OwnNothing Consent Scanner: Scoring Methodology Whitepaper

Section 1: Header and version

Product: OwnNothing Consent Scanner Score name: the Own Nothing Score (locked 2026-07-12) Whitepaper version: 0.1 (draft for board and AI-counsel review) Methodology version (pinned): engine_v3 Owning board: OwnNothing Contract-Scoring Advisory Board (Ralph Nader anchor seat; Lina Khan, Max Schrems, Matt Stoller, Cory Doctorow, EFF policy seat, Jony Ive design seat, Aaron Day framework seat) Status: draft. Not published. One hard dependency remains (Legal Analysis Board clause-type review + defensibility ruling, run as AI counsel).

Every score the scanner displays is stamped with the methodology_version that produced it. When the engine changes, the version changes, and old scores remain tied to the method that generated them. This whitepaper documents engine_v3, the deterministic scoring engine at _research/ownnothing_scanner_prototype/engine_v3.js.

What a score is, in one sentence: it is this methodology's measurement of how far a specific contract departs from voluntary, informed consent and how much it strips from the person who accepts it. It is a methodology output, not a legal conclusion. It carries a stated confidence band (Section 5). The exact verbatim clause that drove the score is displayed next to the score, so you can read the evidence and judge it yourself.

Section 2: Author and ecosystem

The Consent Scanner is a property of OwnNothing.org, part of the Daylight Freedom ecosystem founded by Aaron Day, author of *The Final Countdown: Crypto, Gold, Silver, and the People's Last Stand Against Tyranny by CBDCs*.

The scanner names your capture, the Atlas cites the systemic proof, and the exit routes to Daylight (money-dimension exits route to DitchTheDollar). Day2026 is deliberately not linked from the scanner (campaign-finance firewall).

Section 3: Purpose and what is measured

Corporate contracts are written to be unreadable precisely because they are written to be unfair. Every "I agree" transfers rights most people never knew they held. The scanner makes that transfer visible.

Definition of what is measured. A contract's score measures its departure from voluntary consent on disclosed facts, across three layers:

1. Consent ceremony, does the instrument bind you by use alone, change itself without asking, and foreclose your ability to contest it? (This is scored first, before any clause counting.)
2. Custody and title, for instruments that hold your money or property, does title actually pass away from you, and can they freeze or seize it at will?
3. Clause content, across eight rights dimensions, how many redline, medium, and low-severity clauses appear, normalized for document length, offset by any genuinely user-favorable clauses.

Posture: neutral and consistent (the same engine scores every contract, with no per-company hand-tuning in the code), evidence-based (every point traces to a specific matched clause), and actionable (each verdict ends in a concrete exit act).

Section 4: Dimensions and weights

Layer A: Consent ceremony (dimension zero), 0 to 65 points

Scored per document as booleans, not per-1000-words. An instrument either binds you by use or it does not. Point values (linear, additive; capped at 65):

Offsets (subtracted from the layer, floored at 0):

Signature-ceremony suppression (Stoller fig-leaf fix, Round 3): a genuine signature block suppresses the "bound by use" penalty, but only if the instrument does not also amend itself unilaterally. A signature card on a self-amending instrument is ceremony, not consent, and does not earn the suppression.

Layer B: Custody and title, 0 to 40 points

Scored per document for instruments that hold funds or property (Round 3 addition):

Offsets (combined offset capped at 8): FDIC/NCUA coverage -4, notice-before-freeze or an appeal process -4.

Layer C: Clause content, eight rights dimensions

Each dimension is scored by counting matched clauses at three severities and crediting genuinely user-favorable clauses. Displayed labels are in plain English:

Per matched clause, per dimension: redline -30, medium -12, low -4 points of loss. Genuinely user-favorable clauses credit +2 each, capped at +4 per dimension. Matches are drawn from a redline lexicon organized into pattern classes (per the disclosure boundary, the patterns themselves are not printed here). At the level of counts: redline classes exist for three dimensions, medium classes for five dimensions, one low class, and positive-signal classes across five dimensions.

Length normalization. Clause counts are converted to a per-1000-word density: a count is multiplied by 1000 and divided by the document's word count, with a floor of 600 words (documents shorter than 600 words are treated as 600 so a short adhesive clickwrap is not artificially inflated).

Reading-burden penalty. Length itself is penalized above a 1000-word baseline at 4 points per additional 1000 words, with step additions at 5000 (+6), 10000 (+10), and 20000 (+15) words, capped at 20 points total. Rationale: an unreadable contract is itself a consent defect.

No compound bonus, no category multipliers (removed 2026-07-12). Two earlier mechanisms were removed by the Contract-Scoring board on review. A flat +12 clause-loss "compound bonus" for co-occurring arbitration and class-action language was cut: because that pairing is near-universal in adhesion contracts, the bonus behaved as a near-constant offset that inflated almost every score without distinguishing one contract from another, and it counted arbitration a third time on top of the two intentional counts (Section 9). Blanket category multipliers (healthcare, finance) were also cut: the stake weighting below already raises the weight of high-stakes relationships on a per-contract basis, so a fixed industry multiplier on top of it double-weighted the same concern. Neither mechanism is in the current engine.

