MR-81 free and reduced-price lunch reports, 1998–2025

Source. Ohio Department of Education, Office for Child Nutrition, “LUNCH MR 81 Report for October”, one directory per year. Type. Primary source — the department’s own annual child-nutrition report. Location. Two, and the second is the one this entry used to deny existed. https://public.education.ohio.gov/MR81/ is an open directory listing of seventeen year directories, MR81_October_1998 through MR81_October_2014. It really does stop there — the Internet Archive holds 101 URLs under it and the newest is October 2014.

The report did not stop; the directory did. From October 2015 it is an attachment on Data for Free and Reduced Price Meal Eligibility, one .xlsx an October, and that page links back to the open directory for “years prior” — the department treats them as one series. Two of the eleven paths still carry MR81 in them. Three different parent segments and three filename conventions, one of them misspelling Eligibilty, so the eleven are scraped off that page rather than composed from a pattern. See #16.

THIS ENTRY DESCRIBED THE WRONG SOURCE. It was written from the directory listing and said MR-81 holds “October enrollment by district, as reported on the MR-81 form”, and the corpus planned a phase around it as an enrollment archive. Opening a file says otherwise: MR-81 is the free and reduced-price lunch report, published by the Office for Child Nutrition, one row per school site grouped by sponsor. It carries an enrollment column because a lunch claim needs a denominator, not because it is an enrollment report.

Everything below is from the files. The slug is left alone because it is cited, and renaming it would break the citation to make the mistake less visible.

What it contains. Per school site: sponsor, county, site IRN, kitchen type, enrollment, and free and reduced-price lunch applications, each also as a share of enrollment, with their total. Sponsor subtotals in the rendered files.

Why it is worth more than the enrollment framing suggested. Seventeen consecutive Octobers of free and reduced-price counts is a district poverty series, and the corpus has none. Ohio’s disadvantaged pupil impact aid runs on an “economically disadvantaged” count whose definition R.C. 3317.03(B)(21) hands to the department, and free-lunch eligibility has historically been the department’s operative test. A seventeen-year run of it is the closest thing available to a long series of the count DPIA is paid on.

Four breaks, each of which produces a wrong series if unhandled:

• The enrollment column changes definition in 2010. AdmCount and PctFreeAdm become CECount and PctFreeCE. The 2014 header states what CE means: “Current Enrollment (CE) — highest daily number of students with access to the program.” That is not average daily membership and not the same quantity as what preceded it. A reader resolving columns by position would splice the two silently.
• The poorest sponsors leave the main file in 2012. From FY2012 the report splits into Traditional, Provision 2, and Community Eligibility (CEP/CEO) streams — in 2012 as three subdirectories, in 2013 and 2014 as separate files. The 2014 Traditional header says outright that it “Excludes Provision 2 and Community Eligibility Option (CEO) sponsors.” Districts adopt CEP precisely because their poverty is high enough to qualify for universal free meals, so a series built from Traditional alone drops the highest-poverty districts, and drops more of them every year as adoption spreads. This is the single most dangerous property of the source.
• Sponsors are not districts. The 2005 file’s first rows are a Residential Child Care Institution and a county Board of MR&DD. Any district panel needs a sponsor-type filter.
• The 1998-2000 files have no sponsor-type column at all, so there is nothing to filter on. They are one row per school with the district named and numbered on it, and the type has to come from 2001, which leaves some thirty-five sponsors a year — three thousand to eight thousand pupils, against 1.8 million — with no type anything states.

Layout eras. Four, all resolvable by column name:

1998         comma, school-centric: BREAKFAST IND, SCHOOL, SCHOOL IRN, ..., DISTRICT IRN
1999-2000    comma, same columns, published as a "Delimited" file
2001         comma, sponsor-centric: County, SponsorIRN, ..., AdmCount, ...
2002-2009    tab, sponsor-centric; ProvisionYear appears in 2003
2010-2011    tab, sponsor-centric, AdmCount renamed CECount
2012-2014    three program streams, separately published
2015-2025    one workbook an October, streams on a column, counts masked under ten

Every year from 1999 carries a delimited file alongside the rendered text; 1998 is comma-delimited already. File naming is inconsistent to the point of needing a per-year table — MR81-Oct2002-TabDelimited.txt, MR81_Oct_2010_delimited Revised 0911.txt, MR8_Oct_2011_Delimited.txt — and one year’s name is misspelled.

