Methodology

What a grade means

How far you can trust what this person says about markets, on everything we have been able to check.

Every person with a live public account is graded. The letter starts from a fixed point, BBB, which means we checked and found no red flags. It moves down for what we find, and it moves up only for a track record that beats chance.

Where the rating stands today

People tracked

1,184

Carry a grade

1,125

Published NR

59

Read this before reading a grade. 1,125 of 1,184 people carry a grade, and 59 are published NR, each with the reason. Of the conduct checks, two count today: referral links and removals. Selling paid access and soliciting paid promotion are shown on profiles but not counted until they pass a precision audit, and regulator lists have not yet been checked. The counts are read from the database on every rebuild.

The bands

Eight letters, plus NR. Each band is one step. Bands only: no point score is published, because a decimal invites false precision. Every move away from BBB is listed by name on the profile instead.

Within a band, people are listed in an order set by an internal key that is never shown. The order is not a finer grade.

Who gets a grade: the evidence checklist

A person is graded when all of these hold. Every live public account can meet them.

1. Their profile or channel description has been read, including the link in their bio.
2. At least 10 of their own recent posts or video descriptions have been read, or every item they have ever published, however few. A small account read in full is complete evidence, not a partial sample.
3. They post publicly. A private account is NR.

Two labels sit beside the letter when they apply.

Limited material

Fewer than 30 of their own items have been read. A rule demanding a fixed number would exclude small channels for being small, which grades size, not conduct.

Provisional

Their removals have not yet been watched for 30 days.

The checks, and what moves a grade down

Each check is defined the same way on every platform. A check that was measured and found nothing never lowers a grade. A check not yet measured is not counted, and the profile lists it as not yet checked. Moves are in steps, and each step is one band. Steps add, and the total is rounded down to whole bands, so a single half step changes nothing on its own.

Whether a check counts today depends on its precision audit, further down this page.

Never a finding: disclosing a sponsorship, or ticking the platform's paid-promotion box. Disclosure is what we ask for. Penalising it would reward concealment, and the first build of this method did: 60 channels would have scored higher by not disclosing.

Caps

Caps are applied after the steps and named on the profile. A cap sets the highest grade a person can hold.

Regulator warning lists and enforcement records have not yet been checked, so no grade is capped on those grounds today.

Before any C or D is published. A C, a D or a severity cap is published only with the evidence attached, with a right of reply, and inside the published response times of the corrections process.

The precision audit: which checks count today

A check can lower or cap a grade only once its detector has been audited at 95% precision or better, on a sample of at least 100 cases, each judged against a written rubric by an independent AI review, published with the result. Until then its findings are shown on the profile and not counted. A detector changed after an audit must pass a fresh sample, because tuning a rule on the cases used to measure it flatters the measurement. Results as of 7 October 2026:

Track record: the only way up

Calls are the only thing that can lift a grade above BBB. Each public call is checked against market prices, and the person's hit rate is compared with what chance would have scored on the same kinds of call. A record that beats chance by more than luck explains moves the grade up. A record reliably worse than chance moves it down. A record in line with chance leaves the grade where conduct put it: making calls never costs an honest analyst with no edge.

Someone whose findings add up to a full step can climb back to BBB through their record, but no higher.

What counts as a call

• A target call names an asset, a direction and a price it says will be reached. It comes good if the market reaches that level inside its window. Where the author gave no deadline, the window is sized to the asset's own recent volatility.
• A trade setup names a level that starts it and a level that would prove it wrong. Once it starts, it comes good if price moves as far in its favour as the author put at risk before touching the level that would prove it wrong.
• A setup is the easier claim: chance reaches about half of them, against about a third of target calls. So a setup counts as 0.6 of a target call, and a record is tested once it holds ten target calls, or seventeen setups, or a mix worth the same. A call restated while the same call is still open, such as a daily post repeating “above $59 offers a long”, counts once; the repeats stay visible in the call list, marked as such.