Stake weighting (across layers)

Certain stakes raise the weight of the relevant scores:

• Monetary stake (title transfer or set-off present): the combined consent + custody total is multiplied by 1.5 (+50%).
• Biometric stake (biometric, precise location, faceprint, or fingerprint language): the two data dimensions (Your privacy, Your data monetized) are multiplied by 1.3 (+30%).
• Children stake (children/minors language, gated to exclude boilerplate "we do not knowingly collect" COPPA disclaimers): the same two data dimensions are multiplied by 1.4 (+40%).

Section 5: Composite score and bands

The published score is built from two components:

1. Consent + custody component = (consent deficit, 0 to 65) + (custody, 0 to 40), then multiplied by 1.5 if a monetary stake is present. This is the dominant component and is deliberately not rescaled: every point traces directly to a marker hit.

1. Clause component, rescaled to 0 to 35 by a square-root curve against a fixed anchor: 35 * sqrt(total_clause_loss) / sqrt(76), capped at 35. The anchor 76 is the maximum raw clause loss measured across the calibration corpus (the Redfin agreement, 2026-07-11). The square-root curve compresses very high raw clause loss so a single pathological contract does not dominate the scale.

Final score = (consent + custody component) + (clause component), capped at 100, rounded to one decimal.

Because the consent + custody component can exceed 65 after monetary weighting, the 100 cap does real work: the worst instruments (accounts where title passes and they can freeze at will, on a bound-by-use self-amending clickwrap) saturate at 100.

Confidence band (Kahneman rider, required next to every score). Each score is displayed with a confidence band that reflects two things: whether the underlying document is a full fetched text or a partial/thin fetch, and whether the displayed clauses have passed model-judge review and human editorial verification. A score on a full, judge-reviewed, human-verified document carries high confidence; a score on a thin or partially fetched document carries a stated lower confidence and is flagged as provisional.

Result bands. The scanner presents bands, not bare numbers, as the primary read (final band names and colors are a design-board deliverable; the ranges below are the working calibration):

Score name (LOCKED 2026-07-12): the Own Nothing Score. The earlier working names ("Rights Loss Index," "Rights Surrender Certificate") are retired: they asserted a legal conclusion about the target, which the Legal Analysis Board ruled out unanimously. The replacement was produced by three boards (X Viral, OwnNothing Strategic, Meme) generating a 20-candidate field, the Legal Analysis Board clearing each candidate against the legal-conclusion gate (two killed: "Rug Score" and "Coercion Score," both asserting a specific legal wrong as fact), and the standing joint panel ranking the cleared field. Own Nothing Score was the pick of all three generating boards and the panel, confirmed by the accountable owner. It is score language, asserts no legal conclusion, and its polarity is self-explaining: a score of 100 means the instrument leaves you owning and controlling nothing.

Section 6: Research and scoring process phases

1. Source acquisition. The actual product or service agreement is fetched (a deposit account agreement, cardmember agreement, brokerage customer agreement, connected-car terms, the uniform mortgage instrument, and so on), not a generic website terms-of-use page. Sources come from Open Terms Archive declarations, the ToS;DR API, and per-company search for the real instrument. Every URL is status-verified before fetch. A single browser driver fetches sequentially (a fan-out of fetchers against one shared browser is a known failure mode).

1. Deterministic floor scoring. engine_v3 scores the fetched text exactly as documented above. This is fully reproducible and runs offline.

1. Model-judge review. Every displayed clause must carry a model-judge ruling before it can be shown. Rulings are keyed by clause-text hash so they survive re-extraction when a document is refetched. Clauses without a ruling are held from display. (The judge layer is separate from engine_v3; see Section 8.)

1. Human editorial verification. Displayed clauses are editorially verified before a named score is treated as launch-grade (the corpus is currently marked unverified pending this pass).

1. Publication gate. No score publishes without its underlying evidence. A contract that produces zero matched clauses renders in an "evidence under review" strip rather than as a score.

Governance: a standing joint panel drives the build. The OwnNothing Strategic Board chairs; the Legal Analysis Board, X Viral Board, and a design board are pulled in at four fixed gates (data seed, scoring calibration, design draft, ship). Scoring changes route through this Contract-Scoring Advisory Board (three rounds on record).

Section 7: Data sources and verbatim gate

Sources. Deposit and cardmember agreements (Chase, US Bank, Truist, Amex, Discover), brokerage customer agreements (Fidelity, Robinhood, Schwab), rental-car jackets, connected-car terms (OnStar, HondaLink, FordPass, and peers), the Fannie/Freddie uniform mortgage instrument, tenant-screening bureau terms, and consumer-platform terms. Two fetch lanes: a real-browser lane for bot-walled sites, and a curl-plus-PDF-extract lane for documents that need no browser.