Status. Wired in full. Twenty-eight Octobers in thirty-two files, across four readers: a fixed-width one for 1998–2000, a delimited one for 2001–2014, one for the printed report — the only form the 2013 community-eligibility stream was ever posted in — and a workbook reader for 2015–2025. The panel is crates/dispersion/fixtures/mr81-sponsor-panel.csv, 33,281 rows.

Two things the workbook era carries that the delimited one did not. Its Notes sheet states the community-eligibility rule this repository had previously measured off the 2014 file — the printed percentage is the directly-certified count times 1.6 over enrollment — so the ceiling half of the corpus’s band is the program’s own arithmetic rather than an inference. And it masks a count printed as <10, which is not a bound: when either of the free/reduced pair falls under ten both are masked, so the value behind one can be 15 or 19. The masked count is recovered from the percentage printed beside it, on an identity that reproduces all 42,983 unmasked cells exactly. censored on each row says how many arrived that way.

What the extraction established, beyond the rows.

• The 1998–2000 files are fixed-width, not comma-separated. Every row is exactly 170 bytes with the thirteen separators at the same offsets, so the seven district names a year carrying a comma never shift anything. The earlier reading of this entry — that they need “a last-field rule” — was describing a problem that only exists if the file is split on its delimiter.
• They were written in 2004, not in their own years. All three carry the same 4,246 schools, zero-filled where a school was not participating, and the directory dates them 5/2004 and 8/2004. October 2000 is the one year whose contemporaneous printing also survives, and the two agree on 4,233 of 4,236 schools — which is what makes the back-fill usable.
• 2001’s delimited file was being misread, and the corpus had published the result. It is the only comma-delimited year; nine rows carry a comma inside a school name; two of them are Cleveland City SD, and read positionally they put site IRNs 00026450 and 00093153 in the enrollment column. The statewide public figure was 1,921,522 against a true 1,802,937 and the poverty share was 27.7% against 29.5%. Repaired by anchoring the row’s nine trailing fields from the right; after the repair, free and reduced approvals match the printed report exactly for all 1,227 sponsor blocks.
• The three streams count three different things. Traditional is current-year applications. Provision 2 is applications frozen at the base year the file names — the same sponsor reports the same counts in 2012, 2013 and 2014. Community eligibility collects no applications at all; its PctFreeRedCE is the directly-certified count times 1.6 capped at enrollment, which reproduces in all 735 of the 2014 rows. Joining Traditional alone is the trap this entry always warned about; joining all three by their approval columns is a different trap with the same shape, and both make poverty appear to collapse.
• 2013’s community stream has no delimited file. Only the printed report, which names its sponsors and neither numbers nor types them, so the identifiers come from the delimited files either side. Believe to Achieve-Canton is filed under Cuyahoga in 2012 and Stark in 2013, so the lookup falls back from name-and-county to name.
• The 2012 community file’s header is one column wider than its rows. A blank column name sits before CEOEligibleStudents, so resolving that column by name points one past the end of the data — silently, since the trailing columns are all numeric.

Other caveats:

• This is a count of applications, not of eligible children, in the years before direct certification. Under-application among eligible families is real and varies by district.
• Two Octobers are not a reading of Ohio. Under USDA’s nationwide free-meal waivers, in force for 2020-21 and 2021-22, a sponsor could serve every student free without collecting one application — and almost every one stopped filing. October 2020 and October 2021 carry 296 and 261 sponsors against about 850, and a quarter of the enrollment; their band reads twenty points above the years either side and every point of that is about who filed. The feed marks them comparable: false and the page prints “not the state” in the row.
• Traditional is emptying into community eligibility. 2,849 sites in October 2015 against 1,782 in October 2025, while CEP goes 844 to 1,754. So no site count can serve as a floor on a parse in this era, and the extractor floors the October instead.
• District identity is not stable across twenty-eight years. Consolidations, closures and IRN changes mean a naive join produces a panel whose membership silently varies — the same problem dispersion::ohio_panel now measures rather than assumes for the F-33, where the FY2022-23 directory names 124 fewer FY2012 agencies than FY2022 ones.

Also read by #

• crates/dispersion/src/mr81.rs — the panel and the aggregates over it, including the band the split years carry instead of a share.
• The meal_program block of the feed, which is the card “What the poverty weight is counted on” on the site’s /history route.
• DPIA, whose fourteen-year poverty series is this file.

The enrolled ADM node records the three-observation limitation this source was expected to relieve; on the corrected reading it would relieve it only as an inference, since CE and headcount are neither of them ADM.