A subscription usually outlives the channel it pays for. Counting each publisher once, the median run from a channel's first published call to its last is 76.5 days, half have under three months of history, and the ordinary ending is silence.
How long does a signal channel keep publishing?
Seventy-seven days at the median, counting each publisher once. That is the gap between the first call and the last call we hold, measured across the 20 independent publishers behind the 22 accounts in our index that hold at least one published call. Ten of those 20 ran for less than 90 days.
| Run from first published call to last | 22 accounts | 20 publishers |
|---|---|---|
| Shortest | 14 | 14 |
| Median | 62 | 76.5 |
| Longest | 1,182 | 1,182 |
| Mean | 174.5 | 185.8 |
| Under 90 days | 12 | 10 |
Across the 22 accounts as recorded, a quarter ran 36.75 days or fewer and a quarter ran longer than 203 days. Five accounts ran for less than 30 days.
Those are runs, and a run is a duration, so this is a different question from the one in how many of the channels we index are actually alive. That piece counted who was publishing on the day we looked. This one measures how long the publishing lasted, for everyone in the index who ever started.
Why does the raw median read 62 and the deduplicated one 76.5?
Because three of the 22 accounts are one feed sold under three brand names, and their runs are identical. Sorted by length, the 22 runs put those three 62-day values in the middle of the list, which is exactly where the median of a 22-item list is read. The raw median of 62 is therefore a reading of one publisher, repeated three times and landing in the centre.
Collapse that trio into the single publisher it is and 20 observations remain, and the middle of that list comes out at 76.5 days. Every headline figure in this article uses the 20, and the raw 22 appears wherever the point is about the accounts as recorded.
Fourteen days is a small difference. The mechanism behind it is worth more than the fourteen days: any statistic that counts channels can be shifted by an operator running one feed under several names, and it shifts whichever way that operator's own numbers happen to point. A fourth account carries part of the same stream, sharing 72 signals with the trio, and it runs for 16 days rather than 62. We left that one in place, since a partial republisher with a different run is a different thing from three copies of one feed.
Why is the mean nearly two and a half times the median?
Because one channel drags it. The longest run in the set is 1,182 days, from November 2021 to February 2025, and it is the largest single reason the mean sits at 185.8 days across the 20 publishers while the median sits at 76.5.
That gap is worth keeping in view. A few channels do run for years, and those are the ones a reader is most likely to have heard of, so the category picks up a reputation for durability that the middle of this distribution does not support. The typical publisher here lasted a little over two months.
Averages across this population mislead in every direction for the same reason, which is why the figures here are medians and counts rather than averages.
How long have the quiet channels been quiet?
Half of them posted within days of our cut, and a hard tail has been silent for over a year. Measured on 17 August 2026, the median channel had been quiet for 4.5 days.
| Silence at the 17 August 2026 cut | Channels of 22 |
|---|---|
| More than 30 days | 8 |
| More than 90 days | 7 |
| More than 180 days | 4 |
| More than 365 days | 4 |
The longest silence is 562 days. Note the shape of that column: past six months the count stops falling, and the same four channels account for every silence beyond a year. In our data a channel that has been quiet for six months has been quiet for far longer than that, and the middle ground is thin.
What does the end of a channel actually look like?
It looks like nothing happening. Across all 47 accounts in the index, the status split is 14 active, 1 paused, 7 stopped and 25 we cannot parse at all. None of those statuses was set by hand: our index holds zero manual overrides, so every one of them was derived from posting behaviour.
Two counts of the same population are in circulation and both are right, so here is the bridge. Twenty-two accounts publish calls our parser reads today, and the remaining 25 hold none we can read. The status columns and this article's 22 therefore count the same feeds from two directions.
Of the seven channels silent for more than 90 days, the median run before the silence started was 230 days, with quartiles at 22 and 423.5. Two shapes sit inside that spread.
One is the short burst. A channel publishes 82 calls in 14 days and then stops, and has now been quiet for 479 days. Another publishes 94 calls in 21 days and has been quiet for 378.
