Before there were service centers

“The directory cannot say why” wired the delimited era, 2008-09 through 2023-24, and deliberately left the fourteen years before it alone. Its stated reason was not difficulty:

it would be a second reader built for no consumer

— the F-33 finance panel starts at FY2009, so no committed fixture wanted a directory year before 2008-09. The backlog issue that carried it forward said the same thing more sharply: “a reader built ahead of its reader.”

What changed since is where the unanswered question sits. “The order was never the states” established that a territory transfer is ordered by an educational service center governing board, not by the State Board of Education, and the phase after it committed five Auditor of State reports that recite those orders — one of them an educational service center’s account of its own dissolution (Auditor of State — district and service center audit reports).

So by the time this was taken up, the question the directory could not answer had moved. It was no longer only why did a district stop but what are these governing boards, and where did they come from. Nobody had asked the older years that.

Wire 1994-95 through 2007-08, giving thirty consecutive school years with no gap: crates/dispersion/fixtures/ccd-lea-directory.csv, 30,655 agency-years, up from 17,618.

Read the thirteen fixed-width years from a per-year byte-position table, fixtures::CCD_FIXED_WIDTH, and hold every record to the length its own layout document states. Read 2007-08 through the existing alias table, because it is tab-delimited despite sitting in the middle of the fixed-width filenames.

Take the status column from the file where the layout document is wrong about it, and say so in the table rather than silently.

Add one cross-year check to the reader: every agency absent from a year must have been filed closed in the year before. Publish types_by_year so the two vocabulary recodes are visible.

Publish no successor for any departure, exactly as before. The county-to-compound-name correspondence is stated in prose as an inference and committed nowhere.

Leave 1986-87 through 1993-94 unwired, and say why in the connector rather than in a note somewhere.

The bodies that make the orders were themselves consolidating, and the register denies it. Ohio has 86 county boards of education in 1994-95 — the CCD types them 3, supervisory union administrative center. Thirty-nine of them disappear between 1995-96 and 2001-02, in ones and fives, while multi-county centers appear beside them carrying the departed counties in their own names: Athens and Meigs leave and Athens-Meigs arrives; Ross and Pike leave and Ross-Pike arrives; Guernsey, Monroe and Noble leave and Guernsey-Monroe-Noble arrives. Fourteen such centers join, so type 3 is 61 rather than 47 by 2001-02; sixty of those are recoded to type 4 the following year and the sixty-first, Carroll-Harrison, leaves. Forty type-3 departures in all.

Every one of the thirty-nine is filed under code 2 — “closed with no effect on another agency’s boundaries”. So the defect the previous phase found in the district rows is the source’s standing practice, it is not particular to school districts, and it did not begin in the 2010s. It is being applied to the governing boards whose minute books hold the instruments the corpus spent two phases looking for.

The negative is now made over thirty completely enumerated years. Code 5 — “significant change in geographic boundaries or instructional responsibility” — appears in zero of 30,655 Ohio agency-years. All 689 departures are code 2, without one exception in thirty years: 616 community schools, 66 service agencies, five regular districts, one local district and one state agency. That is the third time this repository has argued a negative by enumerating a population rather than by failing to find something, after the casino floor and the State Board’s minutes.

The one departure with an instrument behind it behaves exactly like the district cases. Geauga County ESC’s final audit, already committed, recites its merger with Lake County ESC into the Educational Service Center of the Western Reserve on 7 November 2019. The directory files Geauga County ESC closed with no effect on anyone. Lake County ESC keeps LEAID 3904786, keeps IRN 047860, is coded 1 Open in the year before, the year of and the year after, and takes the new name in 2020-21. The only trace of the merger in the federal register is a rename on the survivor — which is the derivation the previous phase tested and rejected as unreliable, so this is a case where the unreliable signal happens to be right and there is still no way to know that from inside the source.

The recorded filename convention was one year late, and the recorded record lengths caught it. The catalog said pau9Xdata.zip covered 1994-95 through 1998-99 and agNNNa_dat.zip 1998-99 through 2006-07. The publisher’s own table says pau94datr is 1994-95 and ag981c is 1998-99: the archives are keyed on the year the survey closed. The tell was internal — the nine record lengths recorded in the same note (722, 722, 723, 725, 729, 729, 519, 519, 530) belong to ag98 through ag06, and under the recorded filenames the last of the nine would have been ag071a, which is not fixed-width at all. Each file’s layout document states its school year twice, which is what settles it rather than the sizes.

One layout document states a column’s position wrongly and the file is what decides. The 2003-04 layout gives the status byte as 281 — which is the metropolitan indicator’s position, stated four lines above, in the same document, for the same year — and gives it again elsewhere as 384. 284 is the one byte the document leaves unaccounted for, and it is where the two neighboring years put it. The check that settles it is not neighborliness: at 284 every agency the file flags closed is absent from the following year, and at 281 that fails immediately. A wrong position here would not have errored. It would have yielded a column of plausible small integers and a panel in which nothing ever closed.

Two other documents are wrong in ways that would not have been caught by reading them. The 2005-06 layout is the previous year’s with the heading changed: its prose still says “data for the school year 2004-05” and its observation count is still 18,085 where the file has 18,213 records. The 1999-2000 layout names ag991a.dat in its header while describing the second revision. Both state a correct record length, which is the worst arrangement available — the parts a reader would check with are wrong and the part being checked is right.

