Public data as a decision system.

Maternal mortality rates for 2023 show the Black rate at 50.3 per 100k births, holding near the 50 mark seen in 2022 (49.5), while the White rate fell to 14.5 from 19.0. These are small annual counts (669 deaths in 2023 versus 817 in 2022 nationally), so single-year movement in any group's rate is noisy and may not be statistically significant. The data detected a wider Black-White ratio in 2023 (roughly 3.5-to-1) than in 2022, consistent with the White average moving downward while the Black rate stayed anchored near 50 — the durable finding is this persistent gap, not the year-to-year tick in either bar. No causal claim is warranted: the pattern is observational, and the data is consistent with, rather than proof of, any driver of the White-rate decline. Notably, the 12-month postpartum Medicaid extension postdates the 2023 outcome window entirely, so it cannot be credited with or blamed for the 2023 figures; it stands as a forward-looking lever aimed at closing the gap, not an explanation of what already happened. Decision: The maternal health program office should treat the Black-White gap as the primary target metric going forward, tracking whether the 12-month postpartum Medicaid extension coincides with narrowing in future years rather than evaluating it against 2023 data.

At grade 8, NAEP math scores show no statistically significant recovery to pre-pandemic levels, and the White-Hispanic gap widened, with White students at 284.0 in both 2022 and 2024 while Hispanic students moved from 261.0 to 257.0. The Black student score moved from 252.0 to 251.0, a change within sampling error and not a statistically significant shift in either direction. Because NAEP scores carry sampling error, single-year moves of a point or two like this one should not be read as real gains or losses. The data are consistent with recovery being concentrated where high-dosage tutoring was actually deployed at grade 4 rather than grade 8: Louisiana was the only state to top its pre-pandemic grade-4 reading performance, and Alabama's grade-4 math score rose from 230 to 236 over the same window, coinciding with states that mandated the intervention rather than proving it caused the gain. Grade 8, by contrast, shows a persistent and widening gap with no comparable recovery pattern, suggesting the lever has not yet been scaled to that grade in most states. This pattern — dosage-linked movement at grade 4, stagnation at grade 8 — is the signal worth acting on, not the noisy one- or two-point swings among Black students. Decision: State education agencies should direct new tutoring funds toward mandating high-dosage grade-8 tutoring in states and districts that have not yet scaled it, prioritizing sites serving Hispanic students given the widening grade-8 gap.

Overdose mortality rates across all reported racial and ethnic groups declined from 2023 to 2024, with the national total falling to 79384 deaths, but the drop was not uniform. American Indian/Alaska Native communities remained the highest-rate group in both years (65.0 in 2023, 51.6 in 2024), followed by Black (48.9, then 33.8) and White populations (33.1, then 24.7), while Native Hawaiian/Pacific Islander, Hispanic, and Asian groups tracked at consistently lower levels. This persistent gap means the overall decline masks a continuing, disproportionate burden on American Indian/Alaska Native and Black populations rather than an equal easing of risk. Note that American Indian/Alaska Native overdose death counts are widely understood to be undercounted due to racial misclassification on death certificates, so the true disparity is likely even larger than these figures show. The timing of naloxone expansion coincided with this downturn, but it occurred alongside major supply-side shifts — fentanyl market saturation and precursor-chemical changes — that can dwarf harm-reduction levers in top-line mortality; the data here detected the turn but does not establish that the program caused it. Any interpretation should treat the panel as evaluative surveillance, not causal proof, and should flag the widening disparity for equity review. Decision: Public health program leads should prioritize targeted resource allocation and improved death-certificate classification protocols for American Indian/Alaska Native and Black communities, while commissioning a supply-side-adjusted causal analysis before crediting harm-reduction levers for the observed decline.

The bars pair 2010, when the ACA created the Hospital Readmissions Reduction Program (Social Security Act section 1886(q)), against 2016: raw, unadjusted 30-day readmissions across all Medicare fee-for-service discharges fell from 16.7 percent to 15.6 percent. Read the concentration, not the top line - readmissions for the penalized conditions dropped 13.2 percent against 4.5 percent for conditions the program does not cover (a 2.4 versus 0.7 percentage-point gap), the sign that the lever, not a background trend, moved the number. MedPAC tested the obvious dodge, hospitals shifting patients into observation stays and ED visits, and found near-identical growth for covered and non-covered conditions, so the drop was not mostly relabeling. Two honest limits: on a risk-adjusted basis heart failure readmissions fall further, 23.9 to 20.5 percent, but part of that reflects hospitals coding more diagnoses after the secondary-diagnosis fields expanded from 9 to 24 in 2011 - a measure that improves partly because it got easier to satisfy is a Goodhart risk, so the raw rates are the conservative read - and the raw decline stalled after 2014. On mortality MedPAC found no adverse effect: risk-adjusted mortality fell across every condition studied. Decision: the penalty moved the targeted number, so hold the incentive where it works while shifting oversight toward the coding drift and the post-2014 plateau, and weigh MedPAC's recommendation to fold HRRP into a broader hospital value program.

Pulled from the College Scorecard field-of-study API - Computer Science bachelor's, median earnings four years after completion, Title IV federal-aid recipients only. These six schools are a hand-picked illustration, not a representative sample, so read the spread as a contrast, not the distribution. Even so it lands: for the same credential in the same field, Carnegie Mellon reports 268,121 and MIT 225,141 at the top against CUNY York 73,044 and Fayetteville State 36,164 at the bottom - a five- to sevenfold range that horizon only widens. Three honesty flags travel with it. Small cells: at the regionals the completer counts are thin enough that horizons run non-monotonic (Fayetteville reads 50,940 at one year but 36,164 at four), so the medians are noisy. Surface dependence: the same federal source answers differently by door - the field-of-study API returns no CS figure for Brown, whose single-year cohort is below the privacy threshold, while the public web tool shows about 214,000 by pooling two award years - so the grain of the query decides whether a number exists at all. Selection, not schooling: the cross-school gap blends the school's effect with who it admitted, so it is not a clean read on what the degree itself causes. Decision: the signal is real enough to steer program-level accountability and student guidance, but only if the grain - school x CIP x horizon x aid-subgroup - travels with it, because a single headline ('CS pays 250k') strips exactly the frame that makes it true. Whether publishing earnings actually changes behavior is separately unsettled - studies find only small, advantage-concentrated sorting.

Virginia's SCHEV report links K-12, higher-ed, and employment records to publish post-completion wages — median annual wage five years out for 2017-18 graduates, measured on 2023 Unemployment-Insurance records (nominal). At the credential median the premium is smaller than the folk model assumes: a four-year bachelor's runs 65,249, an applied-science (AAS) associate 58,393, and a transfer associate 49,444 — so 'higher credential = higher pay' holds on average but only barely, and the distributions overlap heavily. That overlap is the real finding: at these medians a strong applied-science associate can out-earn a weaker bachelor's, so credential tier alone is a poor predictor of individual pay. Two caveats travel with every wage here: it is UI-covered, in-state employment only (out-of-state, federal, military, self-employed, and gig work are invisible), and not every graduate has a reported wage — match rates differ by credential, so the medians describe differently-selected subpopulations. SCHEV itself warns against ranking programs on these wages without adjusting for family wealth, residency, and demographics. Decision: fund and advise by field-and-region outcome, not by credential tier — and open the record-level data (today locked behind data-sharing agreements) so families and funders can interrogate the aggregate instead of trusting it. Built to decide; not built to browse.