A channel's advertised percentage was computed by the channel. Building your own version takes about ten minutes, and most of the work is deciding what to do with the calls that never resolved.
Why do you have to build the record yourself?
Because no one else has. A Telegram channel is not audited, not registered, and not obliged to keep a scoreboard, so the number it advertises was produced by the same party the number is about.
That is not necessarily dishonest. It is simply unverified, and the difference matters when the alternative is a spreadsheet you fill in yourself from the channel's own posts, in less time than it takes to decide whether to subscribe.
The work below is a sample. It is not an audit and it does not pretend to be one. Thirty calls will not settle whether a channel is good. What they settle is whether it publishes checkable calls, whether those calls resolve, and how closely the record it advertises resembles the record it produced.
What should you sample?
The most recent complete calls, in order, taken from the feed rather than from the results post. Aim for 30, and if the channel has not produced 30 complete calls, that finding is worth more than any percentage you could compute from what it did produce.
A complete call names a pair, a direction, an entry price, and something that ends the trade. Anything else goes in the sample as an incomplete call and stays there, because dropping it is the exact move that flatters a record.
Two sampling rules keep the exercise honest.
- Take a contiguous run. Start at the most recent complete call and work backwards without skipping. The moment you choose which calls to include, you are doing what a channel does when it builds a showcase.
- Ignore the pinned results. They are the channel's own selection. You are building the thing they were selected from.
What do you write down for each call?
Six columns, and you can do this in a notes app. The columns exist because each one is a place where a record can be quietly improved.
| Column | What goes in it | Why it is there |
|---|---|---|
| Date and time | When the call was posted | Fixes the moment everything else is measured from |
| Pair and direction | BTC long, and so on | Lets you find the chart |
| Entry | The published price | The field that decides whether the trade existed |
| Stop | The published stop, or "none" | Half of them will say none |
| Targets | The full ladder, in order | A six-target ladder is six different possible outcomes |
| Outcome | What you observe on the chart | The only column you fill in yourself |
The sixth column is where the exercise lives. Everything before it is transcription.
How do you decide what counts as resolved?
This single decision moves a percentage more than anything else you will do, so make it before you look at the outcomes rather than after.
A call resolves when the entry filled and then either the stop was hit or a target was reached. Those are the calls you can score. Everything else falls into three buckets that are not losses and are not wins:
- Never filled. Price never came back to the entry, so no trade existed.
- Overtaken. Price reached the target zone before the entry filled, so the move happened without the trade being available.
- Still open. Neither the stop nor a target has been touched, and the call is old enough that it is not going to be.
Our own replay hits this wall at scale. Of 3,167 resolved outcomes as of 7 August 2026, only 1,583 ever produce a verdict. The other half are unfillable, unresolvable, or still waiting. If your hand-built sample comes out the same way, that is the normal shape of this data rather than a mistake in your counting.
What you must not do is delete those rows. A percentage computed only on the calls that resolved, presented without the count of calls that did not, is the single most common way a track record is inflated, and it is done more often by accident than by design.
What do you do with a call that has no stop?
Record it as unresolvable and keep counting. It sits outside both columns, and it will stay there until somebody decides by hand when the trade ended.
You will have plenty of these. Of the 3,227 signals we have parsed as of 7 August 2026, 1,647 - 51% - carry no stop at all. A call with no stop can be held indefinitely, which means only the author can say when it finished and at what price, which means the outcome is an opinion rather than an observation. Why that happens so often is the subject of Why half of signals come without a stop loss.
For your sample, apply one fixed rule to all of them and write the rule down. A 30-day limit is defensible. A 10% adverse move is defensible. Deciding case by case is not, because by the third case you will be doing what the channel does.
What about calls overtaken before the entry filled?
Count them separately and prominently, because they are usually the largest group and they are invisible in every advertised percentage.
In our corpus, calls overtaken by their own thesis are the single most common outcome: 1,554 of 3,167, or 49.1% as of 7 August 2026. That headline needs its caveat every time it is quoted - one high-volume channel supplies 1,197 of them, and across the other sixteen publishers the rate is 21.9%. The full breakdown is in Half the trades never happen.
When you find one in your own sample, the tell is simple. The chart shows the move the channel described, and it shows that price never traded at the published entry after the message went out. The call was right and the trade was not available, and those two things get reported as one thing.
What can 30 calls actually tell you?
