Product Health
What Product Health shows in ErrorDock — Feature health, a per-feature table of adoption, errors and latency, and Priorities, one ranked list that explains every row.
Product Health is two views rather than one number. Feature health shows how each feature is doing. Priorities ranks what to look at first. Neither hides its inputs, because an unexplained composite is not something an engineering team should be asked to trust.
There is no single product-wide score. A number that averages checkout with a settings page tells you less than the two rows do on their own.
Feature health
Feature health, in the dashboard under Performance, is one table with a row per feature:
| Column | What it shows |
|---|---|
| Adoption | Feature users over active users in the window |
| Uses | How often the feature was used |
| Users | Distinct users of the feature |
| Error rate | How often the feature's work failed. It needs withFeature, which records the outcome as well as the use. |
| P95 latency | The time under which 95 percent of the feature's work completed |
Every column shows its change against the previous window, so a feature that got worse after a release stands out from one that has always been slow.
Priorities
Priorities ranks three kinds of problem in one list: issues, latency regressions and underused features.
Each row explains its position. For every input it shows the raw number, the normalised share, the weight and the points that input adds. The weights are versioned, so a ranking from last month stays explainable after the weights change.
Priorities is a screen, not an alert. It is for the moment you sit down to choose what to work on; it does not page anybody.
Worked example
Checkout has 72.4% adoption, a 4.8% error rate and a P95 of 2.8 seconds, up from 1.5 seconds in the previous window. A background export job has a 31% error rate and 40 monthly users.
Sorted by event count, the export job wins. On Feature health the two sit side by side, and checkout is the row that matters: most of your users depend on it, and both its error rate and its latency moved. On Priorities, each row shows the inputs behind its position, so the order can be checked rather than trusted.
What it deliberately excludes
Alert volume, log volume and deployment frequency. All three measure activity rather than whether the product works for the people using it.
Related
Last updated October 20, 2026