A weekly review is most useful when it can end with “no change.” Its job is to check what the process recorded, identify missing evidence and set one bounded next question. It is not a deadline to place a trade or produce a new rule.
Start with the record that existed at the time
Before interpreting a result, collect the dated plan version, market and instrument, observation time, signal inputs, intended size, order type, modeled or actual fill, fees, funding if relevant, exit and any exception. Preserve the original note. Add a correction as a new dated entry so a later explanation cannot overwrite what was known earlier.
Include decisions that did not become trades: a setup rejected by the rule, a cancelled order, a skipped observation, a missing data point or an intentional stay-flat period. A list containing only closed winners and losers cannot show where the decision process changed.
A 20-minute review sequence
- Freeze the window. Record the start and end timestamps, strategy version and data source. Do not shift the dates after seeing which interval looks favorable.
- Reconcile the ledger. Match each decision to a modeled or venue record. Mark missing fills, fees or funding as unknown rather than filling gaps with zero.
- Classify process and outcome separately. Label whether the written rule was followed, then record the outcome in a different field. A favorable result does not repair an unplanned decision; an unfavorable result does not by itself invalidate a correctly followed rule.
- Name one unresolved question. Examples: “Were two modeled fills using the same timing convention?” or “Did the stated fee assumption match the dated venue schedule?” State what evidence would answer it.
- Choose one next state. Continue collecting the same evidence, pause because a data or risk limit was reached, or create a separately versioned hypothesis. Record the next review date and keep the prior version intact.
The time estimate is a suggested meeting format, not a validated performance or behavioral intervention. A review can take longer when the ledger is incomplete; do not force a conclusion to fit the calendar. Use the same date range and inclusion rule when comparing periods. If a record is late, corrected or missing, keep that status visible instead of silently treating the sample as complete.
A fictional review example
Suppose a paper log contains 12 planned observations: eight rule-qualified decisions, two signals that failed a written data-quality check, one skipped observation with a missing candle and one decision made outside the plan. These counts are invented bookkeeping, not a strategy result. Before calculating any rate, define the denominator: planned opportunities, signals that passed a data check, orders submitted or completed trades answer different questions. Do not remove a skipped or inconvenient observation after seeing its outcome. The useful review question is whether the four nonstandard cases are traceable and whether the missing candle changes the evaluation window. The next action might be to repair provenance or keep collecting the same rule; the count alone does not justify changing the entry condition.
| Review field | What to keep |
|---|---|
| Process | Rule version, exception reason, planned versus actual action |
| Market record | Instrument, timestamp, source, missing or corrected observations |
| Execution record | Simulated versus actual fill, quantity, fee and funding evidence |
| Interpretation | Sample size, unresolved assumptions, contrary cases |
| Next step | One question, one owner, a date and the condition for revisiting |
Consider two more fictional rows. In the first, a completed-bar rule qualifies, a one-unit position is modeled at $100, and a later stop exit at $98 records a $2 gross loss before fees. If the written entry, size and exit rules were followed, classify the process as followed and the result as adverse; one loss does not automatically justify changing the signal. In the second, an unplanned entry after a price jump exits at $102 from a $100 entry. Record it as a favorable exception, not as a newly validated rule. The apparent gain does not tell you whether that exception is repeatable. If its fee is unavailable, leave net result unknown rather than treating it as zero.
Keep the process field separate from the outcome field. If enough comparable records accumulate, compare only like-for-like strategy versions, instruments, observation windows and fill conventions, and show excluded and unresolved records alongside any summary. A dozen observations can help find a missing column; it cannot establish a stable personal or strategy-wide rate. The backtest-overfitting literature addresses research selection across tested strategies, not whether a weekly review habit improves trading decisions.
Do not turn review into endless tuning
Repeatedly trying new thresholds against the same history makes the selected result depend on how many alternatives were considered. Bailey and colleagues describe a framework for estimating the probability of backtest overfitting; it is a research-selection method, not evidence that weekly journaling prevents overfitting. If a review changes the strategy, label the new version and reserve a later evaluation period rather than repeatedly reusing the same holdout.
SuperQuant’s journal and review features preserve user-entered records. Its bundled histories are synthetic and its paper exercises modeled; they are not records of live exchange execution. Use the journal guide, journal template, trading-plan template and backtest audit checklist to keep assumptions visible. This is a process aid, not financial advice or proof of an edge.