Gold signal channels on Telegram: over 1,200 signals, a result of zero
10 min read · Published 27 Oct 2026
We scored over 1,200 gold signals from three public channels the way an automated copy-trading system would: entering shortly after publication, with the stop and targets from the message and real costs. The channel with the longest history, over five years, averaged −0.012 R per signal — that is, zero. In none of the channels did the direction of the signals differ from random, and in one of them more than 90% of the message numbers were missing.
R is a result in units of risk (−1 R = a loss equal to the distance to the stop). PF (profit factor) is total gains divided by total losses; a PF of 1.00 means breaking even. Hit rate is the share of signals that ended in profit. p is the share of random draws that gave a result at least as good as the signal; the closer it is to zero, the less likely it is that the signal’s result is due to chance.
Hypothesis
A signal channel promises something simple: someone else analyzes the market, and you just repeat their orders. A typical gold signal gives a direction, an entry price or zone, a stop and one or more targets. Channels are also keen to post announcements of targets reached.
Stated at its best, the hypothesis is this: a copier who repeats the signals mechanically — entering shortly after publication, placing the stop and targets from the message and paying real costs — achieves a positive average result in R. We also asked a second question: does the direction given by the channel add anything beyond a random choice between buying and selling?
It started with a question we hear often: could our system read a channel like this and place the orders by itself? Before connecting anything, we checked the history.
How we tested it
Data. In read-only mode, we downloaded the entire visible history of three public gold signal channels, over 12,000 messages in total. The longest history went back more than five years; the other two were much shorter. We took prices from one-minute gold candles at our broker and converted the time of each message to the server clock.
A parser without artificial intelligence. Signals were extracted from the text by a rigid set of rules, not by a language model. This is a deliberate safeguard: content from someone else’s channel must not be able to steer a model that has access to the account. We counted as a signal a short message with a single direction, an entry price and a stop in the same message. Repeats such as “still valid” and copies of the same signal from the previous 48 hours were counted once. We skipped messages with a direction and a price but no stop, because a copier would not know where to put it. The first channel had 800 of them. We checked the parser on a sample of messages, without looking at results, and fixed how it read the targets before we froze the rules.
Execution as a copier would get it. We counted the entry from the open of the first full minute starting at least 60 seconds after publication, which in practice means a delay of 1 to 2 minutes. We tested two variants: entry at market, and a limit order in the middle of the stated zone, valid for 24 hours. When a signal had several zones (adding to the position), each zone got an equal share of the position, and with several targets the position was split equally between them. The stop was always the one from the message, and after 120 hours we closed the position at market. When the stop and a target fell in the same minute, we counted the stop. Costs included the spread plus the account markup, and positions held overnight paid swap.
Controls. The first placebo moved each signal to a random moment within ±60 days, with the same geometry of entry, stop and targets. It answers the question of whether the result depends on the choice of moment or only on the shape of the trade. The second placebo kept the same minute but drew the direction at random, which shows whether the channel’s direction adds anything. We ran each placebo 200 times.
Limits of the measurement. We froze the scoring rules with a checksum before the first result and did a single run. We did not, however, set pass thresholds, so this is a description, not a test with a “pass / fail” verdict. Nor did we model the channel’s later messages, such as “close now”, “hold” or “move the stop”. We measure the result of a copier who repeats the signal and does not follow every follow-up message; someone who reads everything might have done better or worse. We also counted only what the channel left in its history. Deleted messages never made it into the count, so the result is more likely an upper estimate.
