Most signal channels cannot be scored at all, and the reasons are mechanical rather than statistical. The same reasons tell you when a channel's own numbers are unfalsifiable.

Why can most channels not be rated at all?

Because a rating needs a record, and most channels never publish one. Of the 47 channels in our index as of 17 August 2026, 25 produce output that cannot be graded by an outsider under any method.

That is more than half the index, and the reason is almost never that the channel performed badly. Performance does not enter into it. A channel gets excluded before anyone measures anything, because what it posts does not contain the facts a measurement would need.

This matters beyond our index. Every one of the conditions below is something you can check yourself in about a minute, and each one is a case where a channel's own advertised numbers cannot be contradicted by anybody. A record nobody can check is not a good record or a bad one. It is not a record.

What does a channel have to publish for a record to exist?

Four facts per call, published before the outcome is known: direction, entry price, stop, and at least one target.

Miss any one and the call cannot be scored, for reasons that survive any amount of goodwill:

  • No direction, and there is nothing to be right about.
  • No entry price, and there is no way to know whether the trade was even available. Roughly half the entries in our replay were never reachable at the published price, which is the subject of Half the trades never happen.
  • No stop, and the loss has no defined size, so the call cannot resolve as a loss at all. Over half our outcomes are undeterminable for this reason, and what that does to a published win rate is in Why so many signals come without a stop loss.
  • No target, and the win has no defined size either.

A call with all four is falsifiable: it can be replayed against the candles and it can come out wrong. That is the whole bar. It is low, it is mechanical, and most channels do not clear it.

What happens when the entries sit behind a paywall?

Nothing can be parsed, because nothing was published. The free channel shows results; the calls themselves were only ever visible to people who paid.

In practice this appears as posts marked with a lock, or as a results feed whose entries reference calls that live somewhere else. There is no dishonesty in charging for signals. The problem is narrower and worth stating precisely: a paywalled call that is announced only after it resolves cannot be distinguished from a call selected after the fact. Both look identical from outside, and only one of them is a track record.

This is also the mechanism behind the teaser format, where the lock is the entire product. That pattern is Locked teasers inside VIP channels.

What about channels that only post wins?

They are the largest excluded group, and the giveaway is that the wins have no matching entry posts.

The format is a feed of "TP1 hit", "TP2 hit", "+47% closed", often with screenshots. Read it as a sequence and something is missing: the calls. If a channel published an entry for every result it celebrates, the entries would be there in the history, timestamped before the outcome. When they are not, the results are claims about trades whose existence rests entirely on the claimant.

A channel like this can be perfectly honest. It can also be posting only the winners from a much larger set of calls, and nothing in the feed distinguishes the two. That is what unfalsifiable means in practice, and it is why our pipeline classes such channels as ungradeable rather than as poor performers. How to test a posted result yourself is in How to check a PnL screenshot.

Why does the format have to be consistent?

Because a record is built by machine, and a parser that guesses is worse than no parser.

Every channel writes its calls differently. One prints ENTRY: 0.57, another prints a zone, another puts the leverage in brackets with a Cyrillic character that looks exactly like a Latin x. Reading a channel therefore means writing a parsing rule specific to it, and the rule has to be demonstrated against a real sample of that channel's messages rather than eyeballed.

The failure mode is the reason for the strictness. A bad rule does not crash. It silently takes a target's number for its price, or reads a stop written as a percentage as though it were a level, and the channel ends up with statistics describing trades that never existed. The rule in our own procedure is that a channel whose parsing cannot be demonstrated to the thresholds gets no track record at all, and that failing is treated as an acceptable outcome rather than something to work around. A channel with no record is honest about what is not known. A channel with an invented record is not.

What is a greedy parse, and why should you care?

It is when the reader treats a message that has no trade in it as though it did, and it is the one error with no tolerance threshold at all.

