How this works
1. What we collect
We pull public videos from a fixed list of channels, take the transcript (YouTube captions where available, otherwise local speech recognition), and keep the timestamp of every sentence.
2. What counts as a call
A call must name a specific asset, a direction, and a horizon. "Bitcoin to 120k by June" is a call. "The market looks interesting" is not. Conditional technical setups count when the speaker states which side they lean on and holds a position or a target. A statement with equal weight on both outcomes is not a call, and neither is a prediction someone else made that the speaker only repeats.
A deadline is the speaker's own when they said it: a date, "this week", "by the end of the year", "this cycle" (counted as a year). When they named none, we have to assume one, usually 30 days for a call on direction alone and 90 days for a call with a price target. Such a call is still judged and shown, but its card says deadline set by Oracle Check, and it never counts toward the grade.
3. How a call is judged
| Type | Hit | Stopped out | Miss |
|---|---|---|---|
| Up with a target | High reaches the target before the deadline | Low hits the speaker's own invalidation level first | Deadline passes, target not reached |
| Down with a target | Low reaches the target | High hits the invalidation level first | Deadline passes, target not reached |
| Direction only | Price moved at least 3% the right way by the deadline | — | Anything else |
| Sideways | Price stayed inside a 5% band | — | Band broken |
4. Repeated opinions count once
A channel that publishes daily will restate the same view in every video. Counting each mention separately would measure how often someone posts, not how often they are right. So when the same person repeats the same direction on the same asset while the earlier call is still open, and with a target within ten percent, we link it to the original and leave it out of the score. The card shows how many times a call was restated, because that number is interesting on its own.
5. The grade
A channel's grade looks at the last 12 months. We take every call where the speaker named the deadline themselves and whose verdict came in the last 12 months — hit, miss or stopped out — and count the hits. Open calls do not count, and neither do calls whose deadline we had to assume: a verdict on our assumption would say as much about us as about them. One sentence that names several coins counts once. About half of these calls name a price target; the rest are judged on direction alone, as in the table above.
That hit rate is compared with the typical hit rate: the same count for every channel we track, taken together, over the same 12 months. Right now it is 21%. It moves with the market, so a channel is measured against what everyone else managed in the same stretch, not against a fixed number.
A channel is graded once it has at least 10 such calls resolved in the last 12 months. Below that a percentage is noise, so the page says "not graded recently" and how many it has; its full record is still shown.
Even ten calls can land above or below the typical rate by luck. So for each channel we work out the range its hit rate could plausibly have, given how many calls stand behind it (a 95% Wilson interval: the fewer the calls, the wider the range). Five hits out of ten is 50%, but on ten calls the range runs from about 24% to 76%. Then:
| Grade | What it means | Rule |
|---|---|---|
| A | Confidently above the typical hit rate over the last 12 months | even the low end of the range is above typical |
| B | Above typical, not yet conclusive | at or above typical; the range includes typical |
| C | Below typical, not conclusive | below typical; the range includes typical |
| D | Confidently below typical | even the high end of the range is below typical |
A channel page also shows the all-time record on deadlines the speaker set, all resolved (every resolved call, including ones whose deadline we assumed) and, in the full table, direct calls, which drops conditional setups of the form "if we close above X, then Y" — those depend on a trigger that may never fire. The gap between the numbers says something about how a channel talks.
6. Which period a grade covers
The grade covers calls whose verdict came in the last 12 months, so it reflects how a channel has done lately. The window moves every day: old verdicts drop out and new ones come in, and a letter can change without the channel saying anything new. A channel that stops making calls we can check loses its letter within a year; its record stays. Every channel page says when the verdicts behind its grade came in, and shows the all-time record next to it.
We do not have every video a channel has ever published. We start from its recent uploads and work back through its archive over time, so the all-time record rests on different stretches of history: for some channels a few weeks of 2026, for others years of calls. Every channel page says when those calls were made.
A record also carries the market of its period. In a rising market bullish calls hit more often, in a falling one bearish calls do. Comparing each channel with the typical hit rate of the same 12 months takes some of that out, but not all of it: a channel that always calls the same direction still rides the trend. Compare all-time records from different periods with care.
7. Known limits
Where a channel's captions are missing we use speech recognition, and it mishears coin names. We correct the spelling of tickers against exchange listings and drop anything we cannot identify, which means a small number of real calls are lost rather than logged wrong. Live streams longer than two hours and members-only videos are not transcribed yet, so for some channels part of what they said is missing. Calls about stocks and indices are not tracked at all — this is a crypto scoreboard.
8. What we do not do
A grade describes a channel's public calls, not the person behind it. We do not call anyone a fraud. We do not publish a call without the quote, the date and the price. Where speech recognition mangles a coin name, we correct the spelling but never the meaning. Nothing on this site is financial advice.
9. Corrections
If a call is mis-transcribed, mis-dated, or judged against the wrong level, we will fix it and say so. Every entry links back to the source video at the exact second. What we fix, what we remove and how fast — corrections and removal requests.
30-day direction
The headline grade needs a deadline the speaker set themselves, and most channels rarely give one. So we also ask every directional call a simpler question: 30 days after the words, had the price moved the way they said? We take the daily close 30 days after the call (the nearest candle within three days if that day is missing) and compare it with the price at the moment of the call. Conditional setups (“if we break X, then Y”) and restatements are excluded. This is a looser test — a coin flip lands near 50%, and a trending market flatters whoever repeats the trend — so it is shown separately, never mixed into the grade, and only for channels with 10 or more such calls.
Questions
Why hit rate and not profit?
Profit depends on things a video almost never states: the entry, the position size, the stop and the exit. Two viewers can follow the same call and end with opposite results. What a speaker does state is a coin, a direction, often a target and sometimes a deadline, so that is what we check: did the price get there in time. A hit rate says how often a channel's stated calls came true. It does not say whether trading them would have made money, and we never claim it does.
What about zones and probabilities without a deadline?
A zone or a probability is not a call by itself. A statement that weighs both outcomes the same ("could be 80k, could be 120k") is not logged at all. When the speaker leans one way and names a target but no date, we log it with an assumed deadline, the card says "deadline set by Oracle Check", and it never counts toward the grade. So a zone without a deadline is never held against a channel.
Why the speaker's own deadline, and not 30 days?
Because a call should be judged on its own terms. "Bitcoin to 150k by the end of the year" is not wrong in February, and a fixed 30-day window would call it a miss. So the grade uses only calls where the speaker named the deadline, each checked on that date. The 30-day direction check on this page is a separate, looser number for channels that rarely give dates, and it never goes into the grade.