How Trust Scores work

Why I built this

I started cashing out from these sites because nobody reviewing them had. The going rate for a "review" in this corner of the internet is an affiliate link and a thumbs-up. So the numbers here come from what actually happened when I deposited, ran KYC, and asked for my money back, not from an operator's press kit. Where I haven't tested something myself, the page says so instead of guessing.

The seven factors and their weights

How scoring works

• TRUSTED70–100
• CAUTION50–69
• AVOID0–49

Each factor gets a 0–100 score from documented evidence. Weight those by the table above and you have the base score, also 0–100.

Two adjustments can pull the base down: a freshness decay and a complaint penalty, both defined in the next section. Subtract them, clamp the result to 0–100, and that's the published score, the number stamped on the operator's page.

Tiers come straight off that number. 70 and above is trusted, 50–69 is caution, anything below 50 is avoid. Those map to the TRUSTED, CAUTION, and AVOID stamps on operator pages. No one nudges a score after the math runs: there's no manual override of the published number, which is also why an operator can't sit in the trusted tier at 64. Where editorial judgment does enter, it enters upstream as a recorded input the formula reads, and the adjustments section names the one input that is set by hand.

Methodology v2 evidence gate

The version in force is v2 preliminary evidence gate (2026-08-28); the rule set below describes the version 2 evidence gate.

Methodology v2 does not treat an unexplained number as evidence. A total can publish only when redemption reliability and Terms honesty are verified, plus at least two of the other five factors. That means at least four verified factors in all.

Once that gate passes, the base score is reweighted across verified factors only. A factor marked held, not applicable, stale, missing, or still in review is excluded. If the gate does not pass, the operator page shows that the total is withheld instead of substituting zero or carrying forward an unsupported number.

Version 1 totals remain attached to version 1 history. Moving an operator to version 2 requires a reviewed task after the research deadline, so installing the database change alone cannot rewrite a published score. The version stamp beside each score identifies which rule produced it.

Adjustments: freshness decay and complaint penalty

The base score measures the operator on the seven factors. Two things the factors don't capture on their own can pull the published score below the base, by formula. Every operator page that carries an adjustment shows the arithmetic next to its score breakdown: base, minus each adjustment, equals the published score.

Freshness decay

Evidence goes stale. If a score hasn't been re-verified for more than 180 days, it loses 1 point for every 30 days past that mark, to a maximum of −10. A score re-verified within the last six months carries no decay; one untouched for a year sits near the floor. Re-verifying an operator resets it to zero. The decay is recomputed for every operator on a schedule (below), and each change is written to the changelog.

Complaint penalty

When a documented legal event names an operator, and the event's allegations or consequences include tangible player harm (unpaid or frozen redemptions, confiscated balances, a botched wind-down after leaving a state), that operator takes a penalty scaled to severity:

• −5 a filed lawsuit
• −10 a regulatory action (state attorney general, cease-and-desist)
• −15 an exit-scam alert

A filing that only disputes whether the sweepstakes model is legal somewhere, such as an attorney general or municipal illegal-gambling suit, does not touch this penalty. That is a fact about the state, not about how the operator treats its players, and it is already carried where it belongs: on the legal tracker and in the state adjustments below. It also tends to name whichever sister brand the complaint happened to cite, which would penalize one brand of a group for conduct identical to its unnamed siblings.

Penalties from multiple qualifying events add up, capped at −25 in total. Each legal-event penalty eases by 1 point every 90 days, so a single lawsuit's hit is mostly gone after about 15 months unless something new is filed. The penalty is recomputed the moment a legal event is recorded and on the weekly schedule; the source event is linked from the operator's page and the legal tracker. To see which operators are carrying a penalty right now, and the review sentiment behind the category, read the complaint index.

The penalty has a second component, and it is editor-applied. Where a corroborated complaint cluster is documented, an editor can record a separate penalty against that operator in its own column, and the recompute adds it to the legal-event total before the cap. That component does not ease on a timer: it stays until a human clears it, and the change is written to the score history with its reason like every other. It is the one value in either formula a person sets, and it enters as a recorded input the formula reads, never as an edit to a published score. Mined review sentiment on its own never sets it.