The other is the slow fade, and it is the channel that sets the 230-day median rather than a second example of it: 230 days of calls, then a silence that has now run 562 days. In none of these cases is there an announcement, a closing post, or a scandal in our data. The calls simply stop.
One caution belongs with those three illustrations. Three of the seven channels silent past 90 days are accounts we no longer parse, including two of the three just described, so part of what looks like a channel stopping may be our reading of it stopping. That ambiguity has limits worth stating precisely.
Can you tell a dead channel from one we stopped reading?
Not reliably, and this is the part of the topic we cannot resolve with our own data. A gap in our index means we hold no signals from that channel in that period. It does not establish that the channel published none.
Three things keep that ambiguity open, and all three run in the same direction:
- Our scrape window is finite. We started reading each channel at some point, and that point is not the day the channel opened.
- We hold no scrape history in this dataset. The table that would record when each channel was last read is empty, so we cannot separate "the channel went quiet" from "our reader stopped visiting" on a per-channel basis.
- Five accounts are flagged as unparsed while holding old signals. Those are historical imports from before we required a demonstrated parsing rule. For three of the five, the last call we hold is months old, and that is a fact about our reading as much as about the channel. The other two were current at the cut.
So read every silence figure in this article as an upper bound on what we observed, and not as a claim that anyone shut down. Our activity snapshot of the index can say which channels were not publishing on the day we looked. These figures can say how long each gap has run in our records. Neither can say why it started.
Why 76.5 days is not the age of the channel
A run measures our observation of the channel, from the first call we captured to the last. A channel that had been posting for a year before we started reading it shows up here with whatever slice we hold.
That cuts both ways for the reader. The number understates how long some channels have existed, and it accurately states something else: how much published, timestamped, checkable history sits in front of anyone trying to evaluate them. A record that only exists in the part of the archive nobody kept is not available to you either.
Both readings support the same practical move, which is to scroll to the first complete call in the channel yourself and look at its date.
What does a 77-day median do to a subscription?
It makes the subscription the longer-lived of the two. Three monthly renewals already cover more than the median publishing run in our index, and the billing does not stop when the calls do.
That asymmetry is mechanical. Renewal runs by itself and needs nothing from the subscriber; publishing needs the operator to keep working every week. Whatever ends the calls, a wound-down account or a rebrand or simple boredom, it touches only the side of the arrangement that was supplying the trades. The payment side carries on untouched.
Nothing here says any operator planned that. The point is narrower: the two clocks are independent, and only one of them needs attention to keep running.
How much history sits behind the rating you are reading?
Under three months of it, for more than half of them. Eleven of the 20 independent publishers ran for less than 90 days, and 13 of the 22 accounts did. Whatever a rating says about a channel, ours included, it was computed from that channel's run and from nothing else.
This is where run length meets sample size. Take the lower quartile of our set, a channel whose entire run is 32 days: four and a half weeks of trading conditions can explain its advertised percentage on their own. The confidence bands on a record that short are wide enough to cover almost any underlying skill level. The arithmetic is in How many trades a hit rate needs before it means anything.
Regulators have put a number on how short is too short, and it is worth holding against this distribution. Australia's ASIC treats past performance drawn from "an inappropriately short time period" as potentially misleading, and says that for an operation running under a year such a period "would usually be inappropriate and may be misleading" (ASIC Regulatory Guide 234). A year is nearly five times the median run in our index, and ten of our 20 publishers never reached a quarter of it. ASIC's rules reach financial firms in Australia and stop well short of a Telegram channel, so nobody here is going to be fined. The measurement problem is identical either way, and in this market the reader is the one who absorbs it.
What about the channels that never start a clock?
Twenty-five of the 47 accounts in our index are flagged unparsed, meaning nothing they post today is scoreable, and none of them has produced a single readable call, so they have no run at all.
Those 25 are not quiet feeds. Twenty-two feeds in the index carry a content assessment, and the verdicts come out as 10 promotional funnels, 6 commentary feeds, 3 that post results with no entries, and 3 that post entries with no exits.
