If you only ever see the trades that worked, you are not looking at a track record. You are looking at the surviving subset β and the size of what is missing is unknowable from the outside.
Survivorship bias is what happens when you draw a conclusion from the things that made it through a filter while ignoring the things that did not. The classic example is analysing the aircraft that returned from missions to decide where to add armour, when the informative data was on the aircraft that never came back.
Trading alerts have exactly this shape. What gets posted publicly is the subset that worked, because that is the subset with promotional value. The calls that failed leave no trace unless someone deliberately records them.
A social feed of green screenshots is the purest form. Each individual post can be entirely genuine β the trade happened, the number is real β while the collection is still meaningless, because you have no idea what fraction of all calls it represents.
Ten winners out of ten calls and ten winners out of two hundred calls produce identical feeds. Nothing visible from the outside distinguishes them.
The more sophisticated version is not deleting losers but redefining them. A call that never reached its entry can be quietly dropped on the basis that it was not really a trade. A call that went against you can be reclassified as a watchlist mention. A target can be adjusted after the move.
Each individual decision sounds defensible. The cumulative effect is a sample that has been filtered by outcome, which is the same problem in a more respectable outfit.
Four things, and all four are checkable. Every call is recorded at the time it is posted, before the outcome is known. The rule for what counts as a trade is fixed in advance and applied to everything. Outcomes are measured from the stated entry rather than the best available price. And the entire dated log is published, including the periods that make the average look worse.
The last point does the real work. A track record that contains a bad month is evidence of a process; one that does not is evidence of editing.
Across the 13 weeks published on this site, weekly hit rates run from 64% to 88%. The 64% week is on the site with the same prominence as the 88% one, because removing it would make the remaining weeks unreadable as evidence.
The same applies to the 41 calls that never triggered β about 9% of everything posted. They are visible precisely because a record you can only see the good parts of is not a record.
The useful question is not what a service's win rate is. It is: what is the denominator, and can I see it?
If the answer is a complete dated log with the weak periods intact, the numbers mean something, whatever they happen to be. If the answer is a highlights reel, no rate quoted alongside it can be verified β and an unverifiable good number is worth less than a verifiable mediocre one.
It is drawing conclusions from only the trades or traders that survived a filter, while the ones that failed are invisible. In alert services it appears as feeds of winning screenshots: each post may be genuine while the collection still says nothing about overall results, because the failures were never recorded.
Because there is no commercial incentive to post anything else, and usually no system recording it. Losing and never-triggered calls have to be logged deliberately at the time they happen; a service that has not built that cannot publish them later even if it wanted to.
Look for a continuous dated series rather than selected highlights, with weak periods still present. A record with no bad weeks in it has almost certainly been filtered. Also check whether calls that never reached their entry are reported at all β that category is the first thing to disappear.
13 weeks of results, week by week, with the flat weeks left in.
See the weekly recapsEducational and informational only Β· Not financial advice Β· Trading is risky and most active traders lose money Β· Past performance is not indicative of future results.