Read the source layers before the dashboards. Always that order.

The rule

Audit truth layers before their mirrors.

The QA doctrine states it as a workflow step and it applies just as directly to the Saturday pass: inspect the Journal, Weekly Summary, EV Scorecard, and gate state before inspecting anything that displays them. The reason is mechanical rather than stylistic. A dashboard is a rendering function — it takes upstream values and presents them well, and it has no capacity to notice that a value is wrong. An error introduced in the log propagates upward through the rollup into the dashboard, arriving in the operator's attention as a clean, formatted, authoritative figure with no visible trace of where it came from. Downstream display tabs mask upstream errors by design, because presenting things cleanly is what they are for.

Why the instinct runs the wrong way

The dashboard is where the interesting numbers are.

The pull toward reading top-down is strong and understandable. The dashboard holds the summary figures, the grades, the things that answer the question the operator actually has. The log holds forty rows of typed fields. Starting at the dashboard feels like starting with the answer and descending only if something looks wrong — which would be a reasonable protocol if a wrong value looked wrong. It does not. A capture efficiency of 0.61 computed from a mis-entered MFE renders identically to one computed correctly, and the only place the error is visible as an error is the row where it was introduced.

• An error is most detectable at the point of entry and least detectable at the point of display.
• Descending only when something looks wrong means never descending when the error is plausible.
• The pips-translation and data QA stamps exist to catch what a source-layer read would catch — use both.

The typed-value problem

The efficiency lab is deliberately isolated, which puts the burden on entry.

This discipline matters more in the efficiency lab than almost anywhere else, because the lab is intentionally not wired live into CP3 or the broker. Values are typed, pasted, or transferred by hand from trade records. The isolation is a genuine benefit — it keeps experimental interpretation from contaminating production risk authority — and it comes with a cost that has to be paid at review time: a hand-transferred value has no upstream guarantee behind it. In a linked system, a wrong figure usually implies a formula fault. In an isolated lab, it usually implies a transcription fault, which is both more likely and easier to fix, provided anyone looks at the row.

In practice

Three passes, in a fixed order.

The Saturday sequence is short. First, the source pass: open the trade log, confirm every row is complete, and read the data QA and pips-translation stamps before reading a single efficiency number. Second, the aggregation pass: check the weekly rollup's sample and outcome group, confirming the trade count supports a grade at all. Only then, third, the reading pass: the efficiency counts, the quality flags, the week grade. The order costs a few minutes and it means that every number read in the third pass has already been established as trustworthy — which is the difference between reviewing a week and reviewing a rendering of one.

The key idea

Polish is not evidence, and the most polished number is the least audited.

Every layer between the trade and the dashboard makes the number cleaner and further from the fact. Reading against that gradient is uncomfortable because it means starting with the least interesting material, and it is the only order in which an error has anywhere to be caught.