Report card connector

Decision proposals approved nine connectors, all of them finance-side: department payment reports, tax abstracts, LSC appropriations, statute and case archives, facilities projects, Census F-33, NCES CCD, and BLS CPI. That list was drawn when the corpus had no outcome variable and no intention of acquiring one — every metric in scope measured what a district was given, could raise, or spent.

The outcome-axis phase changed that. Evaluating an external white paper that correlated the Ohio School Report Card’s Performance Index against its expenditure-per-equivalent-pupil figure required both variables, and the corpus held neither. The evaluation could be written from the paper’s published tables, and was, but every figure in it was secondary — quoted rather than computed — and the corpus’s central objection to the paper was a claim about the data that could only be settled by holding the data.

crates/connect/src/registry.rs pins the connector list at nine in a test, deliberately: “A connector dropping out of this list is a decision, not an oversight, and should fail here first.” Adding one is the same kind of event and is recorded here for the same reason.

Three publications were identified, all on the report card’s public download API (edu-prd-reportcard-datarefresh-api.azurewebsites.net/api/v2/Categories, which enumerates every file with its storage URL):

24-25_Achievement_District.xlsx Performance Index, 607 rated traditional districts, with the 2023-24 and 2022-23 scores in the same row. 2425_DISTRICT_SPEND_PER_PUPIL.xlsx Expenditure per equivalent pupil, federal and state-and-local components. FY25 Expanded List.xlsx Operating expenditures with BOTH weighted and unweighted ADM, per IRN, on separate sheets.

The third is why the connector earns its place. It carries one numerator against two denominators, which turns the corpus’s denominator objection from an argument into a measurement.

Approved. A tenth connector, dew-report-card, publisher Ohio Department of Education and Workforce, feeding metric and education-agency, status Wired, with the three sources above. The registry test that pins the connector list is updated from nine names to ten.

It is a separate connector rather than three more sources under dew-foundation. The two publish different things for different purposes on different infrastructure: dew-foundation retrieves finance files from education.ohio.gov under Finance and Funding, and this one retrieves report-card files from reportcardstorage.education.ohio.gov under an accountability program with its own release calendar and its own methodology revisions. Merging them would put a Performance Index redesign and a payment-report revision behind one status flag.

One operational detail is recorded because it will otherwise look like a bug. The storage host requires a read-only SAS query string, which the report card’s own public web application embeds in its JavaScript bundle in clear text and which expires 2031-07-28. Without it the storage host answers ResourceNotFound with HTTP 404 rather than 403, so a fetch that omits the token looks like a moved file rather than an authorization failure. The token is committed in the registry alongside the URLs it belongs to.

The alternative was to leave the two new metrics carrying only secondary figures, which the corpus’s own provenance rule makes untenable: “Every numeric claim in the corpus should reach a catalog entry in one hop.” Quoting a white paper that quotes the department is two hops, and the intermediate hop is exactly where the corpus disagreed with the paper.

Against that, the cost is real and worth naming. The connector count is a rough proxy for how much surface this repository maintains, and the nine were chosen to bound it. Ten is a deliberate expansion, justified by the specific fact that the report card is the only publisher of an Ohio district outcome measure, and the corpus cannot answer an outcome question without one.

The narrower option — adding only the Expanded List, which alone settles the denominator question — was rejected. It would leave the Performance Index itself uncomputable, and the denominator finding is only interesting because of what it does to a correlation with the Index.

A note against a future reader’s assumption: approving this connector does not approve the outcome measure as an adequacy standard. See corpus/metric/performance-index.yml, which records that Ohio publishes an index and no threshold, and corpus/doctrine/adequacy.yml, which records that a spending-outcome correlation is not an adequacy test in either formulation.