What a citator reaches

“The last three connectors” wired the four DeRolph opinions and left two clauses in still_blocked:

trial-level rulings such as the 2025 EdChoice decision are not in the supreme court archive at all, and citing_cases needs a citator rather than a document

Both are true sentences and neither is a finding. The first says where a document is not; it does not say whether the document is anywhere. The second says what a property needs; it does not say whether the thing exists. The backlog issue that carried them forward was explicit about the second — “worth recording whether a free or public citator covers Ohio appellate decisions, because if not this property is unfillable and should say so rather than sitting open”.

Meanwhile EdChoice Constitutional Challenge (Franklin County, 2025) carried three [open] marks — the caption, the docket number, and the appellate status — and rested on reporting for all of them, while the scholarship-program nodes leaned on it for the constitutional question.

Wire Columbus City School Dist. v. State, 2024-Ohio-1217, 10th Dist. No. 24AP-60, the one appellate decision in the EdChoice case, into a fixture of its own.

Do not wire the merits ruling, and record why in terms of the clerk’s conditions of use rather than in terms of difficulty.

Do not fill citing_cases, and replace the assumption behind that refusal with a measurement: the recall of the free citator’s citation graph against its own full text, named cases included.

Rewrite the connector’s still_blocked so it states what was measured instead of what was assumed.

A free citator covering Ohio exists. CourtListener, run by the Free Law Project, serves this project’s user agent without complaint on /api/rest/v4/search/, needs no credential for it, and holds 87,949 Ohio Supreme Court opinions and 110,756 Ohio Court of Appeals opinions. Full-text search reaches opinion bodies, checked with a control phrase. So the property is not unfillable, which is what the issue asked and is the opposite of what “needs a citator” implied.

And filling it from there would publish a list missing the two cases that matter most here. Against DeRolph I, counting only decisions other than DeRolph’s own later opinions:

• the citation graph identifies 10 Ohio decisions;
• full-text search for 78 Ohio St.3d 193 finds 25;
• and the graph is a strict subset — every decision in it is in the text set.

The fifteen it misses include Simmons-Harris v. Goff (1999), the Ohio Supreme Court’s own voucher decision, and State ex rel. Ohio Congress of Parents & Teachers v. State Bd. of Edn. (2006), the community-school constitutional challenge. A corpus about school funding that published a citing_cases edge from this graph would be asserting that Ohio’s voucher case does not cite DeRolph. That is worse than an empty property, because an empty property is visibly empty.

So the refusal stands and the ground under it has changed completely: it used to be “no citator is available” and it is now “the available one has 40% recall against its own corpus, and the missing 60% is not random with respect to this subject.”

Six properties of the citator, recorded so nobody derives them twice.

• cites: takes opinion identifiers, not cluster identifiers. A DeRolph cluster holds five to seven opinions. cites:(6775348) — the lead opinion, the one that looks like the opinion — returns 0. All five siblings together return 29. The wrong query returns a plausible zero rather than an error, which is the failure mode this repository keeps meeting.
• Each DeRolph opinion is in the database twice, once from the reporter feed and once from Ohio’s own WebCite feed, and the citation counts split between the two: 129 and 3 for DeRolph I, 49 and 1 for II, 29 and 1 for III, 12 and 7 for IV.
• citeCount is not a count of citing cases. It counts citing opinions, each concurrence and dissent separately; the search counts decisions. The two differ by roughly four to one.
• A nonexistent court identifier returns count: 0 under HTTP 200. The question “does CourtListener hold Ohio common pleas decisions” was first answered 0 from a court code that does not exist. A coverage negative from this API means nothing without a positive control, which is the same discipline this repository applied to the CCD’s unused status code and to the State Board’s minutes.
• Anonymous requests are limited to five a minute. A connector walking a citation graph would be pacing against that, not against the network.
• The endpoints that expose the graph directly — /opinions-cited/ and /clusters/ — return 401. They want a free API token nobody here holds. And courtlistener.com itself, the human site as opposed to the API, is 403 CDN-blocked to a non-browser agent. So the API is more open than the website, which is the reverse of the usual arrangement and worth knowing.

