Most investors keep exactly one record of their decisions: the brokerage statement. It tells you what you bought, at what price, and what it's worth now. It tells you nothing about why you bought it, what you expected to happen, or how you felt at the time. So when a position works out, you can't separate skill from luck — and when it fails, you can't tell whether your reasoning was wrong or just early.
A decision journal fixes this. It's a structured, written record of the thinking behind each investing decision, captured before the outcome is known. By the end of this guide you'll be able to set one up, know exactly what to log, and run a review cadence that turns your own history into your best research tool.
TL;DR
- A decision journal records your reasoning at the moment of decision, locked in before the outcome — the only way to defeat hindsight bias.
- Log five fields per decision: thesis, expected outcome, conviction, what would prove you wrong, and emotional state.
- The value is in the review, not the entry. Re-read old entries on a fixed cadence and grade the reasoning, not the result.
- Separate process quality from outcome quality — a good decision can lose money and a bad one can make it. Track both on a 2×2.
- Start with a plain template you'll actually use. A journal you maintain beats a perfect one you abandon.
Why outcomes are a misleading teacher
The core problem is hindsight bias: once you know how something turned out, your memory rewrites what you "knew" beforehand. A stock you bought on a vague hunch that tripled becomes, in memory, a decision you reasoned your way into. The lesson you extract — "trust my hunches" — is the opposite of what actually happened.
A decision journal works because it timestamps your reasoning and freezes it. When you review, you compare the recorded thesis against reality, not your reconstructed memory of it. This is the single mechanism that makes the whole practice work, and it only works if the entry is written before the outcome is known.
| What you have | What it tells you | What it hides |
|---|---|---|
| Brokerage statement | What you did, current P&L | Why you did it, what you expected |
| Memory | A story that feels true | Rewritten by hindsight bias |
| Decision journal | Your reasoning, frozen in time | Nothing — that's the point |
The five fields of a decision entry
A useful entry is short and structured. Capture these five fields every time. Anything more and you'll stop doing it; anything less and the review is toothless.
| Field | Question it answers | Example |
|---|---|---|
| Thesis | Why am I doing this, in one or two sentences? | "Margins are expanding faster than the market expects; re-rating likely within 4 quarters." |
| Expected outcome | What specifically do I think happens, and by when? | "+25% over 18 months; revisit if flat after 12." |
| Conviction | How sure am I, and how big is the position because of it? | "Medium — 3% position, not 8%." |
| Disconfirming evidence | What would tell me I'm wrong? | "Two consecutive quarters of margin compression." |
| Emotional state | What am I feeling right now? | "FOMO — it's up 15% this week and I'm afraid of missing more." |
The disconfirming evidence field is the one most people skip and the one that matters most. Writing down in advance what would prove you wrong is the difference between an investment thesis and a hope. It also gives your future self a pre-committed exit condition, made while you were calm.
The emotional state field looks soft but earns its place. Patterns like "every entry tagged FOMO underperformed" are some of the clearest signals a journal surfaces — and you can only see them if you logged the feeling honestly at the time.
Process vs. outcome: grade the right thing
The hardest discipline in reviewing a journal is refusing to grade decisions by their results. Markets are noisy; a sound decision can lose money and a reckless one can win. If you reward yourself for outcomes, you'll learn to gamble. Grade the process.
Use this 2×2 every review:
| Good outcome | Bad outcome | |
|---|---|---|
| Good process | Deserved win — repeat it | Bad luck — keep doing this |
| Bad process | Got lucky — don't repeat it | Deserved loss — fix the process |
The two diagonal cells are where the learning lives. The "got lucky" cell (bad process, good outcome) is the most dangerous, because the win tempts you to keep a broken process. The "bad luck" cell (good process, bad outcome) is where most people abandon a sound strategy at exactly the wrong time. A journal lets you label each decision honestly because the process is written down and can't be re-litigated by the result.
A review cadence that actually compounds
An entry is a deposit; the review is the interest. Without a fixed cadence, journals become write-only and die. Run three loops:
- Weekly (10 min) — Did I open any positions without an entry? Backfill or close the gap. This is hygiene, not analysis.
- Quarterly (45 min) — Re-read every entry whose "revisit by" date has passed. Score each on the process/outcome 2×2. Note any thesis whose disconfirming evidence has actually appeared — those are forced action items.
- Annual (2 hr) — Tag patterns across the year. Which thesis types worked? Which emotional states preceded your worst decisions? What's your honest hit rate on conviction-weighted calls?
The annual review is where individual entries become a data set. You're no longer asking "was this trade good?" but "what kind of investor does my own record say I am?"
What to log — and what to leave out
A common failure is logging too much. The journal is not a research dump; it's the decision record. Keep the supporting research wherever you already keep notes and link to it. The entry itself stays to the five fields.
- Log: every buy, sell, and deliberate hold (a hold is a decision too — you chose not to sell). Log decisions you considered and rejected; "why I didn't buy" entries are often the most instructive.
- Don't log: routine rebalancing, dividend reinvestment, or anything mechanical with no judgment in it. Noise dilutes the signal.
A "deliberate hold" entry deserves special mention. Most people never record the decision to not act, so their journal is silent during exactly the periods — drawdowns, melt-ups — when their judgment is most tested. A one-line "holding through this, thesis intact, disconfirming evidence not present" is a real entry.
Common mistakes
- Writing after the outcome is partly known. Even a day's price move contaminates the entry. Write at decision time or not at all.
- Grading by result. The fastest way to learn the wrong lessons. Use the 2×2.
- Skipping the disconfirming-evidence field. Without a pre-committed "I was wrong" condition, every loss becomes "just be patient."
- Over-engineering the template. Fifteen fields and a scoring rubric guarantee you'll quit in a month. Five fields, every time.
- Never reviewing. An unread journal is a diary. The review cadence is the entire point.
- Editing old entries. Append corrections; never overwrite. The frozen original is the asset.
Summary + next step
A decision journal converts your scattered, memory-distorted history into a clean data set about one investor: you. Capture five fields before the outcome, separate process from outcome when you review, and run a fixed weekly/quarterly/annual cadence. The discipline is small; the compounding is not.
The practical blocker is usually retrieval — entries pile up and you can't find the three from two years ago that rhyme with the decision in front of you today. That's a search problem, and keyword search fails here because you rarely remember the exact words you used. Meaning-based retrieval does, which is exactly what [semantic search](../ai-notetaking/what-is-semantic-search.md) is for: ask "times I bought into a hyped runup" and surface every FOMO-tagged entry regardless of the words you wrote. Keep the journal in JustJot.ai, tag each entry with its conviction and emotional state, and let your own past decisions become searchable research.
Try this: Open the last position you took. Write its five fields as of the day you bought it — honestly, without peeking at how it's done since. That single backfilled entry is your first data point.