Building a Prediction-Market Trading Journal That Changes Decisions

A prediction-market trading journal is not a profit diary. It is a structured record of the forecast you made, the price you paid, the evidence you had, the rule you relied on, and what happened. Done properly, it lets you separate luck from skill and locate the specific step in your process that fails. Done as a list of wins and losses, it teaches you almost nothing.

The test of a journal is simple: six months later, can you tell whether a losing trade was a bad decision or a good decision with a bad outcome? If the record cannot answer that, it is missing the fields that matter.

Record the decision, not just the result

Most traders record the market, the side, the size and the profit. That is enough to compute a return and not enough to improve.

A useful entry has nine parts.

1. Timestamp and market. The market name, the URL, and the time of the decision. Prices move, and memory rewrites why you acted.
2. Thesis in one sentence. "I believe X because Y." If it takes a paragraph, the edge is not clear enough to trade.
3. Your probability and the market price. Record both the midpoint at the moment of the decision and the price you actually paid.
4. Catalyst and date. What will resolve this, and when.
5. Resolution source and rule. Quote the exact sentence from the rules that decides the market. Polymarket's documentation is explicit that the rules, not the title, define resolution.
6. Size, fees and spread. Position size in dollars and in percent of bankroll, the fee paid, and the spread at entry.
7. Falsifier. The specific fact that would prove the thesis wrong. This is the field most people skip, and it is the one that makes the review possible.
8. Exit plan. Target, invalidation, time stop, and whether you intend to hold to resolution or exit on repricing.
9. Outcome and post-mortem. Resolution date, payout, profit or loss, and what you would do differently.

Why the rule quote matters

The most expensive mistakes in prediction markets are not directional. They are definitional. A trader buys "will the official be indicted" because a leak says an indictment is coming, and the market resolves on whether a charge was formally announced by a named authority. The trade was based on a true story and a false reading of the contract.

Recording the exact rule at entry does two things. It forces you to read the rule before you trade, and it creates evidence of what you believed the rule said. When a resolution surprises you, the journal shows whether the rule was ambiguous or whether you misread it.

The same applies to resolution mechanics. Polymarket resolves through the UMA optimistic oracle, with a proposal bond, a two-hour challenge period, and a longer dispute process if the outcome is contested. A market can resolve 50/50. If your journal never records the resolution process, you will attribute a dispute loss to bad luck rather than to a risk you could have priced.

A template in prose

For example, an entry might read:

"14 October, 09:40. Market: the agency's rule publication before the December meeting. Rule quote: 'Resolves Yes if the agency publishes the final rule in the Federal Register on or before 31 December.' My estimate: 0.62. Mid at decision: 0.54. I bought 800 shares at 0.56, which is 4.5% of bankroll; fee $7.88. Thesis: the agency has published a draft, the comment period closed, and the remaining procedural steps normally take six weeks. Catalyst: the agency's open meeting on 21 October. Falsifier: the meeting agenda omits the item, or the published rule is an interim final rule with a different effective date. Exit plan: hold to resolution; sell half if the price exceeds 0.75 before the meeting. Outcome: resolved Yes on 18 December. PnL +$344. What I would repeat: reading the Federal Register definition before sizing. What I would change: I did not check whether the interim-rule variant would count, which was the main ambiguity."

The market name is a placeholder; the structure is the point. Note what the entry does not contain: a prediction about the price next week, a narrative about the news cycle, or a claim about being right. It contains the decision inputs and the falsifier.

Review cadence

Three reviews, at different frequencies, catch different errors.

After each resolution. Five minutes. Fill in the outcome, the payout, and one sentence on whether the thesis was right for the reason you expected. This is the raw data; do it while the position is still fresh.

Weekly. Group the entries by category and by catalyst type. Look for repeated mistakes: the same rule misread, the same size error, the same tendency to enter before verification.

Monthly. Review calibration and execution, not individual trades. Individual outcomes are noisy; the aggregate is where the signal is.

Metrics that matter

Expectancy per trade. Win rate multiplied by average win, minus loss rate multiplied by average loss, net of fees. A 55% win rate with a 1-to-2 reward-to-risk ratio is a losing strategy. A 40% win rate with a 2-to-1 ratio can be a good one. Win rate alone is not a metric; it is a component.

Calibration. Bin your probability forecasts, for example 50% to 60%, 60% to 70%, and so on, and compare each bin with the realised frequency. If your 70% forecasts come true 55% of the time, you are overconfident, regardless of whether you made money.

Brier score. The mean squared difference between your forecast probability and the outcome, where the outcome is 1 or 0. Lower is better. It rewards being both accurate and honest about uncertainty, and it lets you compare forecasts across markets with different prices.

Execution slippage. The difference between the midpoint at the moment of your decision and the price you actually paid. This separates a good forecast from a good trade. A trader who is right about the event but pays three cents through the spread has a different problem from one who is wrong about the event.

Rule surprises. The count of markets where the resolution differed from your reading. Even one is worth a review; a pattern means you are trading rules you do not understand.

Capital lock-up. Days from entry to resolution, and the return per day of capital committed. A 5% gain over six months is not the same trade as a 5% gain over three days.

Maximum drawdown. The peak-to-trough decline in bankroll, and how it compares with what your sizing implied. A drawdown much larger than the model predicted is evidence that your positions are more correlated than you thought.

Process versus outcome

The hardest discipline is judging a decision by the information available when you made it. A good decision can produce a bad outcome: the probability was 70%, and the 30% happened. A bad decision can produce a good outcome: you bought an ambiguous contract for the wrong reason and the resolution went your way.

The journal is the only defence against this confusion, because it preserves what you knew at the time. If the entry records the thesis, the price and the falsifier, you can ask whether the process was sound. If it records only the profit, you will learn to repeat whatever worked, including the mistakes.

What the journal cannot tell you

It cannot turn a small sample into evidence. A few dozen resolved bets will not distinguish a 55% forecaster from a 50% one, because the binomial noise is larger than the difference. You need hundreds of resolutions, across different market types, before calibration numbers mean much.

It cannot correct for a changing market. If the crowd gets smarter or the fees change, the historical edge may not survive. It cannot capture position impact in thin books, and it cannot verify that your self-reported entry price was actually available to the size you claim.

What it can do is make your mistakes visible and easier to fix. Research on expert performance emphasises deliberate practice with immediate feedback, which is exactly what a written record provides: a specific decision, a specific result, and a gap to work on. That is the whole point. The traders who improve are not the ones with the best memory; they are the ones with the best records.