The 2025 ruling is not withheld by an archive. It is published under terms that forbid redistribution. The Reporter of Decisions’ own source list is the supreme court, the twelve district courts of appeals, the Court of Claims and Miscellaneous — no common pleas court at all, so this is not about EdChoice or about any one case. The ruling is on the Franklin County Clerk of Courts’ Case Information Online, which returns 200 to this project’s user agent, and whose conditions of use say the data “is not intended for distribution by other persons, entities or organizations” and that an organization wanting files “must submit a Public Records Request”.

That is the third time this repository has arrived at R.C. 149.43 as the remaining step, after the educational service center minute books and the pre-1997 session laws. It is worth saying plainly that this is not a research problem.

What the appellate decision does settle. 2024-Ohio-1217 states on its first page that the parties are Columbus City School District et al. v. State of Ohio et al. and that the trial court case is Franklin C.P. No. 22CV-000067. Two [open] marks resolved from a primary source one hop from the node, on a case whose identity the corpus had been carrying from reporting.

It is also substantive about the litigation without being about the merits: the plaintiffs subpoenaed the President of the Ohio Senate for a deposition on 22 March 2023; he moved to quash on legislative privilege; the trial court modified the subpoena to twenty written questions; he appealed; the Tenth District dismissed for want of a final appealable order because he had not shown that an appeal after final judgment would leave him without a meaningful remedy. So the case’s entire appellate record, as of this revision, is one dismissed interlocutory appeal about who may be deposed.

And no Ohio appellate court has decided the merits appeal. Checked against an index current to 31 July 2026 for the Tenth District — a positive control, because a coverage claim about a recent filing is worthless without one.

One defect in this repository’s own reader, found by wiring a second court. decided_on took the date from an opinion’s first page by finding Decided and then reading to the closing .) of the supreme court’s parenthetical, falling back to the end of the line. The courts of appeals print Rendered on March 29, 2024 bare on a line with no parenthesis, so the fallback never fired and the search ran on to the next .) anywhere in the document. The first appellate opinion wired here came out with sixty-eight lines of holding in its date field. The rule is now whichever of the two comes first, and there is a test with one document of each shape.

And an assumption in the rebuild, caught before it cost anything. The opinions fixture was built from every source of the ohio-courts connector, which was the same thing as “every DeRolph opinion” for exactly as long as the connector held one case. Sources already declare which fixtures they feed; the rebuild now selects on that. Otherwise a file whose first line reads “DeRolph v. State, the four opinions this corpus cites” would have held five.

Fill citing_cases from the citation graph and note the recall. Rejected. A footnote does not survive the trip: an edge in a graph is read by things that do not read the note, and the corpus would be asserting the absence of Simmons-Harris v. Goff from DeRolph’s citing cases. The measurement is published instead, where a reader who wants the edge can see exactly what it would cost.

Fill it from the full-text hits instead, which is the larger set. Rejected, and this is the more tempting of the two. A string match on a reporter citation is not a citation — it catches a case quoting a party’s brief, and a string cite in a dissent, and it is still bounded by one database’s corpus rather than by Ohio’s reports. Twenty-five is a better number than ten and it is not the number, and publishing it as the number would be the same error at a different scale.

Get a CourtListener API token. Not rejected — deferred, and it is the obvious next step for anyone who wants this. It is free, it unlocks /opinions-cited/ and /clusters/, and it may change the recall measured above, which was taken through the one endpoint that answers anonymously. It is an account rather than a fetch, so it belongs with the other things a person can unblock and an agent cannot.

Retrieve the merits ruling from the clerk anyway, since one document is not bulk. Rejected. The terms do not turn on volume for the part that matters: the data “is not intended for distribution by other persons, entities or organizations”, and committing it to a public repository is distribution. Reading it is fine and this repository is not a reader, it is a publisher. The narrower point — that “efforts to mine large quantities of data … will be detected and stopped” — was not tested and has nothing to do with the decision.

Wire the appellate decision into derolph-opinions.txt as a fifth record. Rejected on the file’s own first line, which says it holds four DeRolph opinions. A separate fixture costs one constant and keeps the litigation record honest about what is in it.

Leave the connector still_blocked sentence as it was, since both clauses are still true. Rejected. They are true and they are the wrong sentences: one describes an archive that was never going to have the document, and the other describes a missing tool that turns out to exist. A blocker that names the wrong obstacle is read afterwards as a fact about the world, which is the failure this repository has now recorded six times.