Cadence

Both adjustments are recomputed automatically every Monday at 04:00 UTC, and the complaint penalty is also recomputed immediately whenever a new legal event is recorded. The formulas run on a schedule, and every resulting change to a published score lands in the changelog with its reason. The one hand-set value either formula reads is the editorial complaint penalty described above; it lives in its own column, so a recompute consumes it and never overwrites it.

Related: our operator-by-operator payout study shows how this plays out in practice: which operators have actually paid, how fast, and where the receipts came from. Alongside it, the sweepstakes complaint index tracks what reviewers report across the whole catalog and the complaint penalties currently applied.

State adjustments: the same operator, scored by where you live

The published score is a national number. But a sweepstakes operator that is sound in Texas can be a worse bet in a state that bans the redeemable-coin model, sets a higher minimum age, or has an open lawsuit against it. On every operator page you can pick a state above the score; we re-derive that operator's score for that state from the law and the operator's own terms. The factor breakdown lower on the page always reflects the national base score. The state layer adjusts that base; it doesn't re-run the seven factors.

Five conditions can change the picture. Each one shows its source on the page; nothing is applied without a citation you can check.

Not available in the state

If the operator's terms exclude the state (we record this per state in operator_states), there is no score. We say so and link to operators that do serve the state. Operator opt-out takes precedence over every other adjustment.

Recent state exit

If the operator pulled out of the state within the last 90 days, the score reads N/A for that state, dated to the exit. A balance you still hold there is a support question, not a rating.

Sweeps Coins not redeemable (cap at 50)

In a state where sweepstakes are prohibited, or where an enacted ban has reached its effective date (11 states at this build, listed on the legality tracker), the redeemable-currency model is broken: you can't cash out. We cap the state score at 50 (the top of the Caution band) rather than zeroing it, because free Gold-Coin play can still be legitimate. The cap cites the state's statute. A state whose ban is enacted but has not reached its effective date is not capped: Sweeps Coins still redeem there until that date, so we show the date instead.

Minimum-age conflict (−10)

If the operator's terms set a minimum age below the state's legal floor for sweepstakes play (for example terms that allow 18+ in a state that requires 21), a resident at that age is non-compliant even though the sign-up would go through. That is a real risk to the player, so the state score drops 10 points, citing the state age statute. See minimum age for how the floor is set.

Active state action against the operator (−15)

If there is an open, operator-named state action in that state (a lawsuit, cease-and-desist, or attorney-general action recorded in legal_events), the state score drops 15 points, citing the filing. This is on top of any national complaint penalty already in the base score; the state layer reflects that the heat is concentrated where the action was filed.

The arithmetic

The point adjustments stack, then the cap (if any) applies, then the result is clamped to 0–100:

state score = clamp_0–100( min( base − age − action , 50 if SC-banned ) )

If none of the five conditions apply, the state score equals the national score and we say so plainly.

Assumptions and privacy

Some state availability is inferred from an operator's published restricted territories rather than a first-hand sign-up. Those rows are labelled inferred. The page uses the state you choose and can remember it in your browser and the page URL. When you follow a /go/ link, our redirect service separately uses Cloudflare country and region data to enforce availability restrictions. A state selected on the page does not override that check.

Freshness: how current a score is

Every operator page carries a dated provenance strip under the score so you know what was checked and when. It names each event and then dates it, one row per kind: an editorial review is a conclusion a person reached, a data refresh is a scheduled job re-reading a public page, and a first-hand test is real money leaving the account and, sometimes, coming back. Those stay separate on purpose. A single “verified” label never said what was verified, and on an operator nobody had reviewed it implied a human review that had not happened.

Every date on that strip is a real column. There is no “recently”, no fallback to today's date, and no default reviewer name: an event we cannot date renders nothing at all rather than a vague claim.

A score doesn't stay fresh forever. Once it goes more than 180 days without re-verification, it starts to lose freshness: the badge turns amber, and the published score takes the freshness decay defined under adjustments. Re-verifying the operator clears the amber and resets the clock.

