A subscription to a signal channel is usually pictured as somebody's opinion on the market. In our index it is mostly a feed of names that appear in one channel and nowhere else, and bitcoin turns up in fewer than one call in a hundred.
What do signal channels actually call?
Almost everything except the two coins most people mean when they say crypto. Across the 3,988 signals our parser has read from the channels in our index, dated 9 November 2021 to 17 August 2026, the calls cover 497 distinct symbols. Bitcoin appears in 33 of them. Ether appears in 28.
Every signal in the set carries a symbol, so nothing here is an artefact of missing data. What is missing is a centre of gravity: the single most-called name in the whole corpus accounts for 2.18% of the flow, and the list falls away from there.
This piece is about what that shape means for somebody paying for the feed, and about the three things in our own sample that make the headline number smaller than it looks.
How much of the flow is bitcoin and ether?
Under two calls in every hundred, taken together.
| Symbol, all parsed signals to 17 August 2026 | Signals | Share of 3,988 |
|---|---|---|
| BTCUSDT | 33 | 0.83% |
| ETHUSDT | 28 | 0.70% |
| Both together | 1.53% | |
| Everything else | 98.47% |
That is the figure with every quirk of our sample still in it, and the quirks are large. One publisher supplies 48.35% of the corpus, and three of the accounts we index turn out to be one feed under three names. Both effects push the headline share down, so the honest version of the finding needs the same measurement on cuts that remove them.
| Cut of the corpus | Signals | BTC | ETH | Together |
|---|---|---|---|---|
| Everything we parse | 3,988 | 0.83% | 0.70% | 1.53% |
| Minus the republished clone feeds | 3,253 | 0.65% | 0.52% | 1.17% |
| Minus clones and the largest publisher | 1,325 | 1.58% | 1.28% | 2.87% |
Take out the duplication and the one channel that dominates the sample, and bitcoin plus ether still come to 2.87% of the calls. The direction of the finding survives every cut we can make. Its magnitude does not, and anybody quoting 0.8% without the third row is quoting one channel's posting habit.
What sits at the top of the list instead?
Names that mostly appear in one or two feeds. Here is the top of the corpus by call count, with the number of separate channels each symbol appeared in.
| Symbol | Signals | Share of 3,988 | Channels calling it |
|---|---|---|---|
| BEATUSDT | 87 | 2.18% | 3 |
| ESPORTSUSDT | 84 | 2.11% | 3 |
| AKEUSDT | 74 | 1.86% | 8 |
| LABUSDT | 74 | 1.86% | 6 |
| BANKUSDT | 72 | 1.81% | 7 |
| EPICUSDT | 57 | 1.43% | 3 |
| USUSDT | 53 | 1.33% | 7 |
| VELVETUSDT | 52 | 1.30% | 2 |
| UBUSDT | 49 | 1.23% | 4 |
| DEXEUSDT | 49 | 1.23% | 3 |
| CAPUSDT | 48 | 1.20% | 1 |
| GWEIUSDT | 47 | 1.18% | 1 |
| KORUUSDT | 44 | 1.10% | 1 |
| OPUSDT | 42 | 1.05% | 10 |
| BTWUSDT | 41 | 1.03% | 3 |
Three of the fifteen most-called symbols in our whole corpus come from a single channel each, and the widest-reach entry in that table touched ten.
Is this just one channel's habit?
Partly, and the arithmetic of the sample says so plainly. Of the 47 accounts we index, 22 have published signals our parser can read. The largest of those supplied 1,928 calls, which is 48.35% of everything. The top three accounts supply 65.15% between them, and three of the accounts in that group publish an identical stream: 334 of 335 signals shared, with a fourth account sharing 72 of them. Dropping the republishers removes 735 signals from the corpus.
Strip the duplication and the dominant publisher, and 1,325 signals remain. The top of the list changes character completely.
| Most-called symbols, clones and dominant channel removed | Signals | Share of 1,325 |
|---|---|---|
| ZECUSDT | 25 | 1.89% |
| SOLUSDT | 21 | 1.58% |
| BTCUSDT | 21 | 1.58% |
| XRPUSDT | 19 | 1.43% |
| OPUSDT | 18 | 1.36% |
| NEARUSDT | 18 | 1.36% |
So the claim that channels barely call bitcoin holds in both views, while the claim that bitcoin is a rounding error belongs only to the view with the pump-style publisher included. That publisher is why the two views exist. It is also why we report both.
