Methodology
RuleScore v0.9 — the five-factor rubric
| Component | Weight | What it detects | Evidence base |
|---|---|---|---|
| ● Source dependence | 24% | Counts and qualifies required confirmers; flags named-outlet rules, dual-government confirmation, press-consensus language, live-feed dependence; credits designated primary sources. | Sanders rally ($3.6M), Biden–Trudeau, Iran ceasefire extension ($203.6M), Melania mute case. |
| ● Definitional vagueness | 24% | Detects undefined operative terms (with a dispute-history term bank), magnitude adjectives without thresholds, 'widely considered' delegation, mention-market mechanics; credits explicit 'means/defined as' clauses. | Zelenskyy suit ($237M), TikTok ban ($120M), Venezuela invasion ($10.5M), Netflix-pronunciation. |
| ● Scalar carve-out | 18% | Checks numeric thresholds for boundary handling (at-or-above), rounding rules, and first-print-vs-revision policy on revision-prone sources. | 2025 Oscars viewership (first print 18M, final 19.7M), UFC 327 scorecard correction. |
| ● Disclosure timing | 17% | Identifies which clock a market runs on (event-time vs disclosure-time), timezone designation, hard cutoffs, early-expiration interaction. | Strategy-BTC 8-K timing ($60–85M), Khamenei halt sequence ($54M claimed). |
| ● Oracle manipulation | 17% | Scores the adjudication layer itself: UMA token-voting exposure, unilateral-discretion language, clarification rights; credits structured review processes. | Ukraine-minerals governance attack ($7M), WSJ oracle-concentration findings (May 2026). |
Notes & formulas
— Grades are produced by a deterministic rubric (RuleScore v0.9) over the live rules text retrieved from venue public APIs at grading time — every flagged phrase is quoted on the grade page, and every grade page prints the SHA-256 hash of the exact rules text graded, so any grade can be independently reproduced character-for-character.
— Scores are 0–100 where higher = greater dispute risk, banded Minimal / Low / Moderate / Elevated / High. Weights: source 24, definitions 24, scalar 18, timing 17, adjudication 17, plus a bounded category prior (e.g., mention markets +10, league championships −6) justified by the dispute record.
— RuleScore grades the risk that settlement is contested — never the probability of any outcome, and never the merits of a past resolution.
— v1.0 (in progress) adds LLM-assisted parsing with human review on every high-volume market, calibrated against the dispute database; the rubric stays published either way.
— Index methodology (EMU v1.1): constituents per basket = top 25 matched markets by 24h VOLUME (≥$500), weights = volume capped at $2M per market; EMU_B = 100 × Σ(wᵢ·4pᵢ(1−pᵢ))/Σwᵢ; composite = equal-weight basket mean. Polymarket data only, by data-rights design. v1.0 selected constituents by weight×uncertainty — selecting on the measured quantity, which simulation showed inflates the level roughly 9–11 points; v1.1 (Jun 11, 2026) corrects this. We found it, we fixed it, we disclosed it — that is what an index administrator does. Daily closes are computed by a scheduled job and committed to a versioned git registry — the commit is the timestamp (repository public from launch).
— Fee math: Kalshi taker = 0.07·C·P·(1−P) per order, rounded UP to the next cent; designated series carry maker fees of 0.0175·C·P·(1−P) (same rounding; most series have none) — source: fee schedule PDF + docs.kalshi.com Fee Rounding, checked Jun 11, 2026. Polymarket = C·rate(category)·p·(1−p), taker-only (docs fetched Jun 11, 2026). Arb economics are quoted per contract at a stated 100-lot; the 'actionable' badge additionally requires a LIVE order-book depth check (≥25 contracts executable at quoted prices).
— Data rights: Kalshi market data is licensed for personal, non-commercial use; commercial price-derived products launch Polymarket-first while a Kalshi license conversation proceeds. Facts and prices are not copyrightable (NBA v. Motorola), but API terms are contract law — we honor them. The same rule binds the Brierly API: it serves grades, subscores, hashes, short quoted excerpts, and Brierly-authored data — never full venue rules text or bulk raw market data.
— Dispute-count method (v1): Brierly counts Polymarket markets, via the venue's public Gamma API, whose UMA resolution history includes at least one dispute escalation and whose market end date falls in calendar 2026. The count runs daily and is self-validating — if the API filter cannot be verified to actually filter on a given run, no number is recorded and the site keeps citing The Defiant's published count instead. Dispute escalations are a lower bound on contested settlements (informal disputes that never reach the oracle aren't counted).
— Rules-change monitoring: a scheduled job hashes the rules text of the covered universe (top ~200 markets by 24h volume, both venues) every run, stores SHA-256 hashes with grades (never full text), and emails subscribers when a hash changes or a grade crosses a band. The change log is served at /api/v1/rules-changes and summarized on the RuleScore page.
— The track record is canonical in a versioned git registry: registered calls and daily index closes live in data/records.json, updated by a scheduled job, so the no-retro-editing claim is checkable in the commit history rather than asserted (repository public from launch; the link appears on the Track Record page once live).
— Licensing: site and engine code under PolyForm Noncommercial 1.0.0; the records registry (data/records.json) and the dispute database under CC BY 4.0 (cite Brierly); methodology text under CC BY-ND 4.0. Full terms in LICENSE and LICENSE-DATA.md in the repository.
Score bands
Independence policy
No venue pays for grades, previews grades, or can alter grades. Conflicts are disclosed on the page they touch. If exchange-affiliated capital ever invests in Brierly, ratings governance is firewalled in writing. The methodology changes only by published version bump with a changelog entry — including the ones that correct us.
Questions or grade disputes: trevorsardis@gmail.com — responses are published.