A crypto trading journal should preserve what you knew before a decision, what you authorized, what actually happened and what remains uncertain. Review those records separately from profit and loss. A profitable rule violation remains a violation, while a loss within a correctly followed plan does not by itself prove the plan was mistaken or effective.
The journal's immediate job is to make decisions auditable. This article offers an original two-pass review and a four-decision worksheet. It does not claim that journaling produces an investment edge or that the example scores measure trading skill.
Write the first record before the outcome
Capture the exact venue, instrument, environment, observation time, data sources and hypothesis before an account action. Include the condition for acting, reasons to reject the idea, the intended order and the assumptions behind modeled loss.
Preserve what was unavailable. If a news claim had not been confirmed, write that explicitly. If a candle was still forming, label it. If an indicator's method was unknown, do not improve the old note later by adding a calculation you only discovered afterward.
The AI decision-card walkthrough supplies a detailed pretrade example. A journal extends that card through authorization, execution and later review. It should not become a polished explanation written only after the result is known.
Outcome bias describes evaluating a decision differently because of its result. Baron and Hershey's original experiments examined that distinction in judgment tasks, not crypto trading returns. They motivate separating a process review from an outcome review without proving that this journal method improves performance. Original outcome-bias research.
Keep four kinds of evidence apart
| Record | What belongs there | What it cannot replace |
|---|---|---|
| Decision | Inputs, hypothesis, eligibility and rejection conditions | An actual order receipt |
| Authorization | Account, mode, side, amount and approved rule version | Exchange acceptance or fills |
| Execution | Order identifiers, statuses, cumulative fills, fees and remaining orders | The predecision rationale |
| Review | Corrections, process findings and future test proposals | The original historical record |
Version the records or append dated corrections. Do not silently rewrite an earlier thesis so it appears to have predicted the outcome. A screenshot can support a record, but it needs a time, source and context; a chart captured after an event may contain information absent at entry.
A useful journal can be a simple document or spreadsheet. Its value depends on the evidence retained, not its visual complexity. Avoid storing API secrets, passwords, private keys or unnecessary personal records alongside educational notes.
Four fictional decisions, one misleading profit total
Assume a constructed review period with a 50 USDT modeled-loss policy for each eligible decision. All values below are invented and net outcomes already include the example's assumed fees. The 50 USDT amount is a teaching convention, not recommended risk and not a guaranteed maximum loss.
| Decision | What happened | Net outcome | Process finding |
|---|---|---|---|
| A | Eligible idea, authorized size, verified final fills | −40 USDT | Stated process followed within this example |
| B | Position enlarged beyond the policy; modeled loss became 150 | +80 USDT | Risk-policy violation despite a gain |
| C | No action because a news claim remained unverified | No realized trade | Rejection condition followed; no booked profit or loss |
| D | Submitted order timed out; account state not yet reconciled | Unknown | Incomplete execution record; do not score as zero |
Known realized outcomes from A and B sum to +40 USDT. That positive total does not erase B's violation or establish the strategy's future quality. D could change the account result after reconciliation, and C has no realized trade to add.
Decision A's −40 divided by its 50 modeled-loss reference is −0.8R under this worksheet's definition. B's +80 divided by the same nominal 50 would show +1.6R, but that headline hides the fact that its modeled exposure was tripled to 150. Keep the actual sizing and policy violation visible instead of comparing the ratios as though the decisions used equivalent risk.
An R multiple is only as meaningful as the chosen denominator. State whether R means planned price loss, a cost-adjusted modeled loss or another measure. None guarantees that the actual loss could not exceed it.
First pass: review the process with the result hidden
Where practical, review the predecision and authorization records before revealing the financial outcome. This is a proposed review technique, not a proven cure for bias. Use concrete fields rather than an intuitive score of one to ten.
Check evidence readiness, rule eligibility, sizing and authorization. For each field, mark verified, violated or unknown and point to the supporting record. “Unknown” should not be awarded half credit; it identifies evidence that still needs recovery.
For A, the worksheet can verify its own process fields while remaining agnostic about future strategy quality. For B, the sizing violation is visible before the +80 appears. For C, the unverified claim is enough to evaluate whether the stated rejection rule was followed, without imagining what a trade might have earned.
A preregistered replication of an outcome-bias experiment also found that outcomes affected decision evaluations in its medical-scenario setting. It does not establish a trading intervention or justify applying its measured effects to investors. Aiyer and colleagues' replication paper.
Second pass: reconcile the outcome before explaining it
Add actual exchange records: executed quantity, average price, fee currency, realized proceeds, open quantity and remaining orders. Match them to the intended decision. An application label such as “submitted” is not a completed fill record.
For D, inspect the original order identity and cumulative execution before another request. The timeout and reconciliation guide explains why a missing response cannot be recorded as “nothing happened.” Keep the journal item unresolved until the account evidence supports a state.
When comparing planned and actual prices, use consistent units and explain fee conversion. A fee paid in an asset needs a stated valuation method if the review reports quote-currency net results. Keep deposits, withdrawals and unrelated holdings separate from the specific trade's outcome.
Then describe the gap between plan and execution. Was the intended order never accepted, only partially filled, executed at another price or manually changed? Do not automatically attribute every difference to market skill or model quality.
Record no-trade decisions without inventing profits
A rejected idea is useful evidence about the process, particularly when the system was expected to abstain on missing inputs. It is not a profitable trade merely because the market later moved unfavorably. Likewise, a missed favorable move does not by itself prove the rejection was wrong.
If you study counterfactual outcomes, define a separate simulation with explicit entry, fill, exit and cost assumptions. Label it simulated. Do not mix its hypothetical gains with realized account outcomes. Otherwise a journal can accumulate imaginary profits whenever restraint happens to look wise afterward.
The news verification ledger provides a concrete reason for abstaining while a claim remains unresolved. The data freshness guide provides another: a connection may be alive while the required observation is not ready.
Turn recurring findings into one testable change
At review time, group findings by their evidence, not only their profitability. Repeated missing timestamps suggest a data-recording problem. Repeated unauthorized size changes suggest an authorization problem. Repeated unknown orders suggest a reconciliation problem. They call for different fixes.
Choose a bounded next change and define how it will be checked. For example: require a source timestamp field before an idea becomes review-ready, then verify it on the next predetermined sample of notes. Do not claim that such a process fix improves returns unless a suitable study establishes that separately.
Avoid redesigning the strategy after every loss. Each new threshold, filter or parameter is another research choice. Preserve variants and evaluation periods so the next backtest review can distinguish discovery from untouched assessment.
If a journal stores model confidence, retain its exact definition and version. The confidence-score audit explains why a review score and a probability of profit cannot be pooled simply because both use percentages.
Make the next reviewer able to disagree
A useful final note states the finding, supporting evidence, unresolved alternative explanations, proposed action and verification criterion. “Be more disciplined” is hard to inspect. “The second order exceeded the saved 50-USDT modeled-loss policy; inspect why the authorization check permitted the change” names a problem someone can investigate.
Keep incident material from gaining authority merely because it is stored. A pasted research page or tool response remains untrusted evidence in a journal. The prompt-injection guide explains why old notes must not silently become new execution instructions.
Use the AI bot evaluation checklist when a recurring limitation concerns the selected system rather than one trade. KLINEO's AI analysis supports review, but this article does not assert a built-in journal feature or certify a trading process.
Prepared with AI assistance under the editorial policy, this worksheet separates original hypothetical examples from cited behavioral research. A complete journal can expose missing evidence and inconsistent actions. It cannot guarantee a profitable strategy, prevent every bias or make crypto losses acceptable for a reader's circumstances.