Less than you want and more than nothing. A sample this size answers structural questions well and percentage questions badly.
Structural questions it answers: does this channel publish complete calls, do the calls resolve, does it publish losses, is the entry usually reachable, and does the advertised record resemble the observed one. All five are yes-or-no and all five are visible in 30 rows.
Percentage questions it does not answer. If your sample shows 70% across 20 scored calls, the range of true rates consistent with that result runs from roughly 48% to 86%, which spans "worse than a coin flip" and "genuinely good" at once. That band is a confidence interval, and the construction used there is the Wilson interval, which NIST recommends because the simplest textbook formula misbehaves on small samples (NIST/SEMATECH e-Handbook, section 7.2.4.1).
Regulators reach the same conclusion from the other direction. Australia's ASIC treats performance drawn from "an inappropriately short time period" as potentially misleading, and says that for an operation running under a year such a period "would usually be inappropriate and may be misleading" (ASIC Regulatory Guide 234, June 2026).
So use the sample to decide whether the channel is checkable at all. What a percentage needs behind it before it settles anything is worked through in Win rate: why 90% accuracy can still lose you money.
Where does the count usually break down?
In four places, and each one has a note that keeps the sample usable rather than abandoned.
Edited posts. A message can be edited after publication. If a call looks unusually precise about a level the market later hit exactly, note it and move on; you cannot audit an edit history from outside, and one suspicious row does not invalidate 29 others.
Deleted posts. Gaps in the numbering of a channel's own call sequence, or a results post referencing a call you cannot find, are worth recording as a row of their own labelled "missing".
Target ladders. A call with six targets is a call with six possible stories. Fix the rule before you start: score the first target, or score the last one reached before the stop, but use the same rule for all 30. Our own replay runs both conventions separately and never mixes them, which is set out on the Methodology page.
Partial and scaled entries. "Enter in the 0.412 to 0.418 zone" is three calls wearing one hat. Take the worst price in the zone and be consistent about it.
The ten-minute version
If you only have ten minutes, this is the order that gets the most out of them.
- Scroll to the most recent complete call and start a list. Date, pair, direction, entry, stop, targets.
- Work backwards to 30 without skipping. If you run out of complete calls before you run out of feed, stop and write down how many you found.
- Open a chart for the three oldest calls in your list. Did price trade at the entry after the post? Did it then reach the stop or a target?
- Mark every row as scored, never filled, overtaken or unresolvable. Four buckets, no exceptions, no deletions.
- Count the buckets before computing anything. The ratio of scored to everything else is the first real finding.
- Compare against what the channel advertises. Not the percentage, the denominator: how many calls does it claim its record covers, and how many did you find?
- Look for a published loss. One clearly labelled losing call in 30 tells you more about the operation than any winning streak.
A channel that survives steps 5 to 7 is one whose numbers can at least be argued about. Where that fits in a full pre-purchase sequence is set out in How to verify a crypto signal channel, and how to test an individual results image is in How to check a profit screenshot.
What this does not prove
A 30-call sample does not establish that a channel is profitable, unprofitable, honest or dishonest. What it establishes is whether the channel produces evidence at all, which is a prior question and the only one you can settle before paying.
Your outcomes column will also disagree with a careful replay in places. You are reading candles by eye, using one target convention, without fees or funding, and without accounting for whether a limit order would actually have filled at the price you saw. Those gaps all run in the generous direction, which is worth knowing when your hand-built number comes out better than a published one. The limits of replay as evidence, including ours, are in What a backtest can and cannot prove.
Our own figures describe our index as of 7 August 2026: 3,227 parsed signals from 17 publishing accounts, 3,167 resolved outcomes, 1,583 of them scored. That index is 50 channels and is not a random sample of the market, the cancelled rate is dominated by one publisher, and every figure moves as the replay catches up.
Nothing here alleges misconduct by any channel, and a channel whose calls do not resolve is not thereby doing anything wrong.
The practical read
Three numbers come out of ten minutes of counting, and together they say more than any percentage on a sales page.
How many complete calls did this channel publish in the sample you took? How many of them ever resolved into a scoreable outcome? And how does the denominator the channel advertises compare to the one you counted?
Current ratings for the channels we can score are at Signal Providers, and how many channels in our index still publish at all is counted in Only 14 of the 47 signal channels we index still post. Nothing here is financial advice.