The test, in detail

Chance baseline.

What someone would score posting the same kinds of calls in the same weeks, going long as often as this person does, but choosing long or short at random. For each call it mixes how other callers' long and short calls of the same type fared within seven days either side. A permanent bull in a rally is measured against that rally's long calls; someone who switches side at the right time is credited for it. Nobody is measured against their own calls.

Edge.

Their hit rate minus that baseline, with each call weighted for recency. How likely a record this good, or this bad, is to arise by luck is computed exactly (the count of hits is a Poisson-binomial, because every call has its own chance), not from a bell-curve approximation, which made short lucky records look stronger than they are.

Calls open at the same time.

Calls open at the same time on the same side win or lose together: a long on BTC and a long on ETH in the same week share part of their luck. The test counts such calls for less, so a record of many overlapping calls carries the evidence of fewer independent ones.

Moving up.

When luck alone would produce a record this good less than once in a thousand (one-sided p ≤ 0.001), the grade moves up one step: from BBB, to A. The bar is 0.05 divided by the number of people tested, set once a quarter (50 in October 2026), which keeps small the chance that anyone on the roster is lifted by luck. Only the person's own record enters: other people's results never move it.

Staying up.

Once moved, the step is kept until the record no longer clears one in twenty (p > 0.05), so a grade does not flip on a single resolution.

Further steps.

A second step needs an edge of at least 10 percentage points once discounted for record length (edge × n ÷ (n + 20)), and a third needs at least 20. Both need at least 50 effective calls, meaning calls counted at their weights. From BBB, that is AA and AAA. A step already held is kept until the discounted edge falls 2 points below its bar.

Moving down.

The same test and the same rule in the other direction. A record worse than chance at one in a thousand loses one step, and a second and third at discounted shortfalls of 10 and 20 points. It is kept down until the shortfall no longer clears one in twenty.

Neither.

The grade stays where conduct put it.

What a record can show.

Each profile states the smallest edge its record could detect. On today's records that is at least about 23 percentage points, and over 50 for a typical record, so “in line with chance” usually means there is not yet enough evidence either way, not that the person has no skill.

The bar is high on purpose. As of 14 October 2026 nobody tested has shown a reliable edge under this test, so nobody is above BBB on the strength of their calls: the top bands are reached only by long records with large edges. The bar and the allowance for calls open at the same time are re-set once a quarter, and each re-setting is recorded in the decision log.

Specificity is described, not scored

Whether someone names levels and deadlines is shown on the profile as a description of how they talk. It does not move the grade. Scoring it rewarded attaching numbers whether or not they came true, and it pushed people who make calls below people who make none.

NR: when there is nothing to check

NR is used only when there is nothing public to check. The profile prints the reason, which is one of these.

NR is not a low grade. It is the absence of one. It means there is nothing public we can check yet, and we would rather say so than guess.

Accounts that change

Deleted by the owner.

The last grade is frozen, with the date and the words “account deleted”. If open calls or a recent finding existed, the deletion is recorded as a removal.

Suspended by the platform.

Shown as a status, with the date. It counts as a finding only when the platform has said why.

Made private.

The grade is withdrawn after 30 days. The audit trail is kept for disputes.

Renamed, or a new account.

Linked to the same person when the evidence is clear.

Funds and platforms

Funds, asset managers and verification platforms are not people making calls, so the rating of people described above does not apply to them and is not implied. Each will carry an editorial assessment: a written review against the published checklist below, labelled as such on its page, with no engine behind it. Until a review is on file, the entry is published NR. The letter uses the same scale so a reader does not have to learn two.

Collapsed funds and schemes stay listed with their status shown. Nothing is removed.

Ratings are unsolicited

Nobody rated here has paid for, requested, or approved their rating, and no rating is for sale. Ratings are editorial opinion formed from public evidence. They are not financial advice and not a recommendation to buy, sell or follow anyone.

See all analysts