Leverage is the most reliably published field in our corpus: 92.78% of signals carry one, while half carry no stop at all. It is also the field that tells a follower the least, because most calls come from channels that post the same multiple every time.

How often does a signal even state a leverage?

Almost always. 3,700 of the 3,988 signals in our index carry a leverage figure, which is 92.78% of them. Only 288 arrive without one, and 25 of those mention leverage somewhere in the message in a format our parser failed to read.

That makes leverage the best-populated field in the corpus. It also sits in odd company: 2,022 of the same 3,988 signals publish no stop of any kind. The number that decides where the exchange closes you is present nine times in ten, and the number that decides where the trader closes you is missing slightly more than half the time.

A missing leverage is usually a property of the publisher instead of an occasional omission. Of the 288 signals without one, 159 come from a single channel that never posts a leverage line at all, 67 from a second and 21 from a third. Those three publishers account for 247 of the 288.

Everything below describes our index as it stood on 17 August 2026, and the sampling limits are set out near the end.

What number do they actually post?

Ten. The median is 10x, the mode is 10x, and 2,109 signals carry exactly that figure, or 57.00% of everything that states one. The mean is 16.2x, the quartiles sit at 10x and 20x, and the full range runs from 2x to 200x.

Leverage stated Signals Share of the 3,700
2x to 5x 112 3.03%
6x to 10x 2,118 57.24%
11x to 20x 609 16.46%
21x to 50x 795 21.49%
Above 50x 66 1.78%

The bulk sits in one bucket and the tail is thinner than the reputation of the category suggests. 26.27% of signals with a figure ask for 20x or more. 119 ask for 50x or more, and 66 of those go beyond 50x.

The tail still matters out of proportion to its size, because what a multiple does to the distance between an entry and a forced close is arithmetic. The CFTC puts the mechanism plainly in its virtual currency advisory: "leverage amplifies the underlying risk, making a change in the cash price even more significant" (CFTC customer advisory). The full arithmetic is in Liquidation explained: why 20x leaves less room than you think, and this article does not repeat it.

For scale outside crypto Telegram, European regulators restricting a related retail product in 2018 set the limit for retail clients at "2:1 for cryptocurrencies" (ESMA, 1 June 2018). That covered contracts for difference sold inside the EU, which is a different instrument in a different jurisdiction from the offshore perpetuals these channels trade. It is still the only official number in the neighbourhood, and 26.27% of the calls in our index ask for at least ten times it.

Are the extremes real, or parser noise?

Real, and they arrive in the channels' own formatting. The highest figure in the corpus comes from a post whose leverage line reads "Leverage: Cross (200х)". The second highest reads "Leveage: 125X", misspelled in the original message.

The middle of the range is written just as loosely. One feed puts the multiple after the letter x, as in "Leverage: x27", "Leverage: x32" and "Leverage: x14", each pulled from a different call in it. The 125x line is quoted without the emoji the channel prints in front of it, and otherwise every quote here is the message as published.

We record what the message says. A parse that follows the channel's own text, typos included, is the only version of this measurement a reader can check against the original post. That is why the extremes are quoted here as text instead of being smoothed into a range.

Why is the median almost entirely one channel?

Because a single publisher wrote it 1,928 times. The highest-output channel in our index supplies 1,928 of the 3,988 signals, or 48.35% of the corpus, and it has stated exactly one leverage value across its entire history: 10x, on every call, with no second value anywhere.

So the sentence "the median signal asks for 10x" is close to a restatement of "the busiest channel in our index has 10x sitting in its template". Take that publisher out and the largest row in the table above loses 1,928 of its 2,118 signals, because nothing else in the corpus is that repetitive at that volume.

The concentration goes further than one account. The top three publishers supply 65.15% of all signals. Three of the accounts in the index are also one feed sold under three brands, with 334 of their 335 signals identical. A fourth account reposts 72 of the same calls in Spanish, so our raw counts carry those calls several times over. The independence problem that creates is covered in how many of the channels we index are actually alive.

Which channels ever change the number?

Most of them do change it, and that is the wrong unit to count. Nineteen publishers state a leverage at all and thirteen have used more than one value. The six that never did are the ones that write most of the calls.

