Cohort norms answer “how did this student grow relative to academic peers this year?” Baseline norms answer “how does that growth compare to pre-pandemic rates of learning?”

Cohort-referenced results are, as expected, centered on 50 with even conditional coverage. The baseline results are uneven across grade and content — the pattern the next section reads against cross-sectional status. Grade 11 is held back for section 4.

The diagnostic content is the tilt, not the distance from 10%. The residual lumpiness is a property of the model meeting a discrete scale — it is the model’s constraint, not a defect in the data. For baseline norms a level shift is neither: it is the change in the prior-to-current relationship that the comparison exists to detect.

Both norm sets fit the data well. The difference is the reference: cohort centers on this year’s peers; baseline holds the pre-pandemic standard fixed — so a level shift here is signal, not misfit.

A is the baseline SGP. B is the same quantity with the prior dropped. C averages B over the cohort — and lands exactly on the Mann–Whitney statistic behind Ho’s $V$. The cross-sectional comparison is not a rival to the baseline SGP; it is the zero-covariate case of it.

The measurements move in opposite directions across subjects. A program-wide linking or equating event would move grades and subjects together in one year. That is not what the RICAS data show — which argues against a common scale-drift explanation and for reading these as real, subject-specific changes.

ELA sits above the diagonal at every grade and across the whole range — a uniform downward shift, not a tail effect. Mathematics tracks the diagonal closely, with grades 4 and 5 slightly below it.

ELA agrees strongly across the two measurements — grade 7 highest on both, grades 5 and 6 lowest on both. Mathematics sits close to the reference on both: status slightly above 2019, conditional growth slightly below it. That combination describes a cohort arriving in the grade already ahead of its 2019 counterpart, not a scale artifact — which would show up as a grade far from the reference on one measure and unmoved on the other.

The same students, three consecutive years (ELA). This cohort was in the early-literacy grades during the pandemic — the diagonal lets us watch their baseline-referenced growth as they progressed.

Exceptional growth against a pre-pandemic reference, in a cohort whose measured achievement has fallen relative to that same reference, is not a coherent description of Rhode Island eleventh graders. The most likely explanation is the PSAT-10 to SAT relationship, not the students.

Cohort norms are flat, as they must be. The ELA baseline distribution rises monotonically to 28% in the top decile — nearly three times its expected share. A shape like that is a property of the comparison, not of a cohort.

Where achievement gaps are large, within-year growth is not. Cohort SGPs sit within a few points of 50 for every group. Students with disabilities are the one substantial exception, and that gap is consistent across both subjects.

As the horizon shrinks from three years to none, catch-up rates fall and keep-up rates rise — monotonically, in both subjects. Most of the apparent grade gradient is the target, not the students. Grade-to-grade comparisons are not meaningful unless the horizon travels with them.

Students entering below the 25th percentile face the steepest catch-up targets and attain them least often — the central equity concern for accountability use of these growth results.