The layout that was supposed to be the hard part is thirteen lines. The record length moves in seven of the nine middle years, and almost none of that movement is in the first 300 bytes. The agency identifier is bytes 1-7 in every one of the thirteen years, the state identifier 8-21, the name starts at 22. The name’s end and the agency type move once, together, in 1998-99. The status byte moves four times. Four of the five columns this reader takes are constant across thirteen years of a format that changes length seven times.

The arrival code carries the same denial as the departure code. Ohio forms two school districts in thirty years and files both under code 3 — “a new education agency formed with no effect on another agency’s boundaries.” Monroe Local appears in 2000-01 in the same year that Middletown Monroe City is renamed Middletown City, same identifier, same IRN, coded open on both sides. Manchester Local appears in 2004-05 and Adams County/Ohio Valley Local does not change at all. Both new districts carry IRNs from Ohio’s original county-ordered block — numbers Ohio had used before, which the federal register has no way to express.

A third, PEEBLES, IRN 000441, is filed in 2004-05 as “scheduled to be operational within 2 years” and closed the next year without ever reaching open: a district announced and abandoned. It is also the only type-2 departure in the fourteen new years, which means not one operating Ohio school district leaves the federal register between 1994-95 and 2007-08. The five district departures in thirty years are still the five the previous phase named, all after 2014.

So the defect is symmetric, and this was not looked for. The corpus had been asking why a district stopped; the same source cannot say where one came from either.

The agency type is not comparable across three years, and a rule reading type changes as events would find 532 that are not. Measured by following each identifier across the seam, rather than by differencing counts that move for two reasons at once:

• 2000-01: 47 agencies move from type 1 to type 7. They are Ohio’s first community schools, filed as regular school districts for two years. This is the dangerous one — on the type alone Ohio appears to gain 47 school districts in 1999-2000 and lose them again.
• 2002-03: 60 supervisory-union centers and 49 joint vocational school districts move into type 4 together. Type 3 empties; type 4 goes from nothing to 109.
• 2006-07: 376 agencies move from type 2 into type 1. Type 2 empties.

types_by_year exists so a consumer can see this, and the test beside it pins all three.

Counting Ohio’s districts by federal type gives three discontinuities, none of them Ohio. Types 1 and 2 together: 661 through 1997-98, then 675 and 708 in the two misfiled years, then 662, then 613 to 622 from 2002-03 once the joint vocational districts leave. The number underneath barely moves — 661 less 47 joint vocational districts is 614 in 1994-95, and 619 in 2023-24. Meanwhile type 7 goes from 24 county data-processing centers in 1994-95 to 503 agencies in 2005-06 and back to 345. Whatever the community school program did to Ohio, it did not change the number of school districts.

Every superseded revision that is still served was compared. NCES links only the latest edition and leaves the earlier ones reachable. For 1999-2000, 2003-04, 2004-05, 2006-07 and 2007-08 the earlier edition was fetched and its Ohio rows are identical to the wired one in the five columns this reader takes. That is not a claim that the files are identical, and it is recorded as the narrower thing.

No published figure moved. The web feed is byte-identical: nothing in it reads a directory year before 2008-09, and the sixteen wired years are unchanged to the byte.

Take the 2003-04 status column from its layout document. Rejected, and it is the decision this phase turns on. The document is a primary artifact and this repository’s habit is to prefer it to a guess — but 281 is demonstrably the metropolitan indicator’s byte in the same document, and the file itself answers: the closure check passes at 284 and fails at 281. Preferring the document here would have produced a year in which no Ohio agency ever closed, sitting quietly inside a thirty-year panel.

Skip 2003-04, since one bad year in thirty is a small loss. Rejected, and not on completeness. A gap would break the check that makes the whole era trustworthy: closures are verified against the next year’s membership, so dropping 2003-04 would blind the reader for 2002-03 as well and leave two years resting on nothing but their layout documents.

Wire the nine years back to 1986-87 too. Rejected, for the reason the previous phase gave and this one did not find wanting. They are the same format and they would extend the negative, but a negative already made over thirty completely enumerated years is not made stronger by thirty-nine, and Ohio’s county boards are all present and unchanging through 1994-95. If a question arrives that needs the 1980s, the reader is written.

Commit the county-board successors from the compound names. Rejected. It is the strongest form the answer takes anywhere in this source and it is still a reading of a name field: the timing is up to two years out of alignment, and the same field moves one survivor between GUERNSEY-MONROE-NOBLE and OHIO VALLEY and back over four years. Twice, in 1994-95 only, a clerk wrote the answer outright — ASHLAND (SEE ASHLAND-WAYNE) C, WYANDOT (SEE SENECA-WYANDOT) — and never again, which is the measure of how far this is from a field.

Report “the fixed-width era is hard” and close the issue. Rejected. It was recorded as hard on the strength of one probing session, and the two things that made it look hard — a per-era column map, and positions moving in seven of nine years — are both wrong about the columns this reader needs. The thing that was genuinely hard is not in that description at all: a publisher’s layout document that states a position wrongly, twice, in two different places.

Bump the contract and put the older years in the feed. Rejected. Nothing on the site reads the directory; it is a join and a membership record, and adding thirty years of agency rows to a feed nothing displays would be shipping weight for no reader.