The Playbook, Audited

The audit

An audit is a comparison between two versions of the same book, and it takes three columns to state. The first is the number of trades in each group, the second is the win rate, and the third is the average win and average loss in R. From those, expectancy follows for each group, and multiplying by the trade count gives the R each group contributed. In the worked year the process trades earn 0.344R each and produce 61.92R, while forty-five impulse trades at −0.24R a trade cost 10.8R. The year nets 51.12R, which is a good year. The number that matters is the counterfactual. If all 225 trades had been taken at the process’s expectancy, the year would have produced 77.4R, so the impulses cost 26.28R — a fifth of the book carrying about a third of the value. There are two reasons this number is more useful than any advice about discipline. The first is that it is measured rather than argued: it comes from the trader’s own record, so it cannot be dismissed as somebody else’s study. The second is that it identifies a consistent property rather than a random one. The impulse trades did not lose by accident; they lost 0.24R a trade, which means a rule that removes them is worth more than any improvement to the process itself — and that is a statement about where the next hour of work should go. The audit also settles the question the whole subject raises. Every lesson before this one describes a bias that costs money, and each of them sounds plausible on its own; none of them proves that the aggregate is material for a particular trader. The gap between the two groups is that proof, in R, and it is the only version of the argument a trader cannot talk themselves out of. A process whose value is 26R a year is worth the fifteen minutes a week it costs to run, and a process whose value is close to zero is worth changing — both conclusions are available from the same four numbers, and both are honest. One year, in R — Process: 180 trades, 48% win, +1.8R / −1R: 0.48 × 1.8 − 0.52 × 1 = +0.344R a trade → 61.92R ← · Impulse: 45 trades, 40% win, +1.2R / −1.2R: 0.40 × 1.2 − 0.60 × 1.2 = −0.24R a trade → −10.8R ← · The year: 51.12R · All 225 trades at the process expectancy: 77.4R · What the impulses cost: 26.28R — a fifth of the trades, a third of the value A win rate alone would have hidden all of this: 48% against 40% looks like a modest difference, and the expectancies differ by more than a quarter of an R a trade, which over two hundred trades is most of a year.

Making the process easier to follow than to abandon

The audit produces a direction, and the direction is not “be more disciplined”. The impulse trades are not random noise; they have a consistent negative expectancy, which means the work is to make the process easier to follow than to abandon rather than to make the trader sturdier. That is a design problem, and the design levers are the ones the previous lessons established: fewer decisions taken in the moment, larger and pre-planned positions, a checklist that must be completed before an order, a daily loss limit that ends the session, and a journal that tags every trade so the next audit is possible. Two properties of that list are worth noting. Every item on it removes a decision rather than adding a rule, which is why the set can be sustained — a trader who has to remember six prohibitions will break them, while a trader whose orders all pass through the same four fields has nothing to remember. And every item makes the deviation visible: a checklist that was not completed, a position with no written stop, a session past its loss limit. Visibility is what converts an intention into a record, and the record is what the audit consumes. The honest ending is that the numbers will look different for every trader, and sometimes they will not support the work at all. A book where the tagged groups have similar expectancies is one where the process is not the problem, and the correct conclusion is to change the process rather than to double the resolve. That is the value of the audit being arithmetic: it can return the answer that the subject is over, which no amount of introspection can. • Fewer decisions taken in the moment; the plan holds them. • Larger and pre-planned positions, rather than more frequent ones. • A checklist completed before an order, and a daily loss limit that ends the session. • A journal that tags every trade, so the next audit has data. • Be ready for the answer that the process is not the problem. The audit is only as good as the tagging, and tagging is vulnerable to the same bias as everything else: a trade that worked can be remembered as planned. The tag has to be recorded at the time, which is the reason the previous lessons insisted on writing the plan before the outcome.

