Building a personal decision journal with AI help

The highest-leverage self-improvement practice I know is the decision journal. It takes 10 minutes per entry. It pays off in years. Almost no one does it.

Here’s what a decision journal is, why it’s so powerful, what kills most attempts at maintaining one, and how AI decision tools make the practice practical — not replaceable, but practical — for people who bounced off the pen-and-paper version.

What a decision journal is

A decision journal is a persistent record of your non-trivial decisions, made at the moment of decision, capturing:

• The decision itself
• The options you considered
• Your reasoning
• Your expected outcome
• Your confidence
• What would have changed your mind

Later — weeks, months, or years later — you revisit each entry and record the actual outcome. The gap between your expected outcome and actual outcome is where the self-improvement lives.

The practice is ancient (Marcus Aurelius kept something close to it), and Annie Duke’s Thinking in Bets popularized the modern version. The value proposition is simple: you cannot improve your judgment without feedback, and feedback on judgment requires knowing what you thought before the outcome was known.

Why most people bounce off

Three reasons the practice doesn’t stick:

Friction at the moment of decision. You’re in the middle of a busy week. You make a call. The idea of opening a notebook, writing for 10 minutes, and logging the decision feels unaffordable. You skip it. Two weeks later you’ve made six decisions; you intended to journal all of them; you journaled none.

No scaffolding for what to write. Faced with a blank page, most people don’t know what to record. They write too little (“picked HubSpot”) or too much (3 pages of narrative). Neither is useful. The scaffolding — explicit fields for options, confidence, expected outcome, change-my-mind — is what makes the journal reviewable later.

No review cadence. Even if you log decisions, you probably don’t revisit them. Without the review step, the journal is half a practice — you’re writing but not learning. The calibration feedback loop never closes.

How AI tools reduce the friction

The three bouncing points above all dissolve with tooling.

Structured decision tools create the artifact as a byproduct. When you run a decision through Phronesis, you automatically produce a journal-ready artifact: question, options, criteria, ranking, confidence, agreement level. You didn’t do extra work. The journal entry is a side effect of the decision.

The template is built in. Phronesis forces the structure. You don’t have to decide what to write — you filled out the form to run the decision, and the form is the journal entry.

Review is cheap. Every Phronesis run gets a run_id and persists. Visit the history page and every past decision is there, in order, searchable. The review step is a browse through a list, not an archeological dig through notebooks.

This doesn’t eliminate the practice’s hardest part — the discipline of actually recording the outcome later — but it removes the friction that kills most attempts in week one.

What to log (and what not to)

A good decision journal captures the decisions that are:

• Non-trivial. If the decision took less than 5 minutes of thought, skip it. Log only the ones where you actually deliberated.
• Reversible or not, both matter. You want a mix so you can calibrate differently for each.
• Have a clear counterfactual. “Ship this feature or hold” is logable; “work hard this week” is not.
• Have a future outcome you can observe. If you can’t tell whether you were right, you can’t learn.

A good journal does not try to capture:

• Every choice of the day (too much noise)
• Decisions made by others that you merely observed
• Decisions where the outcome is unobservable
• Decisions that are pure preference (pizza vs pasta)

Rough target: 1–3 entries per week. Not more. More is performance; the practice is about quality of attention, not quantity of entries.

The fields that matter

A minimal but useful journal entry has six fields:

1. Date + decision statement. One sentence. “Should I take the new job offer?”
2. Options considered. At least two; three is better.
3. Weighted criteria. What mattered, with rough weights.
4. Decision + reasoning. What you picked and why.
5. Expected outcome + confidence. “Job will be net positive; 70% confident.” The confidence number is the most important field.
6. What would change my mind. A specific condition. “If I learn the team has had 30% annual turnover in the last 3 years, I would reconsider.”

Six fields. Takes 10 minutes. This is the unit of the practice.

For important decisions, add:

• Worst-case scenario. If this goes badly, what does that look like?
• Key assumption. What am I assuming that I haven’t verified?
• Review date. When will I revisit this?

The review cadence

The decision journal is half a practice without review. Two review cadences:

Quarterly. Every three months, open the journal and review all entries where the outcome is now known. For each:

• What was your expected outcome? Actual outcome?
• What was your confidence? Was it calibrated?
• What, in retrospect, should you have weighted differently?
• What pattern do you see across multiple entries?

The quarterly review is the learning. Without it, the journal is a file.

