docs: note the student groups the research does not model

The CRDC also reports arrests for Asian, Hawaiian/Pacific Islander, and
multiracial students. Nothing on the results page said so, which leaves a reader
to assume the four race groups shown are the whole picture.

The methodology block now names the three groups and gives the paper's own
reason. Worth stating precisely, because it is not a small-sample argument:
white_paper.qmd footnote 4 explains the focus on White, Black, Hispanic and
American Indian students as an analytic choice — prior work documents the
starkest disparities for Black and American Indian students relative to White
students, with Hispanic students included for their share of enrollment — and
line 309 gives the second reason, that the stratified specifications are fit per
group, so covering all 14 reported combinations would mean 70 models instead of
40. Small sample size is the stated reason for excluding *disability* status
(line 168), not race groups.

The note says plainly that the absence reflects scope rather than a finding, and
that no estimate for those students should be inferred from this page. Also
mentions the two upstream exclusions a reader of the totals would otherwise
trip over: arrests not disaggregable by both race and sex (including Section 504
students), and schools not enrolling grade 7 or above.

src/components/ChartPanel.jsx

@@ -241,6 +241,22 @@ export default function ChartPanel({ district, state }) {
           without referral-rate covariates. All rates are per 1,000 students. Modelled distributions
           are drawn from the published posterior predictive draws, fetched in the browser.
         </p>
+        <p style={{ fontSize: '0.85rem', color: 'var(--cv-ink-2)', marginTop: 'var(--space-2)' }}>
+          <strong>Student groups not shown.</strong> The CRDC also reports arrests for Asian,
+          Hawaiian/Pacific Islander, and multiracial students. The underlying research does not
+          model these groups: it focuses on the four race/ethnicity groups where prior work has
+          documented the starkest disparities — Black and American Indian/Alaska Native students
+          relative to White students — with Hispanic students included for their large and growing
+          share of U.S. enrollment. Fitting every reported group would also have meant fitting 70
+          stratified models rather than 40. Their absence here reflects that scope, not a finding
+          about those students, and no estimate for them should be inferred from this page. See the{' '}
+          <a href="https://crdc-api.civilytics.org/api/v1/" target="_blank" rel="noopener noreferrer">
+            white paper
+          </a>{' '}
+          for the full rationale. Arrests that cannot be disaggregated by both race and sex —
+          including Section 504 students — are excluded upstream, as are schools not enrolling
+          grade 7 or above.
+        </p>
       </div>
     </div>
   )