Score history: every change, charted

Every operator page carries a 90-day mini-chart of its published score. Each point on that chart is a logged change (a re-verification, an adjustment, a new factor reading), and each one records the reason it was made. The full permanent record, all the way back, is the public changelog; the mini-chart is just the recent window of it.

The chart is reconstructed from the same append-only history rows that feed the changelog, and it's anchored to the live published score. That means it can never disagree with the number stamped on the page. The line ends exactly where the current score sits, by construction.

What feeds the factors

Four kinds of evidence: the operator's own terms documents, first-hand redemption tests where I've performed them, state legal records, and player reports filed on review pages. Negative findings link to their source. When the evidence is thin, the score reflects that rather than guessing.

First-hand credit: how depth of testing earns points

When I test an operator in person (a real account, a real deposit, real cash-outs), that evidence feeds the score by formula, the same way every other signal does. Four kinds of first-hand observation each add points, each capped, and each idempotent (logging the same thing twice never counts twice). Every credit lands in the changelog with its reason.

• +5 each delivered redemption → Track Record, up to +20 (four cash-outs = full credit)
• +2 an in-depth session (a first purchase, or a daily-bonus week) → Track Record, up to +10
• +3 a session with 3+ tracked first-purchase offers, confirming the advertised promo was honored → ToS Honesty, once per operator
• +2 a logged daily-bonus week (5+ distinct days) → Reviews, once per operator per month, up to +10

Redemption credit and depth credit sit under separate caps, so first-hand evidence can lift Track Record by at most +30 combined. Only positive evidence is credited today: a failed KYC or an unrecovered cash-out is documented on the operator's page but does not subtract here yet. Nothing is credited that I can't show: the promo-honesty credit is withheld where an operator's terms couldn't be read, rather than assumed.

Daily-bonus value: observed rewards and their limits

Daily bonus records show observed or operator-reported coin rewards. Operator pages show nominal totals over time. The daily-bonus leaderboard orders offers by nominal annual Sweeps Coins, with Trust Score breaking ties. These quantities do not predict a redeemable balance or whether an operator will pay out.

The formula

We publish nominal observed coin rewards and their observation windows. Repeated totals assume the same award continues. Cash projections are withheld: turnover is a wagering obligation, not a divisor of redeemable cash. We will restore cash estimates only with a versioned, applicable model and explicit assumptions.

Wagering requirements and factor scores

A factor score describes the assessment under its applicable methodology. It does not establish a Sweeps Coin wagering multiplier. A specific requirement needs a reviewed source clause, currency, offer scope and observation date. Missing or conflicting terms remain unresolved.

Where the daily figure comes from

I prefer my own first-hand daily-bonus log (the coins actually credited over a real test week) to the operator's marketing claim. When I have several days, I average them and show the testing window; a streak-escalating bonus (more coins on day 7 than day 1) is drawn as a small bar chart. If all I have is the operator's own promotions page, I use that and label it. A daily bonus quoted only in Gold Coins, with no SC, has no cash value (GC is promotional play), so GC-only offers are excluded from the SC ranking. Gold Coins remain displayed separately on operator pages as entertainment coins.

Assumptions

Nominal totals assume the same reward repeats and every daily award is claimed. Observation dates remain unchanged when the site is rebuilt. These totals are coin quantities, not expected profit or a predicted redeemable balance.

Minimum age: state floor vs. operator terms

How old you have to be to play isn't a single number. Two rules stack, and the stricter one wins. Every state sets a statutory minimum age for sweepstakes play (the state floor); separately, each operator sets its own minimum in its terms of service (the operator floor). The age that actually binds you is the higher of the two:

effective minimum age = max(state floor, operator ToS age)

Most states' floor is 18; Alabama and Nebraska are 19. Many operators set their own floor at 18, but a number set 21 in their terms regardless of where you live, and a few set 19 in specific states. So a 21+ operator is 21+ everywhere it accepts players, while an 18+ operator is still 19+ for a resident of a 19+ state. We surface the operator's own ToS age on each review page (when we have sourced it) and the state floor on each state page, and where the two differ we flag the higher number as the one that governs.

One honesty rule, same as everywhere else on the site: if we could not source an operator's stated minimum age from its own terms, the review shows no age claim rather than defaulting to “18+”. An age badge on a review page means we read that number in the operator's terms and link them; its absence means we haven't confirmed it, not that it's 18.

