Last known value carry-forward is the rule that says when a source has not reported since its previous reading, the previous value continues to apply until it does. It is how slow-moving data gets projected onto a finer time grain without inventing numbers.
Measurement inputs arrive at wildly different rates. An endpoint agent reports several times a day. A vulnerability scan runs weekly. An internal audit closes findings monthly. A third-party assessment lands twice a year. Training completion is exported when someone remembers.
A daily metric that draws on all of them has to decide what the audit figure is on a Tuesday in the middle of a month. Carry-forward says it is whatever it was at the last close, which is both the honest answer and the one that matches how a human would reason about it.
The alternatives are worse. Treating an absent value as zero produces a metric that collapses and recovers on a monthly rhythm that reflects reporting, not reality. Interpolating invents readings that were never observed.
An age. Every carried value should carry the timestamp of the observation it came from, so a reader can tell a fresh reading from a four-month-old one holding steady on a chart.
A staleness limit. Beyond some age, carrying forward stops being reasonable and the value should be marked unknown rather than presented as current. The limit differs per source and belongs in the metric definition.
Visibility in presentation. A flat line from a monthly source and a flat line from a genuinely stable daily source look identical, and mean very different things.
A collector fails silently and the last good value carries forward indefinitely. The metric looks stable. Nothing is being measured at all.
This is the single most common way a measurement programme goes blind without noticing, and the only defence is monitoring the age of the underlying observation rather than the value.
Metric Maestro applies carry-forward as a declared property of the metric rather than a side effect of how the chart was drawn, and stores the observation age alongside the value.
Manual entries behave the same way as automated ones. A quarterly figure typed in with its supporting evidence carries forward with its date attached, so a reader can always see how old the underlying observation is.
From the blog