The last three connectors

Five connectors had been unblocked in as many phases, every one by opening a file the corpus had already cataloged. Three remained, and after the last round none of the three had an obviously stale reason. This is the attempt on all three, made together so the result is a statement about what is left rather than another single find.

Probe each properly — not the path the blocker names, but whatever path exists — then wire what is retrievable and rewrite what is not.

Wire ohio-courts for the four DeRolph opinions. Leave the other two declared, and replace both blockers with what actually happens.

ohio-courts: wired, and it makes a load-bearing citation checkable.

The blocker had two clauses of very different quality. “Opinions are PDFs” stopped being a blocker the moment Format::Pdf had a reader — all four DeRolph opinions retrieve on a plain request. “Trial-level rulings such as the 2025 EdChoice decision are not in the supreme court archive at all” is correct and unfixable from here: a common pleas ruling is not in the supreme court’s archive because it is not the supreme court’s.

What wiring it bought is specific. regime_diff::RATES carries the charge-off progression — 20 mills, 20.5, 23 — with a session-law citation on each, sourced to DeRolph I ¶97. Until now that was a claim about a document nothing here held. Every authority string is verbatim from the opinion, and so is the base: “the charge-off is the total taxable value of real and tangible personal property in the district times a certain percentage.” The DeRolph-era half of the valuation-base distinction — the half the corpus got right — is now sourced to the document that drew it.

And I wired the wrong case, then caught it. derolph-iii first pointed at 2001-Ohio-114, which is State ex rel. Smith v. Industrial Commission, a workers’ compensation appeal. The WebCite was guessed from a plausible pattern rather than read off the document; it retrieved a real PDF from the real archive and would have passed any check that did not ask what the file was. The correct citation is 2001-Ohio-1343, 93 Ohio St.3d 309. The test now matches each opinion against its reporter volume and page, which the archive prints on the first line.

That is the sixth wrong citation this corpus has produced, and the second I produced myself while building the thing that catches them.

dew-payment-reports: still declared, and the blocker was wrong in both directions.

“No index and no stable path” is stale — the department lists 38 direct .xlsx payment reports for FY2026 and FY2027 at fixed URLs. “The years before about 2015 are not on the current host at all” is wrong in kind: the department publishes Foundation Legacy Payment Reports (1999–2021), covering the entire deduct era, on its reports portal. The portal gates on sessionStorage.claims and requires an OH|ID account.

So the era this connector exists for is retrievable in principle and behind a login, which is a different problem from an absent index and not one to route around. The era that is wide open is the era in which the deduction does not exist. Recorded rather than wired, because 38 monthly payment files whose analytical value needs an FY2026 calculator the corpus does not hold is effort spent ahead of a use.

ofcc-projects: still declared, and partly by choice.

ofcc.ohio.gov returns 404 to this project’s contactable user-agent and 200 to a browser string. The data is served; the agent is filtered. Sending a browser string would work, and cache already states the position — impersonating one “would be discourteous besides”. The filter is more likely an undiscriminating CDN default than a considered exclusion of researchers, but guessing at intent is not a license to route around it.

Behind the filter the project portfolios are interactive maps rather than files, so even with access a bulk export would have to be reverse-engineered out of a map service. The honest next step is to ask the commission.

Eleven connectors wired, two declared. Both remaining blockers are now things a person could act on — get an OH|ID account, or write to the commission — rather than descriptions of a file format.

Send a browser user-agent to OFCC. It works. Rejected on the ground the repository already committed to for the Bureau of Labor Statistics, where the same question arose and was answered the other way round: BLS refuses an agent with no contact, and cache reads the contact from the environment rather than impersonating a browser. Having taken that position when it cost nothing, abandoning it when it costs something would make it decoration.

Create an OH|ID account and wire the legacy reports. Not rejected on principle — the account is free and the reports are public records. Rejected for now because a credentialed connector is a different kind of thing from the twelve here: it cannot be refreshed by anyone who clones the repository, and the digest manifest’s promise that a fixture can be regenerated from its source would quietly stop holding. If the deduct era becomes the subject rather than context, it is worth doing deliberately, with the credential requirement stated in the connector.

Wire the 38 open payment reports anyway. Rejected as building ahead of a use. Their value is a comparison between what the department projected before a year and what it paid during it, and that needs the matching year’s calculator. The corpus holds FY2027’s calculator and FY2026’s completed payments — the wrong pair.

Reverse-engineer the OFCC map service. Deferred and not attempted. It would mean inferring an endpoint from a page this project cannot fetch without changing how it identifies itself, which puts it behind the same decision as above.