A board memo cites a conversion rate. The figure is recorded, minuted, and filed. Two years later, someone asks what that figure actually measured. The number is still there. What it meant is not.

What survives, and what does not

A decision record typically preserves the output of a digital measurement — the figure itself, the date, the conclusion drawn from it. It rarely preserves the structure that gave the figure its meaning: how the metric was calculated, which system produced it, what counted as a qualifying event, and what attribution logic assigned credit to which touchpoint. That structure lives in dashboards, calculated fields, semantic layers, and configuration files — separate from the decision record entirely, and far more likely to change.

Digital measurement environments are not static. Metric definitions are revised. Attribution models are updated. Dashboards are rebuilt. Source systems migrate. Tracking logic changes as platforms evolve and as privacy requirements shift. Each of these changes can alter what a given metric name means without altering the name itself — so that “conversion rate,” reported consistently across years, may not have meant the same thing in each of them.

Why this is a distinct problem, not a data-quality problem

The number in the original decision record is not wrong. It was an accurate figure at the time it was produced. The loss occurs afterward, when the environment that made the figure interpretable — the definition, the source, the calculation logic — changes or disappears, while the figure itself remains quoted in a record that gives no indication anything has changed. The record preserves an artifact. It does not preserve the conditions required to interpret the artifact correctly.

The executive consequence

This becomes material whenever an organization needs to look backward with precision: reconstructing the basis for a past decision under scrutiny, comparing current performance against a historical figure to judge trend, or defending a prior conclusion that is later questioned. In each case, the assumption is that the old number and the new number can be compared directly. If the underlying definition changed in the interval — and there is no institutional memory of the change — the comparison is invalid in a way that is invisible to everyone making it.

The institutional implication

The correction this territory calls for is precise, not expansive: a decision record that cites a digital metric should preserve enough about that metric’s definition and source — not the full technical configuration, but enough context to know it later — for someone to determine, years afterward, what the number actually meant when the decision was made. Without that, an organization retains its numbers but loses its ability to trust what any of them once represented.