Methodology changelog

What changed on the site and when. Newest first.

Until today, operator review pages implied first-hand withdrawal testing where none had been done. Five seed PayoutTest rows existed in the database with is_sample=True, and the editorial rating math was including them as if they were real measurements. They were not.

What changed:

• Sample tests no longer contribute to any visible rating, JSON-LD Review, or verdict.
• Every operator review page now carries an audit-only banner unless a first-party test (is_sample=False) exists on record.
• Verdict labels changed from "Verified" to "On-chain activity observed". We have not personally tested withdrawals; we have observed wallets paying out.
• Tier labels changed: "Verified" became "On-chain verified", "Cautioned" became "Caution advised". Watchlist unchanged.
• Review.reviewRating is now emitted only when at least one first-party verified test exists for the operator. Zero operators meet this bar today.
• Sample test pages emit no Review JSON-LD at all. The page's sample-data banner already discloses sample status to humans, and the structured data now matches.

Rationale: a site whose domain contains "verified" cannot afford to imply tests it did not run. Strategic review with Codex on 2026-05-28 concluded that flipping sample tests to look real would amount to misleading structured data in a YMYL niche, with a 70–90% one-shot brand-kill probability if discovered. The honest path is to lead with what we actually have, 119 confirmed on-chain payouts across 7 attributed wallets, and label the gap.

Added weekly automated re-checks of every UK Gambling Commission and Malta Gaming Authority license against the regulator's public register. Status changes (Granted to Revoked, Suspended, and so on) are flagged on the operator page with a freshness badge.

Added a candidate-intake pipeline so new operators move from discovery to public review page in roughly 5 to 7 minutes instead of 40. Every step has an audit trail: scraped facts retain their source URL and DOM snippet hash, LLM extractions go to a review queue before any canonical row is written, and a reviewer must accept each fact individually.

No editorial change for readers. These are infrastructure.

We now refuse to activate a payout wallet for tracking without a person clicking through to Arkham Intelligence's labelled entity page and confirming the operator name matches. Two candidate wallets we previously considered for FortuneJack and BetFury did not survive this review and remain inactive. Their absence from the leaderboard is intentional, not an oversight.

First version of the on-chain forensics layer. 119 real USDT payouts from 7 Arkham-attributed wallets fetched from public block explorers, surfaced as the /verified-payout-leaderboard/ cross-operator data product.