How spread out is the flow?
Very. Half of the 3,988 signals are spread across 59 different symbols, and 80% of them across 169. A total of 143 symbols appear exactly once in almost five years of archive.
The picture barely tightens in the control cut. Take clones and the dominant publisher out and the remaining 1,325 signals still cover 414 symbols, with 70 of them needed to reach half the flow and 195 to reach 80%. Of those 414, 173 appear once.
A feed with that shape is not delivering a view on a handful of instruments that a follower could learn. It is delivering a stream of individually rare names, and the rarity is the part that carries consequences for anyone trying to check the record afterwards.
Do channels agree on what to call?
Half the time they have no overlap at all. Of the 497 symbols in the corpus, 245 were called by exactly one channel. That is 49.3%. Another 127 appear in two or three channels, and only 91 reach five or more.
The symbols with the widest reach across feeds are a different list from the symbols with the most calls:
| Symbol | Channels calling it | Signals |
|---|---|---|
| ZECUSDT | 12 | 40 |
| NEARUSDT | 11 | 30 |
| BCHUSDT | 11 | 24 |
| OPUSDT | 10 | 42 |
| XRPUSDT | 10 | 33 |
| SOLUSDT | 10 | 29 |
| AAVEUSDT | 10 | 26 |
| TIAUSDT | 10 | 25 |
| XLMUSDT | 10 | 20 |
The names a subscriber would recognise are there, and they are the ones several channels have in common. They are not the ones the volume of calls goes to.
Per channel, the sets are wide and mostly private. The largest publisher used 191 distinct symbols across its 1,928 calls. The clone feed used 140 across 335. Two smaller publishers used 83 symbols across 159 calls and 100 symbols across 127 calls, which is close to a new name for every call.
What does a feed of rare names cost the person following it?
We cannot price it from this cut, and it is worth being exact about why. Our snapshot records what each channel published: symbol, direction, entry, targets, stop. It holds no order-book data, so it says nothing about spreads, depth, or traded volume on any of these 497 names.
Two things we have measured do bear on it. The first is fills. Across the outcomes in this corpus, 49.86% are cancelled, meaning price reached the target zone before the published entry ever filled, which is the subject of Half the trades never happen. The second is cost arithmetic: a taker round trip of 0.08% to 0.11% of notional consumes 4% to 5.5% of a 2% target before the trade has to be right about anything, which is worked through in What fees and slippage do to a 2% target.
Whether a thinly followed symbol widens either of those effects is the obvious next question, and it is the one our data cannot answer. Anyone who tells you the answer, in either direction, should be asked what they measured it on.
Can you check a channel's record on a symbol like this?
Rarely, and the corpus shows why. When 143 symbols appear once in the entire archive and 245 come from a single channel, there is no second feed to compare a call against and no run of prior calls on the same instrument to judge it by.
That has a direct effect on evidence. A hit rate needs a run of resolved trades behind it before it says anything, and the arithmetic is in How many trades a hit rate needs before it means anything. A per-symbol record built from one or two calls is not a small sample. It is an anecdote with a percentage sign after it.
It also removes the cheapest verification a subscriber has. Searching a published entry verbatim to see whether the same call appeared in other channels at the same time is how republished feeds get caught, and it is the check described in Clone signal channels. On a symbol that only one channel in our index has ever called, that check returns nothing, and returns nothing whether the channel is honest or not.
Does this correct something we published earlier?
It does. In How many signal channels should you follow we wrote that channels overwhelmingly call liquid perpetuals, and that the pool they draw from is therefore smaller than the tradeable universe suggests. That sentence was measured on a different and much smaller archive of 646 parsed calls covering 270 symbols.
The parsed index does not support it. The pool here is 497 symbols with half the flow spread across 59 of them, which is wider than that sentence implies rather than narrower. And we never measured liquidity in either dataset, so the word had no business in the claim to begin with. We are letting the earlier article stand with this correction attached rather than quietly editing the sentence away.