Publisher, by output Signals with a leverage Distinct values used Range Median
Highest output in the index 1,928 1 10x only 10x
Three-brand cloned feed, each brand 335 45 10x to 109x 21x
A 127-signal publisher 127 9 10x to 200x 25x
A 55-signal publisher 55 1 3x only 3x
A 50-signal publisher 50 1 10x only 10x
A 44-signal publisher 37 1 20x only 20x
An 11-signal publisher 11 1 3x only 3x
An 8-signal publisher 2 1 3x only 3x

Those six single-value publishers add up to 2,083 of the 3,700 leverage figures in the corpus. For most calls, then, the multiple was decided once, for the channel, and carried into every message after that. Two of the six rest on eleven and two stated multiples, which is thin evidence of anything; the other four cover 2,070 calls between them.

The cloned feed is the clearest counterexample, with 45 distinct values between 10x and 109x, and its variety belongs to one operation publishing under three brands rather than to three opinions. The unrounded values are what give it away. Counted as published, clones included, 205 calls state 27x, 160 state 14x, 140 state 16x and 132 state 25x, which is not how a person picks a number off a dropdown.

Does lower leverage mean better outcomes in our data?

Sorted by leverage, our outcome data does look like that, and the sorting is the problem. The take rate falls across the range, the pattern is real, and it is still not evidence about leverage, because each bucket below is dominated by a different channel. Scoring every call under our one-take rule, which replays a signal with entry at the published price, one +2% target, the channel's stop or a 10% fallback, and a fixed resolution window, gives this:

Leverage stated Calls scored Reached the take Cancelled: price ran past the target before entry
2x to 5x 25 80.0% 80
6x to 10x 502 80.3% 1,592
11x to 20x 499 59.3% 83
21x to 50x 669 61.6% 110
Above 50x 58 36.2% 8

The drop across the third column is large, 80.0% at the bottom against 36.2% at the top, with the two middle buckets close together at 59.3% and 61.6%. It is the sort of table that gets screenshotted with a caption about discipline, and the caption would be wrong.

Read the fourth column before anything else. The rows are scored on wildly different fractions of what those channels published, which is the first sign that the comparison is between publishers.

Why can that table not be read as cause and effect?

Because the leverage buckets are channels wearing a numeric label. The 6x to 10x row is very largely the publisher that posts 10x on all 1,928 of its calls. The 21x to 50x row is very largely the three-brand cloned feed. The 2x to 5x row is half one publisher that posts 3x and nothing else, and most of the rest is two more channels with a single value each.

So comparing the rows compares publishers. Those publishers differ in every way that decides an outcome: which pairs they call, how far their entries sit from the market, how wide their stops are, and which weeks of which year they were active in. The leverage column is the one difference we happened to sort by.

The shape of the problem has a standard name. The CDC's epidemiology course glossary defines confounding as "the distortion of the association between an exposure and a health outcome by a third variable that is related to both" (CDC, Principles of Epidemiology, Glossary). Swap in our terms and the third variable is the channel, which decides the multiple and influences the outcome at the same time. Experimental design calls the same failure aliasing, which the NIST/SEMATECH handbook describes as the case where "the estimate of an effect also includes the influence of one or more other effects" (NIST/SEMATECH e-Handbook, section 5.7). Once two effects are aliased, no care taken later in the arithmetic pulls them apart.

Three specific things break the causal reading.

  1. The signals with no leverage at all do best. Calls that state no multiple score 84.1% on 182 outcomes, above every bucket in the table. If the number drove the result, leaving it out would be the strongest setting available, which nobody believes and which no exchange would implement. What that row really shows is that the channels which omit the field are different channels.
  2. The denominators are not comparable. The 6x to 10x row discards 1,592 calls whose entry was never reached, against 8 for the top row. Across the whole corpus our one-take rule scores 1,935 outcomes and leaves 1,958 cancelled that way. A row built on the residue of a channel with heavy cancellation is not the same measurement as a row where almost everything filled.
  3. Our replay never reads the leverage field. Whether price reaches the first target before the stop is set by the pair, the entry and the stop distance, none of which the multiple changes. What a multiple does change is what the move is worth in money and whether a real position survives long enough to see it, and neither of those is what this take rate measures. The column we sorted by has no way into the number we sorted.

