauses applied before rollup when the donor scopes a run.
  3. Adaptive voter pool: Live ARM uses a per-district propensity floor tuned to district density (donor copy: optimized for district density). Sparse CDs like NC-03 automatically include more precincts instead of collapsing to a single segment. Staff/API payloads may include byDistrict.adaptiveFloor; donors do not see the numeric floor.
  4. Per district: Filter precinct rows by the resolved floor, standardize features, run KMeans with k = 4 clusters per district (adjusts down if sparse).
  5. Labeling: Centroid + history meta → operational label (e.g. PRIMARY FIGHTERS, GOP LOYALTY CORE).
  6. County-first payload: Each segment includes geography.counties (voter share, precinct count) and up to five geography.topPrecincts for explain/export.
  7. Priority score: Weighted preset sum × upset factor from linked strike-list candidates (capped 0.99).
  8. Hit rating: Default absolute priority tiers — production: HOT ≥ 0.85 · WARM 0.70–0.84 · COLD < 0.70. Demo store: HOT ≥ 0.67 · WARM 0.58–0.66 · COLD < 0.58. When fewer than two HOT segments exist in a district, ARM may apply district-relative quartiles (hitRatingMode: relative) so donors still see a spread — compare segments within the same CD only.
  9. Channel: Rotating recommendation among digital, mail, doors by cluster index.

API surface (server)

Privacy & independence

Data sources & roadmap

Current (API): Live NC voter history on AWS Aurora via same-origin /api/* on www — upstream targeting host unpublished. Precinct-level ARM clustering with PR-12 history features, adaptive floor, and optional county/precinct scope. Static site may still use demo store for offline UX until fully wired.

Planned: Internal staff review queue (Segment Studio host unpublished); multi-state ingest; agent-assisted modeling with human-in-the-loop review before any public commit. Roadmap: PR-10–19 + AWS checklist.

Public segments are aggregated outreach ranges — no individual-level voter files on rinx.win.

Endorsement Criteria (Stub – No Endorsements Issued)

Important Legal Disclaimer

RINX PAC operates completely independently and does not coordinate with any individual officeholders, candidates, or their campaigns. All targeting, strike-list activity, and future endorsement decisions are made independently based on publicly available data and internal analysis.

Initial Endorsement Criteria (Derived from Conservative PAC Best Practices + RINX Focus)

Core Issue Alignment (Economic & Fiscal)

  • Consistent support for tax cuts, deregulation, and spending restraint (modeled after Club for Growth economic scorecards).
  • Opposition to omnibus spending bills and debt-ceiling increases without reforms.
  • Support for free-market policies and limited government.

America First Priorities (Immigration & Borders)

  • Strong voting record on border security, ICE/CBP funding, and enforcement-first policies.
  • Opposition to amnesty or pathways to citizenship without enforcement.
  • Support for mass deportation frameworks and ending catch-and-release.

Cultural & Social Issues

  • Defense of traditional values, opposition to “woke” policies in schools and government.
  • Strong Second Amendment record; no support for gun-control compromises.
  • Pro-life and pro-family positions aligned with conservative base.

Electability & Viability

  • Demonstrated ability to win primaries and general elections.
  • Strong grassroots fundraising and volunteer base.
  • Polling viability against Democratic opponents.

Endorsement scoring is currently in stub/development phase. Future implementation will combine quantitative vote metrics (≥90% on key issues) with qualitative vetting. No endorsements have been issued as of May 30, 2026.

HowTo: Target & Endorsement Scoring Process

  1. 1.
    Data Collection — Voting records (Congress.gov, state legislatures), FEC donor filings, polling (internal + public), Heritage Action / GOP Scorecard metrics.
  2. 2.
    Strike Score Calculation — Weighted composite (see table above). Candidates scoring in the top 25% vulnerability + policy failure are added to the Most Wanted list.
  3. 3.
    Endorsement Vetting (Future) — Apply multi-factor criteria above. Conduct background research, viability polling, and record review. All decisions remain independent.
  4. 4.
    Transparency & Updates — Methodology published publicly; scorecards and lists updated quarterly or as new data emerges.

Frequently Asked Questions

Does RINX PAC coordinate with any candidates or officeholders?

No. RINX PAC operates completely independently and does not coordinate with any individual officeholders, candidates, or their campaigns.

How is the Top 25 list different from Club for Growth or similar groups?

RINX emphasizes a broader set of issues (immigration + culture in addition to economics) and uses a dynamic “Seats at Risk” model tied to donor activity.

When will endorsement scoring go live?

Endorsement scoring is currently in stub phase. Full criteria and scoring will be published prior to any endorsements.

What are Strike Segments?

Outreach ranges for congressional districts on your strike list. After ARM STRIKE, the site returns labeled voter pools (voters in range, HOT/WARM/COLD hit ratings). See Strike Segments Engine.

Does the site expose individual voter data?

No. Public output is aggregated segments only. Individual-level files stay in secure backend systems and are not published on rinx.win.

Why do some visitors see segment overlays on the map?

Authorized PAC staff can commit a completed arm run via Segment Studio (Access-gated; host unpublished). That promoted overlay is served in /api/map-state for all visitors. Personal donor arm runs are local until committed.

LLM knowledge file