What this does not prove
Our index is not the market. We hold 50 accounts, and 22 of them publish signals we can parse. That set was assembled from channels large enough to be worth indexing and formatted consistently enough to read, which is a selection, not a sample.
One publisher decides the headline. At 48.35% of the corpus, its posting habits are half of any unweighted figure on this page. Every claim above that depends on it is reported alongside the cut that removes it.
The archive is mostly one summer. 3,604 of the 3,988 signals are dated 2026, and 1,956 of those fall in July alone. The period runs back to November 2021, but the volume is recent, and nothing here supports a statement about how channel behaviour has changed over the years.
We measured strings, not instruments. The symbol is recorded as the channel wrote it. We have not checked that each one maps to a contract a follower could have traded on the venue they use, and a ticker that looks unfamiliar in this table is unfamiliar to us too.
No liquidity, volume, spread or price data is in this cut, so the reasons a channel might prefer rarely called names are outside what we can demonstrate. Nothing here alleges misconduct by any channel, and a wide symbol set is compatible with several explanations at once.
The counts are dated. They come from an index snapshot whose most recent signal is 12 August 2026, and they move as channels publish, as signals resolve, and as accounts enter and leave the index.
The practical read
Before the next subscription renews, three questions come straight out of this data and none of them need our tooling.
Scroll the last thirty calls and count how many names repeat. A feed that names something new every time is a feed nobody can build a record on, including the channel itself.
Take the three most recent symbols and search each one alongside the channel's own history. If the channel has never called that name before, the percentage on its sales page tells you nothing about the trade in front of you.
Then take the same three and search the entry price verbatim across other channels. Silence there is not proof of anything on a name that only one feed calls, which is exactly the position a rare symbol puts you in. How to check a signal channel's track record in 10 minutes covers the checks that still work.
Current ratings for the channels we can score are at Signal Providers. Nothing here recommends any channel, any asset, or any trade.
Sources
- Symbol counts, the top-15 table, the coverage thresholds and the singleton counts:
work/_recount-2026-08-17/slice.json, keycoins, taken read-only from the production index on 17 August 2026. 3,988 signals dated 2021-11-09 to 2026-08-17, all carrying a symbol; 497 distinct symbols; 59 symbols to reach half the flow and 169 to reach 80%; 143 symbols appearing once; BTCUSDT 33 and ETHUSDT 28. - The two control cuts, both computed over their own row sets rather than derived from the headline figures: same file, keys
coins.deduplicatedandcoins.deduplicated_minus_dominant. Clone-dedup cut N=3,253 with BTC 0.65% and ETH 0.52%; clone-and-dominant cut N=1,325 over 414 symbols with BTC 1.58% and ETH 1.28%, 70 symbols to half the flow, 195 to 80%, 173 singletons. - Cross-channel presence, including the counts of symbols appearing in one, two-to-three and five-plus channels: same file, keys
coins.symbols_in_one_channel_only,coins.symbols_in_two_or_three_channelsandcoins.symbols_in_5plus_channels. 245 symbols in one channel only, 127 in two or three, 91 in five or more, maximum 12 channels for a single symbol. - Sample concentration and the clone links: same file, keys
concentrationandclone_links. 22 of 47 providers with parseable signals; largest publisher 1,928 signals at 48.35%; top three at 65.15%; six clone links covering four accounts, 334 of 335 shared signals in the main group and 72 in the fourth; 735 signals dropped as republished. - Per-publisher symbol sets and the 2026 concentration of the archive: same file, keys
channel_life.per_channel,composition.by_yearandcomposition.by_month_2026. 191 symbols over 1,928 calls, 140 over 335, 83 over 159, 100 over 127; 3,604 signals dated 2026 of which 1,956 fall in July. - The cancelled-outcome share: same file, key
duration.excluded, 49.86% of outcomes. The fill mechanics behind it are in Half the trades never happen. - Round-trip cost against a 2% target: What fees and slippage do to a 2% target, which carries its own venue-fee sources.
- Human-readable version of the whole slice, with what changed against the previous count:
work/_recount-2026-08-17/slice.md. The slice itself is rebuilt bywork/_tools/build_article_slice.py.