Which of your numbers stopped meaning what they meant

Four metrics examined, four different ways they broke, what replaces each one — and the one judgment no number left can support.

Your metrics did not get worse. They started measuring something else.

Every number you report went up. None of them is computed wrongly. And most of them were never measurements of anything you cared about directly — they were things that were easy to count and that used to move when the thing you cared about moved.

That connection is what broke, and a broken connection does not make any gauge read zero. This volume names which of your numbers are affected, in which of four ways, and what to put in their place.

The spine of the volume: counting something easy in place of something you care about is a chain of two guesses, and the middle of it came apart.

What moves: three gauges rise together, then the link between the second and third becomes a broken dashed line while both outer gauges carry on rising.

Four metrics, four different failures

the metric	how it broke
counting output	the stand-in stopped standing in for effort or for value
cycle time	the stretch being timed moved, and the queue moved outside it
individual records	who did the work stopped being a decidable question
coverage and its relatives	it was only ever meaningful because it was expensive

The four were chosen because each breaks differently, so the set generalizes. By the third step you should be able to apply them to a number this volume never mentions.

Every condemned number here comes with a replacement, or an explicit admission that there is not one. There is exactly one of the latter. It is about judging individual people, and it is the most important page in the volume.

What you can do afterward

Mark every line on your own dashboard as readable, readable-with-a-caveat, or retired. Compute the one number most teams do not, from records you already keep. Walk into a calibration meeting knowing which number cannot help you. And retire a metric upward without it sounding like an excuse.

The four failures, the replacements and the retirement script are on one page you can print.

Part 1 — what moved underneath your numbers

01Why everything on your dashboard improvedafterward you can recognize your own before-and-after in a constructed oneexhibit

02The middle link snapped and both ends kept movingafterward you can explain why no gauge read zero when it brokereasoning

03Four ways a number stops meaning what it meantafterward you can apply the four to a metric this volume never coversreasoning

Part 2 — the four metrics that stopped meaning what they meant

04Why counting output stopped measuring effort or valueafterward you can say what a count of shipped work still tells you, which is lessreasoning

05A falling cycle time can hide a rising backlogafterward you can read an improving cycle time without being reassured by itreasoning

06Identical records can mean opposite amounts of workafterward you can stop using output records to compare two peoplereasoning

07What happens to a target once it is cheap to hitafterward you can spot the metrics that were only ever proxies for effortsourced

Part 3 — what still means what it used to

08Why outcome measures survived the transitionafterward you can name the handful of numbers you can still read literallyreasoning

09The most useful number you probably do not computeafterward you can compute rework on your own team, from records you already keepreasoning

10Individual contribution, and why no number supports itafterward you can go into a calibration meeting knowing which number cannot helpsourced

Part 4 — what to retire, and how to say so

11How to retire a metric upward without losing trustafterward you can have the conversation that stops you reporting a number you distrustreasoning

12Marking a real dashboard, number by numberafterward you can mark every number on your own dashboard in one sittingexhibit

Every term this page uses

proxy: Something easy to count that stands in for something you actually care about and cannot count directly.
denominator: The count a rate was worked out over. How many, not what share.
attribution: Deciding which person or which change produced a result.
cycle time: How long one piece of work takes from being started to being shipped.
rework: Work that had to be revisited after somebody called it done.
outcome measure: Something measured about the result rather than about the effort spent.
gameable: Cheap to satisfy without doing the thing the measure was meant to reward.
lag: The delay between something happening and the number moving in response.
target: A measure somebody is expected to move, rather than one you only watch.

Taken as already known, and so not defined here: agent, metric, dashboard, pull request, coverage. That list is a claim about who is reading, and it is printed so it can be argued with.