Some measures are worth watching and must never affect a score. Reading rate is the canonical example: it diagnoses, it does not assess.
Mixing them corrupts both. A student who knows speed is graded reads faster and comprehends less — trading away the thing that is graded for the thing that is merely measured.
Bands are anchored to the publisher’s own printed benchmarks, not invented:
| Band | Source |
|---|---|
| Floor | the baseline printed on the student’s own progress graph |
| Target | the rate the teacher’s notes allow for the grade level |
| Ceiling | the top of that graph’s scale |
If you cannot point at where a threshold came from, you have invented a standard and attached authority to it.
A flag combining two signals — high rate, low score — is useful. The wording is where it goes wrong.
Too fast — comprehension slipping asserts a cause. The first two real data points
destroyed it: 175 wpm scored 90, 177 wpm scored 80. Rate moved 2; score moved 10. The
student was tired, which the numbers cannot show.
The flag now reads Check — fast pace, N missed. Same trigger, no claim, and a Note
column beside it for the condition.
A metric without its context becomes misleading the moment everyone forgets. Budget a free-text column next to every automated indicator. It is the cheapest thing in the system and the first thing omitted.
A single reading is noise. The sheet shows the change since the previous one, because the expectation is improvement over a year, not a number on a day.