The Journal and the Review Loop

What to record, and when

A journal entry has two halves and they are recorded at different times. Before the trade, the five plan fields from TR19: setup, entry trigger, stop, target, size — plus the frame, the regime from TR14, and the location from TR13, because those are the conditions that determine whether the setup has an edge at all. After the trade, three things: the outcome in R, whether the plan was followed exactly, and the exit reason. Recording the first half first is what makes the second half evidence rather than narrative: an entry written after the outcome is a story about the past, and it will quietly rationalise every deviation. The tags are where the value concentrates, because they are what allow the record to be sliced. Execution: did the trade follow the plan, or was it impulse, chasing, or a stop moved? Regime: was the instrument trending or ranging? Location: at a graded level, or mid-range? With those three columns, a hundred trades stop being a single number and become several: the expectancy of planned trades in trends at levels, against the expectancy of everything else. That comparison is the audit of the whole subject so far, and it is routinely uncomfortable — which is why the tags have to be recorded honestly and in advance of knowing which way they will help. One entry per trade, one line per statistical summary, and a review on a schedule rather than on a mood. The reason is mechanical: expectancy takes dozens of observations to separate from noise, so a learner who reviews after every loss is reading variance and calling it feedback. Monthly is a workable cadence for a retail trader, with a rule that no parameter changes inside the month and no conclusions drawn from fewer than twenty trades. The two halves of one entry — Before: setup, entry, stop, target, size, frame, regime, location: Recorded while the outcome is unknown · After: outcome in R, plan followed yes/no, exit reason: Recorded once, not edited later ← · Monthly summary: expectancy, average win, average loss, worst run, profit factor: Computed, not felt · Slices: planned vs impulse · trend vs range · at a level vs mid-range: Where the diagnosis appears ← Screen-record or timestamp the entry if the habit is hard to keep. A journal written from memory at the end of the week is a summary of how the week felt, and it will systematically over-report good process and under-report the trades that were never planned.

Reading the record, and changing one thing

The statistics worth computing each month are few. Expectancy per trade, in R. The average win and the average loss, which together give the payoff ratio and therefore the break-even win rate: one over one plus the payoff. The worst losing run, which tells you what the size has to survive rather than what the last month achieved. The profit factor, gross R won over gross R lost. And the count of trades, because a small sample has no verdict in it. Those five numbers are enough to answer the question that matters: is this setup being paid for, and at what size can it be traded? Then the diagnosis, which is the part that requires discipline. If expectancy is positive but the variance is causing observable misbehaviour — skipped trades, size changes, stops moved — the record will show it in the execution tag, and the answer is a smaller size rather than a better analysis. If expectancy is negative and the regime tag shows the setup firing mostly in ranges, the fix is the filter from TR14 and nothing else. If expectancy is positive on planned trades and negative on impulse ones, the fix is the process discipline from the psychology subject. Each of those is one change and one variable, which is why the rule is to change one thing at a time: changing three gives an uninterpretable result and invites the conclusion that the change did not work. Finally, the loop closes back at the beginning of the subject. The journal is what tells you whether the candle, the level, the breakout and the plan are being paid for at the sizes you are trading — and the honest outcome for many learners is that one part of the process works and another destroys it. Finding that out from a hundred of your own trades, in R, with the conditions tagged, is the difference between a methodology and an enthusiasm. Everything in this subject from TR1 onwards was a measurement; this lesson is the one that measures the measurements. • Compute five numbers a month: expectancy, average win, average loss, worst run, profit factor. • Slice by execution, regime and location — the diagnosis usually sits in one of the three. • Read the execution tag before adjusting analysis: variance-driven misbehaviour is fixed with size. • Change one parameter at a time so the effect is identifiable. • Set a minimum sample — twenty trades — before any conclusion is drawn. • A journal written after the fact is a summary of how the month felt. The instinct to reduce a journal to a win-rate tracker is the one to resist, because a win rate without the payoff is uninterpretable: a 40% record at a 3:1 payoff is excellent and a 60% record at a 0.5:1 payoff is a slow loss. The statistics exist to be read together, and the payoff ratio is the one most often missing.