Separately from the law, Bonus Bandit is written for readers 21 and older. That is why the site footer says 21+ even on pages where the operator's terms and the state floor are both lower. It is our own audience policy, not a legal minimum, and it does not change the two numbers above.

Where a score comes from: baseline vs. reviewed

Every score carries a provenance label so you can tell a rubric-computed baseline apart from one a person has individually verified. Both are computed from real evidence with the same formula. The difference is whether a named reviewer has signed off on it.

Baseline

The score was computed by the rubric on this page from documented public evidence (the operator's terms, state legal records, and player reports), but no individual has re-verified it by hand. Its stamp reads BASELINE and the page carries no named reviewer. A score sits here until I work that operator by hand, which is a queue position and not a judgment on the operator. A baseline score is a measurement from evidence, not a guess; it just hasn't had a person's individual sign-off yet.

Reviewed

An operator I've worked by hand: I read the actual terms, recorded the redemption evidence I could obtain (including first-hand cashouts where I ran them), and published a written verdict. These pages carry a “Reviewed by Noah Rafkin” line next to the score breakdown with the date the score was verified. That line has two conditions, and because they are two different facts we count them separately: a named reviewer set or confirmed the score, and a written verdict is published on the page. The line renders only where both are true; an operator missing either one carries no reviewer line at all. A named reviewer has signed off on 150 of 292 operator records at this build. Written verdicts are published on 278 operators.

The distinction matters for honesty: I never attach my name as reviewer to a baseline row. If a page doesn't say a person reviewed it, a person hasn't. The score is the rubric's, computed transparently from the formula above.

Versioning

The formula is versioned. Any change to a weight creates a new version id, and the change is logged in the public changelog. The version in force right now is v2 preliminary evidence gate (2026-08-28) (0cb3d58c). Old scores keep a reference to the version they were computed under, so a weight change can never silently rewrite history.

Independence

Some operators pay this site commissions (disclosure). Those relationships have zero input here: there is no affiliate factor, no affiliate weight, and no way for a commission to move a number. Operators can't pay for placement, score, or removal.

Watch: the research sequence

The complete beginner's guide walks the same four-step process this page documents: understand the model, compare the offer, read the terms, verify the evidence.

Nothing loads from YouTube until you press play.

Analytics and your privacy

We measure traffic with cookieless, aggregate analytics (pageviews, referrers, and which pages get read) via Cloudflare Web Analytics and Plausible. No cookies, no advertising profiles, no cross-site tracking, and no personal profile of you. Nothing that follows you to other sites.

We also count aggregate events that tell us whether the site is useful, for example that an outbound affiliate link was clicked, or that a player report was submitted, never tied to a name, an email address, or an account of yours. Some operational records, such as the one-way hash our /go/ redirect writes, are pseudonymous rather than anonymous. The privacy policy explains exactly what is stored and for how long. Both analytics tools are privacy-first and cookieless by design, which is why you don't see a cookie banner. Read their policies: Plausible's data policy and Cloudflare Web Analytics.

The newsletter

The weekly email is optional. If you sign up, we collect one thing: your email address, no name, no payment details, nothing else. Your address is stored in our own database, on the same infrastructure that holds the trust-score data, and it is never handed to a third-party list broker or ad platform.

What we send: new operator launches, state-law changes, and operator-side terms-of-service edits: the same vetted intel as the site, once a week. We never sell your address, and every email has a one-click unsubscribe in the footer. Unsubscribing removes you immediately and for good.

What's not in the score yet

Two honest gaps. First, bulk complaint-sentiment mining (aggregating Reddit and Trustpilot at scale) is not yet a scored input. (This is distinct from the complaint penalty above, which fires on documented player-impact legal events and on an editor's recorded entry after a corroborated cluster, not on aggregated chatter.) The player reviews factor is currently scored from the evidence I can document directly. Player reports are collected on every review page and feed the redemption stats as they verify. When sentiment aggregation becomes a scored input, the version id above will change.

Second, state-availability tracking is rolling out and not yet complete, so some operator pages don't have verified availability data for every state.