Ten fields are typed. Four compute. The boundary is not arbitrary.

The rule

A field is typed only if nothing in the system could have known it.

Date, branch, risk, outcome, hit flags, break-even status, fees, MAE, MFE, duration and notes are entered, because each records something that happened in the market or in the operator's execution and exists nowhere the sheet can reach. Risk in dollars, outcome in dollars, fees as a percentage of 1R and fees as a percentage of balance are derived, because each is a function of fields already present. The boundary is drawn at that question and no other — not at what is convenient to type, and not at what is quick to look up.

Why duplication is the failure

A field with two sources has no source at all.

The tempting shortcut is to type a derivable field because the number is already on screen from the broker — the dollar risk, say. The moment it is typed, the sheet holds a value that could also have been computed, and the two will diverge the first time a typo, a balance change, or a mid-week deposit intervenes. Nothing in the system can arbitrate the disagreement, because both are marked as fact. Removing the possibility is cheaper than detecting the consequence, which is why the derived fields have no input cells at all rather than defaults the operator is trusted to leave alone. A protection that depends on someone declining to type into an available cell is not a protection; it is a request.

Verified, not trusted

The computed fields are checked, and that is not a contradiction.

The operating instruction is that auto-calculated fields are verified rather than trusted, which sounds like distrust of the formulas and is not. The formulas are reliable; their inputs are the risk. A derived dollar figure that looks wrong is a signal that a typed field feeding it is wrong — a misplaced decimal in risk percent, a balance not yet updated, a fee entered in the wrong unit. The derived column is therefore functioning as a cheap integrity check on the typed ones, and glancing at it costs seconds while inheriting the error costs the quarter. The instruction is really about direction of suspicion: when a derived value looks wrong, look up at its inputs rather than at the formula.

The hard one

Branch identity is typed, and it must be assigned at entry.

Branch is the most consequential typed field and the one most vulnerable to a specific error: assigning it after the fact, from how the trade behaved. A position that ran is retroactively a trend trade; one that stalled becomes Normal. Every figure computed from that classification is then contaminated in a direction that flatters, because the branches have been defined by their outcomes rather than by their intent. The field is typed because only the operator knows what was intended, and it belongs at entry because that is the only moment the intent exists uncontaminated.

• Typed if only the operator observed it; derived if the sheet can reach it.
• Never both — two sources of truth diverge silently and cannot be arbitrated.
• Branch assigned at entry; inferred afterwards it encodes outcome, not intent.

What the boundary buys

Every wrong number has exactly one place it can have come from.

The practical dividend is diagnostic. Because no figure is duplicated and every derived value traces to named inputs, a wrong number anywhere in the sheet resolves to a specific typed field on a specific row. There is no third possibility — no transcription layer, no manual copy between tabs, no summary that was updated separately. That property is what makes an audit tractable at all: the question is never whether the arithmetic drifted, it is always which entry was wrong, and that is a question with an answer.

The key idea

The discipline is about what the operator is allowed to type.

Most record-keeping advice concerns being careful, which is unenforceable and degrades under pressure exactly when it matters. This is a structural version of the same goal: the operator cannot corrupt a derived field because there is nowhere to type it, and cannot leave a typed field to be inferred later because nothing else knows it. Narrowing what a human is permitted to enter is a more reliable protection than asking them to enter it well. It is also the only version that holds during the weeks when attention is scarce, which are the weeks the record most needs to survive.