Reconcile before you review, or the review audits itself.

The procedure

Three sources, checked against each other, before anything aggregates.

The weekly step compares three independent records of the same week: the Journal rows entered daily, the paper capture sheet written near each event, and the broker or platform statement. Each exists for a reason. The Journal carries branch identity and intent that no external source holds. The capture sheet carries what was observed before recollection reshaped it. The broker carries the fills, fees and durations that are not the operator's to remember. Agreement across all three is what promotes the week from entered to reconciled, and only reconciled weeks feed the rollups. Two sources agreeing is not sufficient, because the two most likely to agree are the two written by the same person from the same recollection.

Why the order is the whole thing

A number already seen stops being checkable.

Read the weekly EV first and the reconciliation that follows is no longer neutral. The operator now knows which direction a correction would move the figure, and every ambiguous row — a fill that could be read two ways, a fee that might belong to the previous week — resolves under that knowledge. Nobody experiences this as dishonesty; it feels like judgement. Performing the check before any aggregate has been seen removes the pressure entirely, because at that point there is no direction to prefer. The sequence is doing work that no amount of integrity substitutes for. This is worth internalising because the natural defence — I would notice myself doing that — is exactly the defence that fails, since the resolution feels like ordinary judgement at every step.

What reconciliation is not

It is not a second chance to improve the week.

Discrepancies are resolved toward the source with the strongest claim on the field, not toward the more agreeable figure. Fills, fees and durations belong to the broker and the broker wins outright. Branch identity and intent belong to the Journal, because no external record holds them. Excursions and context belong to the capture sheet, which was written closest to the event. And corrections are made as logged entries rather than overwrites, so the record shows what changed and why — a reconciled week with three corrections is in better standing than a clean one that was never checked.

The trades that need it most

Ambiguity concentrates in exactly the rows that matter.

Straightforward trades reconcile instantly and contribute little. The rows that consume the time are the ones with partial fills, stop moves mid-trade, weekend carry, or an exit taken in two pieces — and those are disproportionately the trades whose classification and excursion figures carry the most information. A reconciliation that feels quick every week is usually a reconciliation not being performed on the rows that would repay it, which is worth noticing as a signal rather than enjoying as efficiency.

• Fills, fees, durations → broker wins. Branch and intent → Journal wins.
• Corrections are logged entries, never overwrites of the original row.
• A week reconciled with three corrections outranks an unchecked clean one.

When the week was bad

The step most likely to be skipped is the one on the most valuable data.

Reconciliation is a scheduled obligation with no visible reward, performed at the end of a week, and it competes directly with the desire to not look at a bad week at all. That makes its failure rate correlated with the informativeness of the data, which is the worst possible property for an audit step. The practical mitigation is to treat reconciliation as separable from review: the week can be reconciled on Saturday and read on Sunday. Splitting them preserves the ordering while removing the excuse, because the unappealing part is the reading rather than the checking. A week reconciled on Saturday and read on Sunday satisfies the ordering completely, and the split costs nothing structurally.

The key idea

This is the step that makes the Journal ground truth instead of a claim.

Every layer above CP3's first treats the Journal as authoritative and none of them can verify it. That authority is not a property the sheet possesses — it is manufactured once a week by a procedure, and it lapses the moment the procedure does. An unreconciled Journal is still a record, still renders, and still feeds every engine above it. What it no longer is, is evidence, and nothing downstream announces the difference.