I traded ICT concepts for about three years. Inversion fair value gaps, change in state of delivery, stop hunts, SMT divergence, m1 and m3 and m5 alignment. I did not abandon them because someone argued me out of it. I abandoned them because my own journal argued me out of it, and it took about nine months of reading my own notes before I listened.
The notes were the evidence#
I log every trade. That turned out to be the thing that killed the method, because the notes stopped being about the market and started being about the rules.
| Date | What I wrote |
|---|---|
| 26 Nov 2025 | hesitation, waiting for m3 cisd is useless, m1 cisd with m3 ifvg |
| 27 Nov 2025 | never wait for m3 cisd. m1 stop hunt and cisd with m3 ifvg entry are enough |
| 3 Dec 2025 | did not wait for a stop hunt or m3 ifvg proper close |
| 16 Dec 2025 | small ifvg? against bias? idk |
| 13 Jan 2026 | 50/50 area where both condition for longs and shorts are present |
| 10 Feb 2026 | plateform is shit first its fake confirmation |
Read those in order. In November I am refining the rules. Three weeks later I am contradicting the refinement. By December I am asking myself a question I cannot answer. By January the setup gives a long and a short signal in the same place. By February I am blaming the broker.
That is not a trader improving. That is a framework failing and a person patching it.
Why it fails the way it does#
ICT concepts are not wrong as descriptions. Price does sweep liquidity. Gaps do get filled. The problem is that they are descriptive, not decidable.
A fair value gap is obvious afterwards. In the moment there are four candidate gaps on three timeframes, and the rule for which one counts is your own judgement. So every losing trade has an explanation available: wrong gap, wrong timeframe, did not wait for the close. The method can never be falsified, only re-specified. That is why my notes multiply rules instead of removing them.
A system you cannot be wrong about is a system you cannot improve.
What replaced it#
VWAP fades. One instrument, one setup, one target.
Price leaves the mean, stays away long enough to prove it was not noise, then comes back. I take the come-back.
The long#
Bars 5 to 14 sit below the lower band. That is the part that matters. Bar 15 is the first M5 candle to close back above it, and that close is the entry. Stop goes under the extreme at 97.5. Target is one R, which the fade reaches before it gets anywhere near VWAP.
The short#
Same thing upside down. Price pushes above the upper band, holds above it, then an M5 candle closes back below.
The rule that does the work#
It is not a reversal trade. I am not calling a top or a bottom.
Price has to break the band, stay outside it, and then revert. The staying is the whole filter. A wick through the band and an immediate snap back is not acceptance, it is noise, and it is the trade that looks most like the setup while being nothing like it.
graph TD
A[Price reaches the band] --> B{Did it close outside?}
B -- no, only a wick --> X[No trade]
B -- yes --> C{Did it stay outside
for several candles?}
C -- no, snapped straight back --> X
C -- yes, acceptance --> D{M5 candle closes
back inside the band?}
D -- no --> E[Wait]
E --> D
D -- yes --> F[Enter on that close
stop beyond the extreme
target 1R]
| Qualifies | Does not | |
|---|---|---|
| Contact with band | closes outside it | wick only |
| Time outside | several candles, held | one candle, snapped back |
| Approach | extended away, returning | already at the mean |
| Trigger | M5 close back inside | trading the wick in real time |
| Target | 1R | held for the full move to VWAP |
Why this is better, and it is not the win rate#
The win rate is not obviously higher. The difference is that I can be wrong about it.
Every term in the setup is a number a computer could check. Did the candle close outside the band, yes or no. Did it stay out for N candles, yes or no. Did an M5 candle close back inside, yes or no. There is no timeframe to argue with afterwards and no fourth gap I should have used instead.
That means a losing month is information rather than an excuse. If the edge stops working I will see it in the data, because the data is answering the same question every time. Under ICT I could never tell the difference between a bad month and a misapplication, so I always concluded misapplication, so I always added a rule.
Three years to learn that the value of a system is not how well it describes the market. It is whether it can tell you that you are wrong.