Verbatim gate (non-negotiable). Every clause the scanner displays must pass a verbatim substring check against the fetched source document. A clause that cannot be found verbatim in the source is not displayed. Junk fetches (navigation pages, login walls, 404s, thin stubs) are excluded by an explicit exclusion list. The verbatim clause is displayed adjacent to every score, which is what makes the score an opinion on disclosed facts rather than an unsupported assertion.

Section 8: Reproducibility and versioning (day-one launch gate)

• Deterministic reproducibility. engine_v3 is pure regex-and-arithmetic scoring with no randomness. The same input text plus methodology_version: engine_v3 yields the same score every time, and it runs offline (the same engine scores the corpus in Node via a batch scorer). Anyone can rerun a score.

• Score-drift regression suite. A scheduled regression suite pins every corpus score against its last-known value and flags any drift, so a change to the engine or the lexicon cannot silently move scores. This regression regime is the anti-gaming mechanism (per the disclosure boundary): the patterns stay private, but the outputs are continuously pinned and auditable.

• Datestamped document versions. Every scored document is archived with the date it was fetched, so any score ties to the exact text it was computed from. When a company changes its contract, the new version is scored separately and the score delta is visible rather than overwriting history.

• Public changelog. The version-control history of the scoring engine and of this whitepaper is rendered as the public changelog. Methodology changes are visible commits, not silent edits.

• Corpus separability (IPFS archival). The evidence corpus (the fetched source documents and their datestamps) is archived independently of the live site, on IPFS, so a legal action against OwnNothing cannot remove the underlying evidence from the public record. Corpus separability and the regression regime are both day-one launch gates, not fast-follows.

Honesty note on what engine_v3 is and is not. engine_v3.js is the deterministic floor only. The pinned model-judge layer described in Section 6 lives in a separate review pipeline (hash-keyed rulings applied on top of the engine's output), not inside engine_v3.js. The whitepaper describes both layers because the product ships both, but a reader auditing the engine file will find the regex floor and not the judge. This separation is intentional and is stated here so the doc does not overclaim what the engine file contains. Score-display consistency (added 2026-07-12): when the review layer rules a displayed clause a confirmed false extraction, that clause is removed from the score as well as from the display. The batch scorer strips the ruled-wrong verbatim text and re-runs the same deterministic engine, so the number can never rest on evidence that has been withdrawn. Because rulings are cached and hash-keyed, this stays fully reproducible: no model runs inside the scoring loop.

Section 9: Limitations, corrections and dispute protocol, feedback

Limitations

1. Score comparability across document scales. A curated clause-concentrate scores differently from a full 40000-word instrument, because the per-1000-word normalization dilutes long documents. Scores are most comparable within a document-length class. This is a known open calibration question flagged for the board before any cross-class ranking is published.
2. The floor is a pattern matcher. The deterministic engine matches patterns. Novel phrasing that a human would read as abusive but that no pattern class catches will be under-scored until the lexicon is extended. The model-judge and human-verification phases exist to catch this, which is why a raw floor score is not published without them.
3. Weights are a judgment. Different weightings would produce different rankings. The weights here are the Contract-Scoring Advisory Board's judgment across three recorded rounds, published so you can disagree with them specifically.
4. Fetch completeness varies. A thin or partially fetched document scores lower than the full instrument would. Confidence bands (Section 5) surface this, and provisional scores are flagged.

Intentional double-count riders (disclosed, not drift)

• Arbitration and class-action waiver count in exactly two places, on purpose. An instrument that forecloses your ability to contest it is worse on the consent axis (Layer A, "dispute foreclosed") and worse on the clause axis (the "Your right to fight back" dimension). The EFF seat marked this as deliberate: consent- channel foreclosure and bad dispute terms are two real harms, not one counted twice. This is the full extent of the double-count. An earlier third path (a +12 clause-loss "compound bonus" for co-occurring arbitration and class-action language) was removed on 2026-07-12: it was never ratified as a third harm, and because the pairing is near-universal it behaved as a near-constant that inflated scores without discriminating between contracts. The current engine counts arbitration on the two named axes and nowhere else.

Corrections and dispute protocol (Barrett Brown provision)

A named company may contest its score with evidence. The contest is logged transparently (the claim, the evidence, the date), reviewed against the verbatim source, and a resolution is published. If the contest surfaces a fetch error, a stale document version, or a misapplied pattern, the score is corrected and the correction appears in the public changelog. The dispute channel and the correction log are a launch requirement, not an afterthought (Legal Analysis Board Option B, Legal Hardening Package).

Feedback

This methodology is published as a draft for board and AI-counsel review. Feedback on dimension definitions, point values, weighting, the rescaling anchor, data sources, and the reproducibility regime is welcome at aaron@daylightfreedom.org.

*Published draft. Owning board: OwnNothing Contract-Scoring Advisory Board. Score name: the Own Nothing Score (locked 2026-07-12). Pending Legal Analysis Board defensibility ruling, run as AI counsel. Methodology version documented: engine_v3.*

Changelog

Every published change to this methodology, from version control. This is the audit trail, not prose.