How we built the Delivery Confidence Score
Every project starts at 100. Everything after that is subtraction — and each subtraction has to name the work item behind it.
The first version of this score tried to predict. It weighted velocity trends against historical outcomes and produced a number nobody could argue with, because nobody could explain it either. A project manager would see 61 and ask what changed. We could not answer without opening a notebook.
So we threw out prediction. The score you see now is rule-based and additive in one direction: a project starts at 100, and each thing we can actually detect subtracts from it.
What subtracts, and how much
Five factors do most of the work, and each one is capped so a single bad signal cannot swamp the rest — you can see them laid out on the delivery confidence page:
- Overdue open items — scales with the share of open work past its due date, capped at 35.
- Items waiting on dependencies — counts open work whose blocker is itself still open, capped at 25.
- Stale items — open work nothing has touched in three days, capped at 15.
- Unassigned high or critical work — five points each, capped at 15.
- Items stuck in review for more than two days, capped at 10.
Sprint pressure is separate
An active sprint adds two more possibilities. If the end date has passed and work remains, that is a flat 20. If the sprint ends within two days and more than a third of the planned points are still open, the score takes a proportional hit up to 15.
Keeping these separate matters. A project can be perfectly healthy and still be in a sprint that will not land, and those are different conversations with different fixes.
Why it is snapshotted daily
A single number is a bad interface for risk. 58 means nothing on its own; 58 after three weeks in the seventies means something specific. We capture one snapshot per active project per day, so what you read is a trend line rather than a verdict.
That also makes the score falsifiable. If it drops and nothing was actually wrong, the factor list tells you which rule fired and you can go argue with the rule.
What it is not
It is not a prediction, and we are careful not to present it as one. It is a decision aid computed from your own data with rules you can read — a prompt to go look at four specific work items, not a judgement about whether you will ship.
The most useful thing it does is not the number at all. It is that the number always comes with the reason.