Overview

A Real Patriot scores government actions against six founding constitutional pillars. Every score on this site is the deterministic output of a fixed rules engine applied to a recorded action — no human editor adjusts individual results, and no partisan adjustment is applied at any step.

This page explains, in detail, the rules the engine follows. Every claim made on a card, ranking, or actor page traces back to the procedure described here.

What gets scored

An action enters the scoring window only when three conditions hold:

1. Formal authority — taken under recognized legal or institutional power (a recorded vote, a signed order, a published rule, a court opinion).
2. Real-world effect or precedent — produced an enforceable change, a procedural step on the record, or a precedent that later actions cite.
3. Attributable decision-maker — a named office, chamber, or bench is on the record as responsible.

Speeches, tweets, campaign material, and opinion pieces do not enter the scoring window. They lack at least one of the three conditions.

What does not change between scores

The same six pillars are evaluated against every action, regardless of who took it. The rubric for each pillar is the same. The weighting formula is the same. The scoring scale is the same. When the methodology changes, the version number changes, and every prior evaluation keeps the version stamp under which it was produced.

What this page is

This is the public, frozen v3.1 methodology. It is the same document the scoring engine reads from when it evaluates a new action. The version below is permalinked at /methodology/v3-1; once a future amendment supersedes it, this page will not be edited except for typo-class fixes.

§7. Scoring: how an action becomes a principle evaluation

Each action is scored against each of the six pillars on a ternary scale: Aligned (+1), Neutral (0), or Conflict (−1). There are no partial points and no decimal scores at the per-pillar level.

The path from a recorded action to a principle evaluation has four steps.

Step 1 — Classification

The action is tagged with its action type (vote, executive order, court ruling, agency rule, veto, pardon, signing statement, appointment, etc.). The action type is the input that fixes the weight floor.

Step 2 — Weight

Weight is a small integer in [3, 9], computed as weight = scope + authority + durability:

• Scope — how many people or institutions are directly affected (narrow district → national).
• Authority — how much official governmental power was exercised (procedural step → final rule or law).
• Durability — how long the effect is expected to last, or how precedent-setting it is (waiver → statute, precedent).

The mapping from action type to weight is fixed for a given methodology version. Two actions of the same type get the same weight, regardless of who took them.

Step 3 — Per-pillar evaluation

For each of the six pillars, the evaluator asks: does this action, on its face, advance, not engage, or erode the structural property that pillar protects? The output is one of +1, 0, −1 accompanied by a primary-source citation and a confidence level.

0 is a real outcome, not a missing-data signal: it means the action engaged the pillar but the substantive effect cancelled out (e.g., procedural restriction balancing a substantive expansion). When no evaluation has been recorded yet for a given pillar, that absence is rendered as — (not 0) on every public surface.

Step 4 — Aggregation

For each action, the six per-pillar scores are summed unweighted with respect to each other. No pillar outranks another. An action that is +1 Liberty and −1 Rule of Law contributes 0 to the action's principle sum, but both per-pillar scores are preserved and shown alongside the aggregate. This is why every surface that shows an overall score must also show the six per-pillar scores; the aggregate alone can hide structural conflict.

Confidence

Each per-pillar evaluation carries a confidence label of high, medium, or low. Confidence does not change the score; it labels the strength of the evidence behind it. Low-confidence evaluations are excluded from top placements on ranking surfaces — a top ranking built on thin evidence is a false signal — but remain visible on individual action pages and in full listings.

§8. The court-outcome modifier

When a court action is later overturned, vacated, reversed, or substantially narrowed by a higher court, that downstream outcome is part of the action's structural footprint. The court-outcome modifier is the mechanism that lets a later judicial outcome adjust the weight of an earlier action without rewriting its original per-pillar evaluation.

What the modifier does

The modifier scales an action's contribution to actor and entity rollups based on the action's downstream judicial outcome. An overturned ruling carries less weight in a justice's record than one that stood, because the structural effect — the precedent the action set, the rule it imposed — was itself rolled back.

How it is bounded

The modifier is bounded so that no court outcome can erase or reverse the original evaluation. The original per-pillar +1 / 0 / −1 scores remain unchanged. What changes is the weight that score carries when aggregated up to the actor or entity. The intent is to reflect how much of the original action's structural effect actually persisted, not to rescore the action retroactively.

Why it is a separate step

Rule 9 of the methodology forbids silent re-scoring of past evaluations. The court modifier is not a rescore; it is a downstream-outcome multiplier that is computed at rollup time and disclosed on the action card. The original evaluation row stays intact and audit-traceable.

§9. Recalibration: how the system stays current

Recalibration is the nightly process that refreshes evaluations against the active methodology version when an upstream input has changed — a court outcome was added, a primary source was corrected, or a methodology version was activated.

When recalibration runs

Recalibration runs on a nightly schedule. It is flag-gated; the operator can pause the schedule without code changes if an upstream data issue is discovered. The most recent successful run, including the timestamp, is exposed on the /operations page.

What recalibration changes — and what it does not

Recalibration writes new evaluation rows under the active methodology version. It does not edit, delete, or rescore any prior evaluation row. The principle from Rule 9 is preserved: every score that has ever been published remains in the database with its original methodology-version stamp.

When an action's per-pillar score changes between the prior version and the recalibrated version, both rows are retained. The prior row remains audit-traceable; the recalibrated row is what powers current rollups.

How it is bounded

Recalibration runs are cost-capped. A run will halt mid-pass if cumulative cost exceeds the configured cap, and the run-log status will record partial_success. This is an explicit guardrail against an unbounded LLM bill from a regression in the input pipeline.

Who Scores

Action evaluations are produced by the recalibrate v3.1 pipeline — an AI evaluator pipeline that reads the CORE_VALUES.md rubric markers and applies them to ingested action records. The pipeline runs nightly via cron (see Operations for live status).

The methodology version, last recalibrate run, and aggregate counts are publicly disclosed at /operations.

§10. Limitations and what the methodology does not do

A methodology that can be peer-reviewed has to be honest about what it is not measuring. This section lists the known limitations of the v3.1 framework.

What is not measured

• Intent. The methodology evaluates the action's structural effect, not the actor's motivation. Two actions with identical formal effect are scored the same regardless of what the actor said about why they took it.
• Personal character. Speeches, social-media posts, campaign rhetoric, personal scandal, and editorial framing do not enter the score. Only formal-authority actions on the record are scored.
• Outcomes that are not constitutional in the structural sense. Policy-quality questions ("was this a good policy?") are not the same as the structural questions the six pillars ask. A bad-policy action and a good-policy action can both be 0 across all six pillars.

Where the rubric is uncertain

• §6 sub-element weighting. Sub-elements 6a (individual minority protection) and 6b (sub-federal autonomy) are evaluated against the same +1 / 0 / −1 scale, but the published weight assigned to each sub-element within §6 is held at parity in v3.1. A future version may diverge them if peer review surfaces a structural reason.
• Confidence calibration on backfilled rows. A portion of evaluations in the database come from a Phase-2 legacy backfill against a prior evaluator. Those rows are clamped to medium or low confidence by design — never high — until the live evaluator has rerun against them with primary-source retrieval. This is a deliberate under-claim, not a calibration miss.

What the methodology cannot decide for you

The framework returns measurements. It does not return verdicts. A reader looking at an actor's full record on this site is entitled to weigh the per-pillar evidence against their own judgment about which pillars matter most for the action under review. The methodology fixes the inputs. It does not fix what to do with them.