Nonpublic enrollment connector

The denominator was held for one year. R.C. 3317.024(E)(2)(d) divides the appropriation for auxiliary services under R.C. 3317.06 and for administrative cost reimbursement under R.C. 3317.062 by the membership in chartered nonpublic schools “as determined as of the last day of October of each school year”. The repository knew that membership for 2023-24 and for no other year, from the Education Landscape fact sheet: 711 schools, 173,156 pupils.

That single observation was load-bearing and it was not closing. Auxiliary Services divided the FY2025 appropriation by it and got $963.39 where LSC publishes $913, inverted the published rate to get an implied 179,713 pupils, and recorded the department’s 711 schools as “short of the redbook’s 747 and the greenbook’s 725” — three separate symptoms of a denominator from the wrong October, each of which read as a defect in one of the sources.

The files were known to exist and nobody had fetched them. #257 recorded in passing that the department’s Enrollment Data page “reaches back to 1978 for public districts and nonpublic buildings”. It is a public CMS attachment page. No connector owned it.

Declare dew-nonpublic-enrollment as a connector in its own right, Wired with a recorded remainder, feeding school, metric and program, over twenty-three sources enumerated from the page’s anchors. Build two fixtures from one read: a building panel keyed on October and IRN, and a sector panel keyed on October and basis.

Enumerated, not composed. The lesson of MR-81 and of the EdChoice designated list, and this page is worse than either: nonpub.xls, nonpub2015.xls, Revised-FY16-nonpub.xls, web_nonpub_2011.xls, oct_08_fy09_adm_web_nonpub.xls, Nonpub_FY2013-Data.xlsx. No pattern generates that set, and a pattern that generated most of it would silently omit the rest.

Its own connector, not a source under dew-scholarship-reports. The criterion that splits a connector is the access regime, which “Scholarship reports connector” settled when the designated list invited the other reading. Both are open fetches, so the regime does not split them — and they are still two connectors, because this is a different office publishing a different page. dew-scholarship-reports is the scholarship channel’s own office describing participation in its own programs; this is the department’s data office publishing an enrollment census of a sector, of which the scholarship channel reaches a part. One connector over both would carry one status for an office whose page this project scrapes and an office whose reports it downloads, and the remainders are unrelated: the scholarship connector is blocked on per-district participation, this one on the district a building sits in. Merging them would make a lifted blocker on either half read as a lift on both, which is the failure dew-payment-reports/dew-scholarship-reports was split to avoid.

Every quantity is a floor, a ceiling and a censored-cell count. A cell printed <10 stands for some number from one to nine, and in a layout of one row a school and grade a small school has almost every cell masked. October 2023 masks 5,113 building-level cells against a published 171,830, and the unmasked cells sum to 146,078: the bound is [146,078, 192,095]. Publishing a point here would be publishing the floor.

Keyed on building IRN, and on nothing else. Ohio’s 607 public districts carry 580 distinct names, which is why every panel in this repository keys on IRN; nonpublic school names are worse — October 2023 holds sixteen schools named St Mary, nine named St Joseph and seven named St Peter. The IRN is the only stable identity the file carries.

The per-pupil figure reproduces, and the statutory denominator is now known. October 2024 is 179,693 pupils in state across 722 buildings. The FY2025 appropriation net of the College Credit Plus earmark, $164,077,754, over that is $913.10, and LSC publishes $913. Over the same count including the 1,478 out-of-state pupils it is $906. So R.C. 3317.024(E)(2)(d) is divided by the in-state count, which nothing in the section says and which no document this project holds states.

Three symptoms in auxiliary-services were one cause. $963.39 against $913, an implied 179,713 against a held 173,156, and 711 schools “short of” 747 and 725 were all the 2023-24 fact sheet standing in for October 2024. The implied membership was right to within 20 pupils. The school counts run 711, 722, 749 across the three Octobers, so the redbook’s 747 and the greenbook’s 725 are later vintages of this series rather than a disagreement with it.

The single-year figure is now a cross-publisher pin rather than the only evidence. October 2023’s state sheet gives 173,156 and its building sheet 711 distinct IRNs — exactly the fact sheet, from a different office. That check is worth more as a check than the number was as a datum.

The sector panel carries two rows for six Octobers and reconciles neither. The 2014-2019 compilation restates FY2014-FY2019 and disagrees with the annual files: October 2017 reconciles exactly, October 2016’s annual total is the compilation’s total rather than its in-state figure, and October 2018 reads 165,346 against 165,511. Keying on basis is what lets both stand.

One file is read against its own label. nonpub_fy17.xls’s building sheet is Revised-FY16-nonpub.xls’s — 5,883 shared keyed cells, none differing — and against the compilation’s per-school totals it matches FY2016 on 86 unmasked schools and no other year on more than six. October 2016 therefore has no building-level data of its own, and the file table says so in a BuildingRows::RepostedFrom rather than in a comment. A pairwise sweep found no other repost in the series.

A remainder that is a real hole, not an unfetched file. The district a building sits in is published nowhere. Auxiliary services flows through that district, so the per-district question the series was wanted for is still unanswerable; county and school type exist to 2006-07 and then stop, so even the coarse geography is gone for the modern years. still_blocked says this, because a reader who sees forty-nine Octobers of building rows will otherwise assume the district is in them.

Fold the files into dew-scholarship-reports as more sources. Rejected, above. Two offices, two pages, two unrelated remainders; the economy is one registry row and the cost is a status that describes neither.

Read the State Total row the historical files carry. Rejected, and this is the trap the source is built around. That row is an Excel SUM over cells already overwritten with the mask text, and SUM ignores text: it equals the unmasked-cell sum exactly on every sheet that carries one. Taken at its label, 1977-78 reads as 243,489 for a sector of at least 260,867. It is a floor wearing the name of a total, and the extraction asserts that identity rather than assuming it.

Reconcile the grade block against the race block into one count a building. Rejected. They are the same pupils measured twice, and the race block — five columns rather than twenty-six — masks a fifth as many cells and bounds far better. But to 1982 some schools reported race and left every grade cell zero, and a few disagree outright. Both are carried side by side, on the same principle as the sector panel’s two bases: the corpus records what does not reconcile.

Match buildings to districts by name or by geography. Rejected as the thing this repository keeps being burned by. A name match against a district roster would produce a district column that is wrong for an unknown fraction of buildings and looks exactly like a right one. An unfilled field that says why is worth more than a filled one nothing can check.

Key the sector panel on October alone and prefer the annual files. Rejected: it would delete the only evidence that the department publishes this sector twice and disagrees with itself.