THE EQUATION OF THE LAW

What we measure

One thing: the drift between what the Constitution's text promises and what documented practice delivers, for a given institution, weighted by how many people it affects. We measure against the written text only, rather than morality, the founders' intent or any individual. That distance can be checked, which is the whole point.

Two layers, two standards

Layer 1 — Scoring an institution

The smallest measurable institution affected by the Constitution that impacts a group of people is the atom. Each documented action it takes is scored on three axes:

The direction test (show your work)

Every direction call must walk these four steps, with a source. No source, no score:

1. Which written clause is at issue? Name the exact text.
2. What does the text plainly promise? State it in plain language.
3. What did documented practice do? Cite the sourced event.
4. How far apart are they? That signed distance is the direction.

50 is the neutral baseline. Near 100 is citizen-side; near 0 is captured. Colour bands: ≥65 green, ≤35 red, everything between renders brown — and any institution carrying a contested record renders brown through the middle range regardless, so a disputed question never displays as a settled verdict.

Three gates before a record can move the number

Written out plainly, because these are the rules that decide what counts:

1. No source URL, no score. An event without a retrievable source is dropped before the sum is taken. It is not weighted down — it is not counted at all.
2. Contested contributes zero. A record marked contested is displayed with both sides and multiplied by 0. It is in the file and visible; it does not move the number.
3. No scored records means unmeasured, not neutral. An institution with records but nothing a human has scored returns no value and shows grey. It never returns 50, because a missing judgment shouldn't look like a middle score.

Layer 2 — Amendment deviation load · SPECIFIED, NOT YET COMPUTED

This layer is designed and documented but is not implemented in the live build. No amendment currently carries a deviation load on the site. We describe it here so the specification is public before the code exists.

Some amendments are built in ways that let bad actors wear down others. We capture that, but we never give an amendment a moral score, because that would be our opinion dressed up as data. Instead:

The amendments end up ranked by deviation load, so the idea that not all amendments are equal shows up in the numbers. The ranking comes from sourced data; we don't set it by hand.

Layer 3 — The fractal rollup

Each tier above the institution aggregates its children. Population weighting applies where a population is known; where it is not, children count equally. The exact behaviour, tier by tier:

Why state law does not colour the state yet (R1). A statute binds every agency in the state on paper, but how far that power reaches the ground is exactly what this map has not measured. Counting a statute at the weight of the whole state assumes perfect reach; counting it together with the local conduct it governs counts the same power twice. Local institutions are where power is expressed on people, so for now they set the colour, and state law is shown in its own tier, never hidden. The intended next step (R2) measures reach from the evidence itself, by comparing local records that operate under a statute with those that do not, and switches on only once enough such records exist. Its threshold will be published here before it is used.

How a method change works. Every human judgment (direction, weight, confidence, clause) is stored on its record and never rewritten by a method change. Every number on the map is recomputed from those judgments each time the map loads. So a corrected method re-scores the whole map at once, and the record of what changed is this page's changelog.

Unmeasured children are left out of the average rather than counted as zero or fifty, which is why coverage is shown beside every number. Change one sourced record and the number updates all the way up. The end goal is an all-green civic map, meaning majority constitutional compliance, with the remaining flaws flagged for amendment where everyone can see them. Winning means accurate numbers and honesty about the gaps.

Specified, not yet implemented

Two transparency modifiers appear in this methodology's design and are not in the deployed formula. They're listed here so nobody mistakes them for live features:

Each will ship with a dated changelog entry when it does. Until then, the number you see is produced by the formula in Layer 1 and the rollup in Layer 3, and nothing else.

Who scores this

Every score on this site was assigned by a human being. No model has ever set a direction, a weight, or a confidence value here, and none ever will. AI is used to locate documents and extract structure from them — the gathering stages return direction: 0, weight: 0, confidence: "" by hard rule, and a person decides every judgment afterward. If this corpus reaches a million records, that means a person read a million documents. There is no model drift in the scores because no model has ever touched them.

The weakness sits right next to that strength: right now the scorer is one person. One reader's judgment is doing work that should be shared, and no method statement fully fixes that. Three things are meant to address it:

1. Every score is checkable. The clause, the promise, the practice, and the source sit on the record beside the number. You do not have to trust the scorer; you can read what they read.
2. Contest it directly. Any record can be disputed with counter-evidence. You become the plaintiff and this site becomes the defendant. Disputes stay on the record where everyone can see them.
3. Agreement gets measured and published. As independent reviewers score the same records, the rate at which they agree becomes a published number. A methodology that cannot be reproduced by a second reader is one person's opinion, and it should be reported as such until proven otherwise.

How it stays honest

Crowdsource facts, publish judgment. Verified locals may update checkable facts — election dates, who holds a seat, whether a vote happened. The scoring method itself stays centralized and published. Facts crowdsource well; judgments crowdsource badly until the method is proven.

Unsourced contributions are inert. They can never reach the score. The structure makes bad data harmless rather than chasing it after the fact.

No score is automated. Discovery can be automated; judgment cannot. Every gathering tool on this project is forbidden from returning a score, and that rule is enforced in code rather than by instruction.

The algorithm is open source. New AIs, scholars, and citizens are invited to break it and improve it. When someone proves a better method, the equation changes. Being open is how we stay honest.

Changelog

Every change to the formula and to this page, dated. If a method changes without a record, nobody can tell whether it was improved or tweaked toward a preferred answer.