A funnel or a commentary feed does not go quiet in any way a subscriber notices, and it accumulates no checkable history either. Which formats clear that bar is the subject of What a channel must publish before anyone can rate it, which asks what a feed has to contain. This article assumes it contains it and asks how long that lasted. So the 25 are neither long-lived nor short-lived. They are unmeasured.
How do you check a channel's run and its silence yourself?
Two dates settle it, and both sit in the channel's own feed. The gap between them is the run, which you can hold against a median of 76.5 days here. The days since the later one are the silence, against a median of 4.5 days across our 22 accounts. The rest is scrolling and subtraction, and none of it needs tooling.
- Scroll to the earliest complete call, not the earliest post. A complete call names the pair, the side, the price you enter at and the level that ends the trade. Its date is where the channel's checkable history starts.
- Find the most recent complete call by the same standard. Promotions, chart pictures and running commentary are not calls and do not mark either end.
- Subtract. The gap is the channel's run. Under 90 days puts it in half of our index.
- Measure the silence. Count the days from the last complete call to today. Past six months, in our data, every such channel has turned out to be silent for more than a year.
- Check whether the run predates the advertised record. A percentage covering a longer period than the visible history is describing trades nobody outside can see.
- Repeat for the other channels you follow. Matching first and last dates across two channels are a reason to compare the calls themselves, not a verdict on their own. Our own clone finding rests on 334 of 335 signals being identical between accounts, which dates alone would never have shown.
What this does not prove
Our index is 47 channels, all of them public Telegram accounts, cut at the last signal we hold, 17 August 2026. It is not a random sample of the market, and it holds no other platform, so nothing here describes signal sellers on any other surface.
Only 22 of the 47 have signals, and once the clone trio is collapsed the headline figures rest on 20 observations. That is small enough for one channel to move a quartile, and the set stretches from 14 days to 1,182.
The corpus behind those channels is lopsided in a second way. A single channel accounts for 48.35% of the 3,988 signals in this cut, and three further accounts publish what is effectively one stream under three names, with 334 of 335 signals in common. Volume skew does not reach these medians, because a run is counted once per account regardless of how many signals it holds. The clone group does reach them, which is why the headline figure counts 20 publishers rather than 22 accounts, and why both numbers are shown side by side.
That correction depends on having detected the clones at all. Our detection compares signal text between accounts, so a brand that rewrites or translates the same calls would still be counted as an independent publisher here. The count of 20 is an upper bound on how many separate operations these 22 accounts represent.
Silence is measured against our last signal anywhere, not against a live check of each channel. A channel that resumed after our cut, or that we simply stopped reading, is indistinguishable here from one that stopped.
Nothing in this data records why anybody stopped, and nothing here alleges misconduct by any channel. Ordinary reasons cover most of it. Our records establish that the calls stopped, and the dates on which that happened.
The practical read
The ordinary ending for a signal channel in our index is not a scandal or a public collapse. The calls stop, and nothing in our records marks the moment. The date on the last complete one simply gets older.
That produces two questions worth asking before money moves, and both are answerable from a phone. How many days of complete, timestamped calls does this channel actually have behind it? And what is the date on the most recent one?
The channels with enough published history to be scored are listed at Signal Providers, and the replay rules behind those scores are on the Methodology page. Nothing here is financial advice.
Sources
- Channel run lengths, silence at the cut, and per-channel first and last signal dates:
Signal.objects.values('provider_id').annotate(n=Count('id'), first=Min('published_at'), last=Max('published_at')), run against our index database. Cut of August 12, 2026, working copy inwork/how-long-a-signal-channel-lasts/data-slice.json. - Activity status split, manual override count, feed assessment verdicts, clone links and scrape-run history:
SocialAccount.activity_status,ChannelFeedAssessment,ChannelCloneLinkandChannelScrapeRun, same cut, same file. - Only 14 of the 47 signal channels we index still post - the activity snapshot this article measures durations against.
- What a channel must publish before anyone can rate it - why 33 of the 50 cannot be graded at all.
- ASIC Regulatory Guide 234, June 2026 - past performance drawn from an inappropriately short period.