Result
| channel, entry | signals | average per signal | PF | hit rate | random direction, p |
|---|---|---|---|---|---|
| first (over 5 years), at market | 1,080 | −0.012 R | 0.98 | 47% | 0.19 |
| first, limit order | 885 | 0.000 R | 1.00 | 41% | 0.83 |
| second, at market | 37 | −0.049 R | 0.88 | 46% | 0.62 |
| second, limit order | 37 | −0.076 R | 0.84 | 43% | 0.79 |
| third, at market | 89 | +0.068 R | 1.16 | 45% | 0.20 |
| third, limit order | 84 | +0.038 R | 1.09 | 43% | 0.52 |
The first channel, after more than five years and 1,080 signals, averaged −0.012 R. Year by year, with market entry, the results were: 2021 −0.23 R, 2022 −0.02, 2023 −0.05, 2024 0.00, 2025 +0.11 and 2026 −0.03 R. Only one year was clearly in profit. The deepest drawdown reached 47.9 R. The same trade shapes at random moments gave −0.034 R (p = 0.25), so the choice of timing added nothing beyond what the geometry alone delivers. A random direction in the same minute gave −0.039 R (p = 0.19). We separately scored 84 messages without an instrument name, in which the channel was most likely trading gold: −0.179 R, PF 0.79.
The second channel gave 37 signals and averaged −0.049 R. Another number matters more than the result: more than 90% of the numbers were missing from its message numbering. Missing numbers mean deleted or service messages. So we were looking only at what the channel had chosen to keep, and even that was in the red.
The third channel did best: 89 signals, +0.068 R and a PF of 1.16. Without the five best signals, however, the average drops to −0.043 R. Of the weeks in its history, 40% ended in profit, and the longest losing streak was 15 signals. A random direction gave a result at least as good in 20% of draws. That is too little data to prove anything. This channel’s numbering was missing 30% of its numbers.
A forward measurement is under way. Since mid-October, we have been recording live every message from one of these channels at the moment of publication, together with edits and deletions, without placing any orders. After about 100 signals we will score them with the same frozen rules. This measurement does not depend on what the channel deletes later.
Why it doesn’t work
The direction is indistinguishable from random. In none of the six rows of the table did the channel’s direction do clearly better than a coin toss in the same minute (p from 0.19 to 0.83). The copier’s result is set by the trade geometry and the cost, not by knowledge of where price will go.
A target hit is not yet a profit. In the second and third channels, the targets were spaced roughly $5 apart, or 0.3–0.5 R, with a stop of $10–13. With a layout like that, the first target is sometimes reached even in signals that end in a loss overall, because the rest of the position then hits the stop. So a “target reached” announcement tells you nothing about the result of the series. What counts is the average R of all signals, stops included.
The author filters the history. A channel can delete a failed signal, and the only trace left is a gap in the numbering. With gaps of about 2% in the first channel, that is a minor correction. With more than 90% in the second, the visible history tells you almost nothing about the channel.
A short history looks better than it is. A few profitable weeks in the third channel are easy to mistake for an edge. With results this widely scattered, that is too little to tell an edge from luck, and without the five best signals the result changes sign.
The copier gets worse terms than the author. They enter later, pay the spread and the markup, and either skip the follow-up messages or read them late. Even if the author has a small edge, delay and cost eat it first where the targets sit close to the entry.
The lesson for you
Judge a signal channel like any strategy: by the average result in R over the whole series, after costs and compared with chance. Announcements of targets hit and screenshots of individual trades are no substitute for that.
Before you start copying:
- Check for gaps in the message numbering. Links to messages in a channel carry consecutive numbers. If a large share of the numbers is missing from the history, you are looking at the author’s selection, not their result.
- Record each signal at the moment it is published: the time, the entry price, the stop and the targets, before you see how it ended. Signals recorded later have already been filtered.
- Count in R, with delay and cost, that is, the way you would really have entered, not at the price in the message.
- Compare with a random direction: the same moments, the same stop and targets, the direction from a coin toss. If the coin does just as well, the channel adds no direction.
- Don’t judge after a few weeks. Calculate the result without the five best signals, and the longest losing streak. Only around a hundred signals give a clear picture.
In your trading journal, record every copied signal with its publication time, entry price and result in R. Once you have a run of trades, the “Luck or edge?” calculator will show the result without the three best trades, the longest losing streak, and how many more trades you need before drawing a conclusion.
Educational content, not investment advice. Past results do not guarantee future results. We describe the result of our mechanical scoring of signals from the visible history of three channels, on our data and with our costs. We do not assess the results of the people who run the channels, or of copiers who follow additional instructions from later messages.