Methodology: How the Sweepstakes Law Tracker Is Verified

How every claim on this site is sourced, labelled, and corrected. Published because a compliance tracker that cannot explain its own verification process is asking for a trust it has not earned.

Source hierarchy

  1. Primary. Bill text, chapter law, statutory code, filed pleadings, a regulator’s own announcement, a court’s own opinion. Where a primary source exists, nothing else is cited.
  2. Secondary. Law-firm analysis and specialist trade press, used to locate primary sources and to establish dates. Never used as the sole authority for a legal proposition.
  3. Operator conduct. Market exits, restricted-state lists, terms-of-service changes. Evidence of what operators believe the law requires, which is frequently more instructive than what they say it requires.

Confidence labels

Verified Confirmed against a primary source. Stated as a bare assertion.
Probable Reported consistently by credible secondary sources; not yet confirmed against a primary one. Hedged in the prose, and the reader is told to check before relying on it.
Unverified A single source, or an inference. Framed explicitly as interpretation, never as fact.

The label and the verb move together. A Probable claim never appears in a Verified sentence.

Correction discipline

Errors are corrected in a separate published note, dated, linked from the original page. The original page is not silently edited, because silent editing is how a tracker becomes unfalsifiable.

The Nevada page on this site is an example. It was initially drafted describing SB 256 as a sweepstakes ban. It is not one — SB 256 never names sweepstakes, and Nevada’s prohibition rests on NRS 463, which predates the model by decades. The page was rewritten before publication and the error is recorded here rather than erased.

Update cadence

Tracker pages carry a last-verified date. Any change in status — a signature, an effective date, an enforcement action, a filed suit — is reflected in the same week. Pages that go stale are worse than absent, because a stale page still looks authoritative.

What is not done

  • No affiliate links. No operator on this site is a commercial partner, and none pays for placement or coverage.
  • No player recruitment. This site is not written for players, and does not chase player search intent.
  • No prediction dressed as reporting. Where a statute’s construction is untested, it is said to be untested.

Last updated: 13 July 2026. Compliance reference, not legal advice.

Who builds this site, and what that means for the numbers on it

This platform is built by one person working with an AI. That is disclosed here rather than left to be inferred, because every compliance argument on this site rests on published figures, and a reader is entitled to know how those figures were produced before deciding whether to trust them.

The division of labour is not decorative. It determines which claims on this site carry a human’s professional judgment and which are mechanical output, and those two categories are treated differently throughout.

What the AI does What the AI never does
Writes the games, the mathematics, the operator console and the site code. Drafts editorial. Runs research passes across primary sources. Executes SEO work. Prepares the weekly statute verification. Decides what is legally true. Assigns the Verified, Probable and Unverified labels that appear on tracker entries and teardowns. Signs off a compliance claim. Publishes anything unreviewed.
Produces the numbers — reel weights, paytables, multiplier arrays, return calculations, simulation harnesses. Chooses the method by which its own output is checked. That rule was set before the code existed and the AI is bound by it, not the author of it.

The verification layer is human and it is not a formality. Statute research is checked against primary bill text before a status changes; a confidence label is a professional assertion by someone whose day job for two decades has been gambling risk and financial-crime compliance. An AI cannot make that assertion, and this site does not pretend otherwise.

The standing rule on every published number

No figure describing a game on this platform is written by a human. Every published return is computed at page load, in the reader’s own browser, from the same constants the running game uses to play. If a figure cannot be produced that way, it is labelled as measured rather than exact, and the measurement runs on demand.

The consequence is that a label cannot drift away from the thing it describes. It is not a claim recorded once and maintained by discipline; it is a measurement of the code as it currently stands, taken again on every visit. Eight of the twelve games publish returns that are exact in the arithmetic sense, because their outcome spaces are small enough to walk in full — 3,375 combinations for a three-reel slot, 512 cards for a scratch game, thirteen binomial slots for Plinko. The remaining four are sampled at load, and say so.

This rule is the reason an AI-built product is publishable at all. Production is close to free now, and an operator generating games or reel mathematics with a model has the same problem this platform had: the output looks correct, and looking correct is not evidence. The answer adopted here is mechanical rather than procedural — remove the human-typed number, and the failure mode it enables goes with it.

The error ledger

Mistakes made building this site are logged with what caught them and what changed. The ledger exists because a verification method that has never caught anything is untested, and the entries are the evidence that this one is not.

Two of them carry the argument. In July, Faro Dorado published a return of 94.7% while a five-million-spin audit measured 92.38% — the paytable did not return what the label promised. It was re-solved and the published figure moved to 94.6%. That error was a human’s. In August, Plinko’s multiplier array was assembled backwards, placing the 40× jackpot in the slot the ball reaches 924 times in 4,096; the page computed the resulting return and printed 1410.82% at first render, before deployment. That error was the AI’s.

Neither was found by review. Both were found by arithmetic that ran in public, and the ledger does not grade them differently by author.

Why any of this is disclosed

Operators are already building with AI. Very few have a method for verifying what comes out, and “who audited the AI’s work” is becoming a question that arrives from procurement rather than from journalists. Publishing the method here is not modesty about the tooling — it is the demonstration, and the games are the working copy of it.

The full ledger, the state-law verification cadence and the framework the console implements are documented across this site rather than summarised in a brochure. An AI wrote most of the code. A human decided what could be claimed about it, and that boundary is the product.