Getting this wrong has a cost beyond the statistics. A reader who concludes that 10x channels are better channels has bought a screening rule that sorts publishers by a template setting, and will happily follow a weak channel with a conservative-looking number on every post.

What would it take to measure the effect properly?

A comparison inside a single channel: the same publisher, the same style of call, the same pairs, some at one multiple and some at another. That is the only version of this question our data could answer honestly.

It cannot answer it today. The publisher with almost half the corpus uses one value, so it contributes no variation at all. The publishers that do vary the multiple vary the pair and the stop distance alongside it. Their per-channel counts are also small, small enough that any split of them lands where a difference of a few percentage points means very little.

Even a clean comparison would answer a narrow question. Our take rate asks only whether price reached +2% before the stop. What a multiple plausibly changes is survival to the target and the size of every result, which is a question about drawdown, and that is measured in A losing streak is normal: leverage decides what it costs.

So what does the leverage line tell a follower?

What the channel had in its template, and very little else. In particular it says nothing about how much of your account is exposed on the trade, because that is decided by position size and by the distance to the stop.

The confusion is easy to hold, because the exchange screen reinforces it: a 2% move at 10x shows as 20% on the position, so the multiple feels like the size of the bet. The bet is the notional. A $1,000 position gains and loses the same dollars at 2x and at 20x, and what changes is how much margin is tied up and how close the forced close sits.

The venues put it in one line. OKX's margin documentation gives the rule as "Initial margin = Position value / Leverage" (OKX, futures margin calculation rules), with position value as a separate input on the other side of the equation. Binance states the same thing in prose: "the Initial Margin is determined by the leverage you select" (Binance, leverage and margin of USDS-M futures). The one place the two do connect runs the other way, since Binance also caps the multiple you are allowed as your notional grows. Turning a stop distance into a position size is the whole subject of Position sizing when the trade idea is not yours.

Two practical readings follow from the measurement rather than from the arithmetic. The first is that a leverage figure repeated on hundreds of consecutive calls tells you the channel is not sizing anything to the individual trade, whatever the number is. The second is that a channel's habitual multiple can be checked against its own published stop. Whether that stop sits inside the distance the multiple leaves before a forced close costs one subtraction, on numbers the post already gave you.

What this does not prove

Our index is not the market. It holds 3,988 signals from the 22 accounts of the 47 we track that publish anything we can parse, and the sample is weighted toward channels formatted consistently enough to read.

The concentration is severe enough to repeat. One publisher supplies 48.35% of the signals, the top three supply 65.15%, and four accounts carry one feed, three brands of it plus a Spanish-language repost, so 735 signals are the same calls counted more than once. Deduplicating those reposts leaves 3,253 signals. Any figure above that averages across the corpus is an average across a handful of channels, mostly one.

The period is short. The formal span runs from November 2021 to August 2026, but 90.37% of the signals were published in 2026, so this is a picture of recent months with a thin historical tail rather than a five-year trend.

The outcome table is descriptive and stays that way. Nothing here establishes that any level of leverage produces better or worse results, and nothing here establishes the reverse either. The honest statement is that our data cannot separate the multiple from the channel that posted it.

Nothing here alleges misconduct by any channel. Publishing a fixed multiple on every call is a formatting choice, not wrongdoing, and a channel is not responsible for how anyone sizes a position from its message.

Finally, the figures move. Signals resolve as our replay catches up with the market, channels enter and leave the index, and everything above is a cut taken on 12 August 2026.

The practical read

Three questions survive the measurement, and none of them is answered by the number in the post.

Has this channel posted the same leverage on its last fifty calls, which would tell you the figure is a template rather than a judgement? Does its own stop sit inside the distance that multiple leaves before a forced close? And what position size, given that stop, costs you what you have decided to accept?

Where the channels we can score currently stand is at Signal Providers, and the rules our replay applies are on the Methodology page. Nothing here is advice about what to trade, or how much.

Sources