The channel you scroll today is not the feed its subscribers traded: remove three deep losses from twenty calls, and the total changes sign while the worst stretch nearly vanishes. Here is what a late joiner sees, what a record made at first read keeps, and what the edit count on a channel page does and does not say.
What does someone who joins today actually see?
They see every message that still exists, and nothing that tells them what used to be there. Telegram's own channels FAQ states both halves: "New subscribers can see the entire message history in a channel once they join," and "If a message is deleted in a channel, it will disappear for all subscribers."
Read together, those sentences describe the late joiner's position exactly. The history is complete in the sense that everything still present is shown. A deleted call leaves no placeholder in the feed and no "message removed" line. Telegram does keep a log of recent actions, deletions included, but reading it requires admin rights in the channel (Telegram API: channels.getAdminLog). A subscriber has no view of it.
So the person scrolling back through six months of calls is looking at the calls that survived. Every one of them can be genuine. There is no need for a fake screenshot or an invented trade. The feed can still describe a trading history that no subscriber lived through.
What deletion does to a win rate, and why Telegram's "edited" mark settles less than people assume, is covered in why a channel deletes a losing call. This page is about the two numbers that move further: the total and the drawdown.
An illustration: twenty calls, three of them gone
The example below is invented to make the arithmetic visible and does not describe any real channel.
A signal channel posts twenty calls in order. Each figure is the move from entry to exit for that call, as a channel would print it, with no leverage:
+4, +6, −3, +5, −22, +3, +8, −2, −18, +4, +7, −4, +3, +5, −25, +6, −3, +2, +4, −2
Three of those are deep stop-outs: the fifth, the ninth and the fifteenth calls. Later, those three messages are deleted. Nothing else is touched, and the small losses stay in the feed, which makes what remains look like an honest record with ordinary bad days.
| The feed as it was posted | The feed after three deletions | |
|---|---|---|
| Calls | 20 | 17 |
| Sum of the call percentages | −22% | +43% |
| Largest fall from a peak of that running sum | 41 points | 4 points |
| Worst single call | −25% | −4% |
| Account risking 10% of balance per call | −2.26% | +4.38% |
Removing 15% of the messages turns a losing feed into a winning one. The share of winning calls only moves from 60% to about 71%, which looks unremarkable. The total switches sign, and the worst stretch shrinks to a tenth of its size.
The account row shows that the direction holds once position size is involved. A sum of call percentages is how channels usually present results, and it says nothing about how much of an account went into each call. How a printed total can drift from the list of calls it claims to add up is covered in how to check a channel's results post.
Why the drawdown moves more than the total
A drawdown is the deepest fall from a high point to a later low. A paper on the measure calls the maximum drawdown "the maximum cumulative loss from a peak to a following bottom". It describes the figure as "a measure of how sustained one's losses can be" (Magdon-Ismail and Atiya, Risk, 2004).
Deleting losers distorts this figure most because the deleted calls are, by construction, the ones that dug the holes. In the full feed of the illustration, the running sum climbs to +12 after the fourth call, then falls to −29 after the fifteenth: 41 points from top to bottom. With the three stop-outs gone, the deepest dip anywhere in the feed is 4 points.
Consider two people looking at the same channel. The first subscribed right after the fourth call, at the high point, and followed everything from there. By the fifteenth call they were 41 points under where they started, and after the twentieth they were still 34 down. The second arrived after the deletions and scrolled back through a feed in which the worst stretch costs 4 points. Only the first of them knows what following the channel was like. Why the moment someone joins changes the drawdown they live through is worked through in the drawdown you inherit when you copy-trade.
This pattern has a name in finance: survivorship bias, the distortion that appears when the failures are removed from a sample and only the survivors are measured. Ibbotson and Chen describe it in hedge-fund databases: "When a fund fails, it is often removed from a database along with its performance history" (Yale ICF Working Paper 06-10, 2006). A channel feed works the same way at the level of single calls. The calls that failed worst are the ones the channel has the strongest reason to remove.
What a record made at first read keeps
A record kept outside the channel holds each call as it read the first time it was seen, and it does not shrink when the channel does. On ChainRated, every signal message is stored as a snapshot at first read, and every statistic on a channel's page is computed from that snapshot. A message deleted afterwards stays in the statistics. A message edited afterwards keeps counting as it was first read, so the edit does not move the numbers.
