Facts and figures connector

Home education was in this corpus only as an aggregator’s series. Johns Hopkins Homeschool Hub — state and district home-education counts catalogs a Johns Hopkins workbook compiling counts the states collect, with four documented defects in its Ohio district sheet and a rule that nothing in it can be [verified], because the source of record is elsewhere.

That entry named the one exception: Home School — 53,051, for 2023-24, in the department’s own Ohio’s Education Landscape fact sheet. It called that figure [verified] on the strength of having been read from the department’s PDF — and then, in the same entry, recorded that every prior-year edition of that same sheet returns an identical 1,245-byte 404, that the older 2013-2014-WHERE-KIDS-COUNT is gone the same way, and that both facts-and-figures index pages archive nothing.

So the corpus’s only verified home-education fact rested on a live URL with no digest behind it, in a repository whose entire retrieval practice is pinning publications by SHA-256, and the publisher’s demonstrated habit is to replace the file in place. The gap was not noticed by any gate: catalog-uncited reports registered sources nothing draws on, and this was the reverse — something drawn on that was never registered.

Reading the sheet for that one cell showed the table around it is worth more than the cell. Its School Options table counts thirteen channels — chartered private schools, five voucher programs, two kinds of community school, open enrollment, joint vocational districts and home education — statewide, for one year, on one page, counted by one publisher.

Declare dew-facts-and-figures as a connector in its own right, Wired with a recorded remainder, feeding metric and program, with Ohio’s Education Landscape 2023-2024 as its single source, and commit its School Options table as a rebuilt fixture.

Kept apart from dew-foundation deliberately. Same department, and folding it in as a fourteenth source was the obvious economy. dew-foundation is per-agency state aid — its note says so — and every fixture it produces is keyed on a district. This one is statewide by construction and keyed on nothing. A connector whose sources answer “how much does this district get” and “how many Ohio children are schooled some other way” carries one status for two questions, and the second is the one with a blocker.

Parsed by column offset rather than by label. The sheet is two-column and pdftotext -layout interleaves them, so Community Schools appears twice on one line — 342 as a count of schools on the left, 116,973 as an enrollment on the right. A label-first reader takes whichever it meets first and produces a plausible wrong number. The School Options heading fixes the column and everything is read from that offset rightward.

The channel list is pinned. A program that stops being reported fails the build rather than leaving the fixture a row shorter, on the same ground the scholarship report’s reader follows: an absent row and a renamed one look identical downstream.

The corpus’s one verified home-education figure is now pinned by digest. That is the whole point. education-landscape-2024 carries a SHA-256 and a byte length like every other publication here, so the next edition replacing this one in place is a digest mismatch to read rather than a number that quietly changed.

Two internal identities are now asserted where the fixture is read. The five voucher programs sum to the published voucher total exactly, and the two community-school types sum to theirs. Both would break if a column shifted, and nothing else would notice.

A cross-source check the corpus could not previously make. The same five voucher programs appear in the 2025 Scholarship Annual Report for the following school year, and every one is larger. Two independent department publications, consecutive years, the same five names, all moving the same way — which is the corroboration that the fact sheet’s table counts what it appears to count.

A series with one retrievable member, said as still_blocked. The prior editions are gone and the department archives nothing, so a refresh that finds no second year is not a failure to look. This connector cannot become a time series from the publisher’s side; it would take the Internet Archive holding editions under a URL pattern the probe has not confirmed.

What it still cannot answer is where. Every figure is statewide. The concentration is the interesting part for a funding question — a population across some 600 districts is thin almost everywhere and heavy somewhere — and only a records request reaches it, tracked at #257.

Leave the figure as prose in the catalog entry. Rejected. It was already being cited as [verified], and a verified claim whose evidence is an unpinned URL on a page that has eaten every one of its predecessors is a claim with a countdown on it. The cost of registering is one registry row and a decision record; the cost of not registering is discovering the loss later with nothing to recover from.

Fold the sheet into dew-foundation as another source. Rejected, above: one status for two access regimes and two kinds of question.

Commit the Johns Hopkins workbook as the fixture instead, since it has twenty years. Rejected on the ground its own catalog entry gives: a Google Sheets export has no stable content address, so nothing here could rebuild byte-identically from it, and its Ohio district sheet has four documented defects including a year label that contradicts the workbook’s own state total. The series stays cataloged and [inference]; what is committed is the department’s.

Extract the whole sheet rather than the one table. Rejected for now. The educator licensure and demographic tables are real data and no question in this corpus asks them anything. A fixture built for a question nobody has is a fixture nobody checks.