A T of 80 and an index of 80 are opposite facts.
On an ability measure, high is good. On a symptom or rating scale, high is the problem. Convert both to z and average them as they stand, and a severe executive complaint quietly cancels a strong reasoning score — leaving a domain mean near zero and a profile that looks unremarkable.
The failure is silent, which is what makes it dangerous
Nothing errors. No value is out of range. You get a number, the number is plausible, and it is the average of a strength and a deficit pointed in opposite directions. On a single case a reader might catch it. In a cohort mean across forty people it is invisible, and it biases the answer toward “nothing here” — the direction that loses a finding rather than inventing one.
The fix: direction is a property of the measure
Every measure in Casebook’s catalog carries a direction, and the sign is flipped on the way in so that negative always means worse.
| Score | Direction | Raw z | Stored |
|---|---|---|---|
| WISC-V index 70 | Higher is better | −2.0 | −2.0 |
| BRIEF-2 T of 70 | Higher is worse | +2.0 | −2.0 |
| WISC-V index 130 | Higher is better | +2.0 | +2.0 |
| BRIEF-2 T of 30 | Higher is worse | −2.0 | +2.0 |
After which ability and symptom scales can sit in the same domain mean and point the same way. A BRIEF-2 T of 70 and a WISC-V index of 70 both read as −2, which is what a reader means when they say both are two standard deviations down.
Why this has to be catalog metadata, not a rule
There is no reliable way to work direction out from the number. T scores run both ways depending on the instrument — some rating scales report adaptive skills where high is good, and sit alongside problem scales where high is bad, in the same form, in the same metric. The only thing that knows is the instrument.
So it is recorded per measure, and it is visible: the catalog page shows every measure with its domain, metric and direction, and lets one be corrected in place. Metadata entered by hand can be wrong, and a measure filed the wrong way round moves a domain mean quietly. Better to make it inspectable than to claim it is right.
The related trap
Direction sorted, the next thing that distorts a domain mean is a composite sitting in the average alongside the subtests it is made of.