Applied to the illustration: if all twenty calls were read before the three deletions, the channel's page keeps the −22% total and the 41-point drawdown, while the feed on Telegram shows +43% and 4 points. The two disagree, and the record is the one that matches what a subscriber who was there experienced.
The site itself cannot quietly rewrite the stored snapshots either. Each day's new snapshots are hashed into a single value written to a public blockchain, so a stored call changed later would no longer match what was anchored. How that works, and what it does not prove, is set out in how we anchor every signal on Arbitrum.
The limit of such a record is when it starts. The snapshot is taken at the first read, which can come after the moment of posting. A call published and deleted before it was first read never enters the record at all, and a call edited before its first read is stored in its edited form. The record is complete from first read onwards. It cannot vouch for what happened in a channel before it was looking.
What the "Edited after publication" line says
The line is a count of mismatches with a date attached. When there is something to report, a channel's page shows this text, with N, M and K standing in for the numbers:
Edited after publication: N of M signals we re-read at the source no longer say what they said when we recorded them, and K are no longer in the channel. That the text changed is a fact; why it changed is not ours to state. Statistics on this page use each signal as first published, so an edit does not move them. Last checked [date].
Each part of it is deliberately narrow:
- M is how many of the channel's signals were actually re-read at the source. Re-reading covers signals published in the last 90 days, so M is the size of the checked window; older history lies outside it.
- N counts signals whose current text no longer matches the stored snapshot. The comparison is of the text itself, and the stored record keeps no list of what changed, so nothing in the line shows whether the entry moved or a typo was fixed.
- K counts signals that are gone from the channel. A disappearance is recorded only when the channel can be shown to be readable at that moment, so a failure on the reading side is not published as a deletion by the channel.
- The date says when the window was last re-read.
A count, not an accusation. Channels fix typos, correct a mistyped ticker, add a note about a partial close, delete a duplicate post. All of those register as changed or missing text, and none of them is manipulation. The line reports that the record and the channel now disagree, and leaves the reason open, because the text cannot show it.
The count does carry weight. It feeds the Honesty axis of a channel's score, which carries a quarter of the total. Inside that axis, the edits-and-deletions measure scores 100 for a channel with none and sits at the floor for one that changed or removed 10% of its recent posts. The full scoring rules are in the methodology.
How often does the text actually change?
Rarely, and in few channels. As of 6 October 2026, the record holds 36,643 signals. Of those, 20,150 signals across 63 channels have been re-read at the source, each while it was less than 90 days old. Among the re-read signals, 29 no longer match their snapshot, and those 29 sit in 3 channels. Another 4 are no longer in the channel, all 4 in a single channel. Together that is 33 of 20,150, about 0.16%.
Three limits apply to those numbers. They cover the re-read window only: an edit or deletion to a signal more than 90 days old goes unseen, so the figure is a floor for the whole record. They cannot include a call removed before its first read, which by construction never reached the record. And they say nothing about why any of the 33 changed.
What they do show is that a changed or missing message is unusual. At least 59 of the 63 re-read channels show neither an edit nor a deletion, which is part of why a count on one channel's page is worth reading when it appears.
What a missing line does and does not mean
The line appears only when something was found. With zero edits and zero deletions, the page shows nothing, because a clean note on nearly every channel would be read past.
That silence can have more than one cause. It can mean the signals were re-read and nothing differed. It can also mean nothing in the channel has been re-read yet, or the channel has posted no signals in the last 90 days. All three leave the same thing on record: no mismatch in the window that was checked. A signal older than 90 days, or a call removed before it was first read, falls outside that window whatever happened to it.
Reading a feed with the gap in mind
The feed is the version of a signal channel's history that the channel chose to leave standing. That can be the full history. A late joiner has no way to tell from the feed alone.
Three habits follow from this:
- Read the total and the drawdown, and treat a high share of winners as weak evidence. In the illustration, the share of winners barely changed while the total flipped sign. Ask what the deepest stretch cost and whether it is plausible that a channel trading for months never had one deeper than a few points.
- Compare the feed with a record made at first read wherever one exists. A channel's statistics computed from first-read snapshots include the calls the feed no longer shows, and a gap between the two is information.
- Read the edit line as a count with a date. Look at N against M, at how recent the check was, and at whether the mismatches look like one cleanup or a pattern. Then read the messages that remain. The number tells you where to look, and the conclusion stays yours to draw.
A record taken when calls are first read cannot recover a call that was gone before it was first read. From that point on, it stops the feed from being the only account of what the channel posted.