What is a Data Confidence Score?
A Data Confidence Score is a single 0–100 measure of how far a project board can be trusted as a basis for decisions. It is not a measure of how the work is going — it measures whether the record of the work is good enough to say how the work is going.
Every status report inherits the quality of the board beneath it
Dashboards are built on the assumption that the board is an accurate record. That assumption is rarely tested. Items lose owners, tickets sit untouched for a month, a dependency points at work that was closed in March, and the same feature is recorded three ways by three people.
None of that produces an error. The burndown still renders, the velocity chart still has a number in it, and the weekly report still goes out. The failure mode is silent: reporting stays confidently wrong until a date slips and everyone asks how it was missed.
A Data Confidence Score makes that risk visible before the miss, by scoring the record itself rather than the work.
How the score is calculated
A board starts at 100 and loses points for each category of problem found on it. The weights are fixed and deliberately explainable — a project manager should be able to read why a board scored what it did without a data team translating it.
What makes the model readable is that a category costs points in proportion to how widespread the problem is. One badly-linked item on a 500-item board barely moves the number; missing owners across half the board move it a lot. That is why the score tracks whether a board can be trusted overall rather than reacting to isolated mistakes.
A category can never cost more than its weight, so the worst possible board scores zero rather than going negative, and no single category can dominate the result.
| Dimension | Weight | What it asks |
|---|---|---|
| Workflow | 30 | Do item states reflect what is really happening? Items marked Done with no resolution recorded, work sitting in review with nobody assigned, tickets reopened without an explanation, and work stuck in progress far longer than it should be. Each one quietly corrupts the cycle-time and throughput figures derived from it — the chart still renders, it just no longer means what it claims. |
| Ownership | 25 | Does someone own the work? Items with no assignee, in-progress work with no due date, tickets with no priority set, and items whose description is too thin to act on. They still appear in every count and every burndown, contributing to the total while telling you nothing — which is how a board can look full and still be unreportable. |
| Dependencies | 15 | Do links between items still hold? Blocked-by links whose blocker has already been closed, and links whose blocker has not moved in more than two weeks. Both hide the real critical path: the first leaves work marked blocked when nothing is blocking it, the second leaves a dependency everyone assumes is progressing. |
| Staleness | 15 | Has anything actually moved recently? Items nobody has updated in weeks, work already past its due date, and blocked items with no follow-up. Until someone moves them they all read as in progress, which is a common reason a status report and reality disagree. |
| Consistency | 10 | Is the same thing recorded the same way? Labels applied with inconsistent casing, statuses outside the standard set, and summaries too short or containing characters that break search. Nothing looks broken — filters and reports simply miss whatever was recorded the other way. |
| Duplicates | 5 | Is the same work counted more than once? Items whose summaries are near-identical to another on the board, flagged above an 85% similarity threshold. Duplicates inflate totals and make progress look slower than it is. Weighted lightest because it distorts volume rather than direction. |
Weights sum to 100. A board starts at 100 and each category deducts in proportion to how much of the board it affects, up to its full weight. A problem touching most of the board costs nearly all of that category's points; one touching a handful of items costs very little. The deductions are summed and subtracted from 100, and no category can take more than its weight.
How to read the number
The bands exist to answer one question: can this board be used to make a decision today, or does it need cleaning first?
| Grade | Score | What it means |
|---|---|---|
| A | 90–100 | High confidence. Reporting is reliable. |
| B | 75–89 | Good. Some issues; the dashboard is mostly trustworthy. |
| C | 60–74 | Medium. Risk of misleading reporting. |
| F | Below 60 | Low confidence. Do not rely on status without cleanup first. |
The practical threshold is 75. Above it, a board can carry a status conversation. Below 60, the honest position is that the board cannot currently support a claim about delivery — and saying so early is considerably cheaper than discovering it at a deadline.
Why these weights
Workflow carries the most weight because its failures corrupt time-based metrics — the figures leadership most often acts on. Cycle time, throughput and burndown are all derived from how items move; when Done items carry no resolution or work sits stuck in progress, every number derived from them is wrong with total confidence.
Ownership is next because a missing assignee cannot be inferred. If nobody owns an item, no report built on it can say who is accountable, and no amount of downstream analysis recovers that.
Dependencies and staleness sit in the middle. Both are weighted lower than workflow, but dependencies are frequently the category that causes the visible failure, since a stale blocker conceals a real critical path.
Consistency and duplicates are weighted lightest deliberately. They distort magnitude rather than direction: they make reports imprecise, not misleading.
What to do with a low score
A low score is only useful if it points at the specific items dragging it down. Scoring a board and returning a bare number moves the problem rather than solving it.
So alongside the headline number ValidFlow returns the category breakdown, the drivers behind it — each with the items responsible and how many there are — and a set of suggested quick wins. The output is a worklist, not an instruction to improve data quality.
Where it applies
The score is platform-agnostic by design, because the failure modes are. A stale item in Jira, a Monday.com board with empty owner columns and a Trello list nobody has touched since the quarter began are the same problem wearing different labels.
Today ValidFlow scores boards on Jira and Monday.com, and analyses repository activity on GitHub and GitLab. Support for Asana, Trello, ClickUp and Notion is on the way.