Pruning the playbook, and the rule against growing it

An audit produces two kinds of finding and they call for opposite actions. A setup with positive expectancy over a workable sample stays. A setup with negative expectancy over a workable sample has three possible verdicts, and choosing among them is the work: the **idea** is broken (the edge was real and is gone), the **execution** is broken (the idea works but you are taking it at the wrong times, sizes or exits), or the **sample** is too small to say (in which case it stays under observation and does not get to use full size). Distinguishing the three means reading the trades, not just the group’s R. A setup that is profitable on the days you took it as written and unprofitable on the days you improvised has an execution problem, and killing it would throw away the edge. The discipline that keeps a playbook honest is to write **kill criteria** in advance, in the same way the risk lessons write stop rules in advance. “I will stop trading this setup if it loses more than X R across N trades, or if the conditions it needs stop appearing.” Written beforehand, the rule can be enforced at the bottom of a drawdown, which is exactly when the judgement is worst. Written afterwards, it becomes an excuse: either the sample is declared too small when the losses are inconvenient, or the setup is discarded after a run of noise that said nothing about its edge. There is a failure mode on the other side that is just as damaging and less discussed: **growing the playbook after every loss**. Adding a filter the day after a bad trade feels like learning and is usually self-curve-fitting — you are fitting rules to a sample of one, and each addition narrows the setup until it describes the trades that already happened. The test is whether a proposed new rule would have been discoverable from the record *before* the trade that prompted it. If it only exists because of that trade, it is not a rule; it is a scar. A playbook should get smaller over time, not larger, as setups are disproved and duplicate each other. • Negative expectancy has three verdicts: broken idea, broken execution, or too small a sample. • Read the trades, not just the group’s R — improvised versions of a good setup hide inside a losing group. • Write kill criteria in advance so they can be enforced at the bottom of the drawdown. • A rule added after one bad trade is a scar, not a rule — the playbook should shrink, not grow. Deleting a setup is not the same as never trading its conditions again. If a group is retired because execution is the problem, the correct next step is a tighter written definition and a smaller size while the new version earns its own sample.

The trades you did not take: keeping a shadow book

The audit in this lesson measures the trades that happened — the process trades against the impulse trades — and it is silent about the third category, which is everything the process produced and the trader declined. That category is where selectivity lives, and selectivity is a behaviour with a sign that nobody knows until it is measured. Every passed-up setup was, in principle, a trade with the process expectancy. If the reason for declining is one of the written filters — the frame disagreed, the level had been tested too recently, the heat cap was full — then passing is the process working, and the shadow book should show the declined trades earning less than the taken ones. If the reason is that the setup did not feel right, or the last one at that level lost, then the shadow book is measuring the cost of discretion: a set of trades that were correctly signalled and skipped for reasons that are not in the document. Keeping the book is cheap because the signals are already generated. Each time a setup appears and is not taken, one line is recorded: the instrument, the trigger, the stop, the written reason for declining, and the outcome that would have followed. Three columns come out of a few dozen lines — count, expectancy in R, and the distribution of the stated reasons — and the reasons column is the one that changes behaviour, because it separates the mechanical filters from the feelings. A trader who discovers that half their declined trades were skipped for reasons that are not in the plan has located a leak of the same kind and the same size as the impulse trades the audit already prices, and it is a leak that the main audit cannot see by construction, since the trades never entered the record. There is a second use that makes the book worth the effort even for a disciplined trader: it measures the **opportunity cost of the filters**. A rule that removes a category of setups should be evaluated on what it removes as well as what it avoids, and without a shadow book the removed trades leave no trace. That is the practical answer to the worry in the previous read — that a playbook which only shrinks will eventually prune away its own edge. With the shadow book, a removed rule can be re-argued with evidence: here are the twenty trades it would have generated, here is their expectancy, and here is the reason it was removed. A playbook maintained this way tends toward the rule set that survives its own review rather than the rule set that survived the last bad week. • Record every signalled setup you decline, with the reason in writing. • Compare the shadow book’s expectancy with the taken trades’ — the difference is the cost or benefit of selectivity. • Separate the mechanical reasons from the feelings: only the first is the process. • Use it to evaluate what a filter removes, so pruning a rule is evidence-based. • A few dozen lines is enough to see the pattern; the count grows with the sample. The book also settles an argument every trader has with themselves after a good move they sat out: whether declining was discipline or fear. The shadow book does not care about the intention. It records the outcome, and over a sample it answers the question with a number rather than with a story about the one that got away.

Sources

• Expectancy, R multiples and process auditingVan Tharp, “Trade Your Way to Financial Freedom”
• Deliberate practice and the feedback loopEricsson, “Peak”
• Reviewing decisions rather than outcomesKahneman, “Thinking, Fast and Slow”