Consider a channel that posts market commentary, promotions and calls into the same feed. A rule loose enough to catch every real call will eventually catch an advert containing two numbers, and that advert becomes a trade in the record. Whichever way that phantom resolves, the resulting percentage is fiction.

The reader's version of this problem is the same shape. When you count a channel's wins from scrolling its feed, you are the parser, and you will do exactly what a greedy rule does: read the ambiguous posts in the direction of the story you already have. Counting from a written-down list of calls captured before their outcomes, in the six-column form set out in How to check a track record in 10 minutes, is what removes your own thumb from the scale.

How few calls is too few?

Under about ten scoreable signals, a percentage carries no information, and our acceptance procedure rejects the channel rather than publishing one.

This is not a stylistic preference. Nine calls with six wins produce "67% win rate", a number that would be equally consistent with a channel that is genuinely skilled and one that is guessing. The band of underlying rates consistent with that sample covers nearly everything. Computed properly for every channel we can score, those bands turn out to be enormous for most of them, which is the subject of What sample size makes a hit rate real.

A channel advertising a percentage without the count behind it has told you nothing, and the count is the first thing to ask for.

What if the levels contradict the direction?

Then either the channel is making errors or the record is being read wrong, and the two have to be told apart before anything is published.

A long call whose targets sit below its entry is malformed. A few such calls across hundreds are ordinary human typos, and they are a fact about the channel worth knowing. But when they run past roughly one in twenty, the more likely explanation is that whoever built the reader mixed up the fields. Publishing that channel's statistics would then describe the reader, not the channel.

Our procedure stops at that threshold and requires a person to look at real examples before deciding. It is the point where a plausible-looking number is most likely to be wrong, and the only defence is reading the raw posts with your own eyes. That principle generalises: a figure produced by pattern-matching over free text needs someone to have checked a sample of the matches by hand, and reproducibility alone does not substitute for it.

What does the ungradeable group actually look like?

Split three ways, our index as of 17 August 2026 comes out as 14 channels publishing calls we can score, 8 that used to and have paused or stopped, and 25 whose output cannot be graded from outside at all.

The 22 channels in the first two groups are the entire basis of every figure we publish: 3,988 signals between November 2021 and August 2026, across 497 distinct trading pairs. Three of those accounts are one feed republished under three brands, so the 17 accounts are at most 15 independent publishers, and treating them as 17 would overstate how much independent evidence exists.

None of that makes the other 25 bad channels. It makes them channels about which an outsider cannot responsibly say anything, which is a different statement and the only honest one available.

How do you use this on a channel you are considering?

Run the four questions before you look at any percentage, because the percentage means nothing until they are answered.

  1. Are complete calls posted before the outcome? Direction, entry, stop, at least one target, timestamped ahead of the result.
  2. Are the calls visible, or only the results? If the entries are behind the paywall and the wins are outside it, there is nothing to check.
  3. How many scored calls stand behind the advertised number? Under about ten, the number is noise.
  4. Does the feed's own history support the claims? Scroll back and match celebrated wins to the entry posts that should precede them.

A channel that fails these is not necessarily dishonest. It is unverifiable, and unverifiable is the condition every bad operator relies on, which is why it is worth treating as an answer in itself.

What this does not prove

Our index is 47 channels, and it is not a random sample of the market. It is weighted toward accounts large enough to be worth indexing and formatted consistently enough to have a chance of parsing, so the market's true proportion of ungradeable channels is probably worse than half rather than better.

The thresholds described here are ours. Another team could set a different minimum sample size or tolerate a different rate of malformed calls and produce a different count. What does not change with the thresholds is the underlying point: a call missing its entry, its stop or its target cannot be scored by anyone, on any methodology, however generous.

These figures move as the replay catches up, and they describe our index only.

Current ratings for the channels we can score are at Signal Providers, and what the score is built from is in How to read a Life Score. None of this alleges misconduct by any channel, and nothing here is a recommendation to follow or avoid one.

Sources