The periods that go unlogged are never a random sample.

Three failures, one record

Wrong, revised, and missing are separate problems with separate defences.

A wrong row is a fidelity failure: the entry exists, is internally consistent, and does not correspond to what happened. A revised row is an integrity failure: the entry was changed after the fact to something more comfortable, which is why corrections are logged rather than overwritten. A missing row is neither. Nothing is inaccurate and nothing was altered — a period simply has no representation in the evidence base. The first two have architectural defences. The third does not, and cannot, because there is nothing present for a check to examine. Every integrity mechanism in the system operates on rows — validating them, preserving them, refusing to overwrite them — and a mechanism that operates on rows has no purchase on the absence of one.

Why gaps are not random

Logging decays under exactly the conditions that make it valuable.

If missing periods were scattered arbitrarily, their effect would be a modest loss of precision and nothing more. They are not scattered arbitrarily. Capture lapses when the week was difficult, when the operator is already defensive about results, when a run of losses makes opening the workbook unappealing, or when a stretch involved trades the operator would rather not classify. Nobody stops updating during a clean profitable month. The gaps therefore correlate with the outcome being measured, which turns an incomplete record into a biased one — a different and much worse object. An incomplete record makes conclusions less certain; a biased one makes them wrong while leaving certainty intact, which is the combination most likely to be acted on.

What a biased record does downstream

Every figure moves in the same direction, and the shape stays plausible.

Random error widens uncertainty and leaves central estimates roughly intact. Directional absence shifts them. Expectancy computed without the worst stretches is overstated. Maximum drawdown, being a maximum, is understated by exactly the periods most likely to be missing. Adherence looks better because the trades taken outside the checklist during a bad run were the ones never entered. The benchmark is fitted to a mix that excludes the operator's difficult conditions, so live comparison flatters. Each figure is computed correctly from what it was given, and the whole picture is wrong in a consistent direction.

Why it is nearly undetectable

A gap produces no symptom anywhere in the pipeline.

The system has no independent source against which to notice that a week is missing. Weekly aggregation runs on the rows present and produces a valid summary of them. Gate classification runs on equity, which the broker still reports — so the account may show a drawdown the journal cannot explain, and the drawdown will be routed correctly without anything flagging the inconsistency. Structural diagnostics read a trajectory built from an edited history and describe it accurately. Nothing errors, nothing blanks, and no reading carries a mark saying which periods it was computed without.

• Gaps correlate with bad outcomes, which makes them directional rather than noisy.
• Maximum drawdown is understated by precisely the periods most likely to be absent.
• No downstream stage can detect a period that was never entered.

The only workable defence

Cheap completeness beats expensive accuracy on the hard weeks.

Because there is no architectural remedy, the defence is procedural and unglamorous: a period is entered before it is analysed, and a difficult week is entered first rather than last. Where full capture is genuinely not possible during a bad stretch, a minimal entry — trade, branch, risk, outcome — is enormously more valuable than a gap, because it preserves the period's presence in every aggregate even if the excursion detail is thin. A record that is complete and partly coarse supports honest conclusions. A record that is detailed and selectively absent does not.

The key idea

Coverage is a stronger requirement than precision, and it is the one that lapses.

Operators tend to treat record-keeping as a question of how carefully each entry is made, when the binding property is whether entries exist for every period. A slightly rough record of every month, including the ones that hurt, produces trustworthy expectancy, honest drawdown, and a benchmark fitted to the real operation. A meticulous record with three months quietly absent produces confident figures that describe a version of the account that never traded — and the operator has no way to see which one they are holding.