Event-triggered. When an outcome lands — you quit the new job, the vendor you picked delivered, the hire you made turned out well — immediately update the journal entry. Don’t wait for the quarterly review to capture the outcome; the memory of what you thought at decision time degrades fast.

The calibration metric

The single most valuable output of a decision journal is calibration. Over time, you build a dataset of “I was X% confident, actual hit rate was Y%.”

If you’re well-calibrated, when you say 70% confident, you’re right 70% of the time. Most people are poorly calibrated — they’re over-confident on their judgment. Say 70%, right 55%. Say 90%, right 70%. This is a well-documented finding across domains and professions.

The journal produces the dataset that lets you see your own calibration. Three months of entries, graded against outcomes, show you whether your 70%-confident decisions actually hit 70% of the time. If they don’t — and for most people they don’t — you now have a personal adjustment factor.

Use the factor. If your stated 70% maps to actual 55%, when you catch yourself feeling 70% confident, mentally downgrade to 55%. Over time your reported confidence will start to track reality, and your decisions will improve because you’re no longer overcommitting to false-confident picks.

The “outcome ≠ quality” trap

One important trap: a decision outcome is not the same as a decision quality. Sometimes you make a good decision and get a bad outcome (bad luck). Sometimes you make a bad decision and get a good outcome (good luck). The journal has to evaluate on process quality not outcome quality.

Specifically:

• If your reasoning at the time was sound given the information you had, count it as a good decision regardless of outcome.
• If your reasoning had a hole — you ignored available evidence, you didn’t consider enough options, you were miscalibrated — count it as a bad decision even if the outcome was good.

This is hard. Our brains want to grade decisions by outcomes. The journal’s review discipline is specifically the practice of separating the two. It’s the mental move that distinguishes bettors from gamblers, and it’s trainable.

The Phronesis-integrated workflow

Here’s a concrete workflow that combines the journal practice with Phronesis:

1. Decision moment: Run the decision through Phronesis. Fill in the question, options, criteria, and get the ranked output.
2. Before committing: Add three extra notes in a plain text file or the review doc — your expected outcome, your confidence, and what would change your mind. These aren’t captured by Phronesis.
3. Save the Phronesis run_id with the notes. The Phronesis run has the structured part; your notes have the subjective part.
4. Set a calendar reminder for the review date (usually 3 or 6 months).
5. At review: Open the Phronesis run. Read the ranking. Compare to what actually happened. Grade yourself on process.

This is 12 minutes at decision time, 8 minutes at review time. For a decision that could ripple for years, 20 minutes total is nothing.

A sample entry

To make this concrete, here’s what a journal entry looks like for a real decision I kept:

2026-03-15 — Whether to accept the speaking invite at conference X

Options: Accept and give the talk / Accept and try to hand off / Decline

Criteria: Audience fit (30%), prep time cost (25%), downstream relationships (20%), my energy level that week (15%), talk quality bar I can hit (10%)

Phronesis winner: Accept and give the talk. Confidence: 62%. Analysts split 2-1 on “energy level” dimension.

My prediction: The talk goes okay-to-good, I’m exhausted for 10 days after, and I get 1 useful relationship out of it.

Confidence in prediction: 65%.

What would change my mind: If a bigger opportunity came up in the same week.

Review in 3 months: 2026-06-15.

Three months later:

Outcome: Talk went well (better than predicted). Exhausted for ~5 days (less than predicted). Got 3 useful relationships (more than predicted). All three predictions were under-confident. Pattern across recent entries: I systematically underestimate upside. Update my calibration adjustment.

That’s the practice. 12 minutes in March, 8 minutes in June, and I now know something specific about my own calibration that will improve the next 20 decisions.

The broader point

A decision journal is one of those practices that sounds fussy and academic and turns out to be transformative. The value compounds across years because the calibration feedback loop runs every quarter. You become a better decision-maker not because you got smarter but because you got more feedback on your own patterns than nearly anyone you know.

The tooling exists to make the practice nearly frictionless. Phronesis produces the structured artifact as a byproduct of the decision. History holds the entries. Calendar reminders handle the review cadence. All that’s left is the 20 minutes per meaningful decision, which is the smallest investment with the largest return in the self-improvement catalog.

Start with your next real decision. Do the 12 minutes. Set the calendar reminder. In three months, grade yourself honestly. By six months, you’ll notice the difference.