What the R column hides

Expressing every trade in multiples of its initial risk is the single best decision in a trading journal, and it carries two assumptions that are worth surfacing. The first is that the trade had **one** risk at entry. A position that is added to, or trimmed on the way up, has several exposures during its life, and recording the outcome as a single R hides which of them produced the result. A winner that was doubled at a better price and a winner that was never touched produce the same number, and they are not the same trade — the first is a decision about management that should be evaluated separately from the entry. The practical fix is to record scale-ins and scale-outs as their own lines with their own risk at the moment they were taken, rather than folding them into the parent trade. The second assumption is that the initial risk stays the right denominator for the whole holding period. A trade held for three months during which volatility doubled had a risk that grew, and a two-R result measured against the original stop is not the same as a two-R result with the stop moved to break-even halfway through. Neither convention is wrong; the point is that the convention has to be applied consistently, because a journal that switches between them is comparing numbers that are not the same quantity. Write the convention at the top of the journal and keep it. The bigger thing the R column hides is **time**. Expectancy per trade says nothing about expectancy per unit of time, and a system with a high expectancy per trade and a ninety-day average holding period is a different proposition from one with the same expectancy per trade that turns over in three days — the second pays far more in costs and gives far more chances for the edge to show. A rolling expectancy computed over the last thirty trades, alongside the average holding period, is the pair of numbers that shows whether the process is drifting or working. That rolling window is what turns a journal from a record of what happened into a control chart for the strategy. Two conventions to fix at the top of any journal: whether a result is measured against the original stop or the current one, and how adds and trims are recorded. Both change the numbers, and neither is wrong — but a record that mixes them cannot be compared with itself.

The two excursions: what your stops should have been

A log records how a trade ended. Two fields also record how it got there, and they turn the two most common feelings in trading — “my stops are too tight” and “I keep giving back my winners” — into measurements with answers. They are the **maximum adverse excursion** and the **maximum favourable excursion**: the worst and the best the position reached while it was held, both recorded in units of initial risk. The first converts the stop question into a distribution. Look at the winners only, and at how far against them the price travelled before they worked. If almost none of them needed more than a third of the stop, the stop is generous and could be tightened, which buys size. If a meaningful share needed nearly the whole stop — say a fifth of the winners touched nine-tenths of it — then the stop is clipping trades that would have worked, and the stop-outs are systematic rather than unlucky. That single histogram answers a question that argument never resolves, and it is the reason the field is worth two seconds at the exit. The second converts the exit question into one. Look at the losers only, and at how far in favour they ran before they turned. Positions that ended at a loss having been up by more than a target at some point are not evidence of bad entries; they are evidence that the exit rule gave back a completed move. The mirror image, winners that never went anywhere, says something about the entry. Splitting the log by these two distributions is how a record tells you which of the two halves of the system is producing the result — which the R column alone cannot, because a trade that was up 2R and closed at zero looks identical to one that never moved. Recording them in R rather than in dollars is what makes them poolable: trades of different price and size become comparable, and the histogram is then over one scale. Partial exits need one convention, and the simple one is to book each leg at its own R and record the excursions of the position as a whole. One caution, and it is the discipline this whole subject applies to parameters. Excursion analysis is a search over stop and target levels, so the number it suggests is an in-sample optimum, and moving a stop to it because the histogram says so is fitting. The honest use is to look for a *range* — a band of stop distances over which the winner distribution barely changes — and to confirm the choice on trades the analysis was not derived from. • Maximum adverse excursion, in R, shows how much room the winners actually needed. • Maximum favourable excursion on losers shows what the exit rule gave back. • Record both in R so trades of different sizes can be pooled into one distribution. • Treat any suggested stop as an in-sample optimum: look for a stable band instead. Two extra columns and one monthly question: what fraction of winners needed more than half the stop, and what fraction of losers were up by more than half a target? Those two numbers describe the system’s binding constraint better than its expectancy does.

Sources

• Deliberate practice and feedback loopsEricsson, Krampe & Tesch-Römer (1993) on deliberate practice
• Decision journals and calibrationAnnie Duke, “Thinking in Bets”; Tetlock, “Expert Political Judgment”
• Expectancy, profit factor and drawdown statisticsVan Tharp, “Trade Your Way to Financial Freedom”