Going through an old trade journal, I noticed that almost every losing trade had a note next to it from some earlier review, and the notes all said some version of "should have seen it coming." The odd part is that reading the original entries cold, months later, I could not reconstruct how I would have seen it coming. The information was not there at entry. What was there was a reasonable setup that failed. The note was me grading my own homework with the answer key open.
This is hindsight bias, and it is one of the better documented effects in psychology. Once you know how something turned out, your brain quietly rewrites your memory of what you knew beforehand. People asked to recall their own predictions after learning the outcome consistently misremember them in the direction of the result. It happens automatically, and knowing about it does not switch it off. Researchers have tried warning people and paying them to be accurate, and the rewrite happens anyway.
For traders this is worse than embarrassing, because the whole point of a post-trade review is to extract lessons, and hindsight bias feeds you wrong ones in both directions.
The two ways it poisons a review
On losers, the failed outcome makes every warning sign glow in retrospect. The divergence on the daily, the funding rate, the headline from two days before entry, all of it now looks like a flashing light you ignored. So you write a rule. Never trade against funding, never enter ahead of a data print, whatever the loss suggests. But if that signal was not genuinely predictive at the time, and often it was not, you have just added a rule that fits noise. Do this for a year and your process becomes a pile of superstitions, each one commemorating a single loss.
On winners it runs the other way. The win makes the thesis feel obvious in retrospect, so you file it under skill and skip the autopsy. Maybe the thesis was actually wrong and the trade got bailed out by an unrelated move. Maybe the direction and the reasoning were right but the sizing was timid. You will never find out, because the winner made sense and nothing forces you to look. Over time you take credit for luck, confidence drifts up, size drifts up, and the process underneath has less edge than you believe.
A snapshot your future self cannot argue with
You cannot debias your memory, so stop relying on it. The fix is a record created before the outcome exists: a thesis written at entry, timestamped, and never edited afterward. Mine has five fields and takes about two minutes per trade.
- The thesis in one or two sentences, specific enough to be falsifiable. "Expecting volatility" does not count. "Expecting X to reclaim the range high within two weeks because of Y" counts.
- A confidence number. An actual percentage for how likely I think the trade is to work as written.
- A time horizon. A thesis with no expiry can never be wrong, only early, which is the same loophole.
- Invalidation. The specific level or event that means the idea is dead, decided now, while there is no open position to rationalize.
- The main risk. One line on what I think kills this trade.
The confidence number is the field people skip and the one that matters most. It feels arbitrary at first, and honestly the first batch of numbers will be arbitrary. The value shows up in aggregate. Once a few dozen trades are tagged, you can bucket them. Of everything marked around 70%, how many actually worked? If your 80% trades win roughly half the time, you have learned something no individual review could tell you, which is that your conviction signal is miscalibrated and your position sizing should stop listening to it.
The other non-negotiable is append-only. If you can quietly reword a thesis after the trade resolves, you will, and you will not notice yourself doing it. Put the entry somewhere edits are visible or impossible. Email it to yourself, commit it to a git repo, post it in a private channel with timestamps. Any medium works as long as the entry can contradict you six months from now.
Reviewing against the snapshot
With the snapshot in place, you can drop the question "was this a good trade," which memory will always answer with the outcome, and ask narrower questions the record can actually settle.
Did the trade work for the stated reason? A win where the thesis was wrong is a process loss that happened to pay, and I grade those as losses, which stings every time. Did I exit where entry-me said I would, or did I improvise? Improvised exits that happened to work are the most dangerous line item in a journal, because they train you to trust the improvisation. And was there anything in the snapshot, the risk line especially, that I flagged and then ignored? If a danger is written in my own entry, the loss was avoidable. If it is not in the entry and I cannot show I saw it, I do not get to claim I should have seen it. That one rule kills most of the obvious-in-retrospect rewrites on the spot.
Separately, grade the decision and the outcome as two different things, which gives every trade four possible quadrants instead of two. Good decisions lose all the time and bad ones get paid all the time, and poker players call the failure to keep those apart resulting. The off-diagonal trades are where the learning lives. A well-reasoned loser mostly needs to be left alone. A badly reasoned winner needs the hardest look in the whole review, and it is exactly the trade you are least inclined to examine.
A few failure modes from my own attempts. Writing the thesis an hour after entry, once price has already moved, contaminates the confidence number more than you would expect, so write it before or not at all. Reviewing only losers turns the journal into a punishment log and misses half the information. And abandoning the process during a drawdown is nearly universal and exactly backwards, since that is when the calibration data can tell you whether the problem is your read or your luck.
If five fields feel like too much ceremony, start with two: one falsifiable sentence and a confidence number, timestamped where you cannot edit them. That alone is enough to catch your memory lying to you, and the first month of results is usually humbling in a useful way.