Reading the statute

ohio-laws had been Declared since genesis, blocked on “codes.ohio.gov serves HTML with no bulk export; section history is rendered rather than published as data.” Its connector note called it “the connector whose absence is most visible in the corpus itself”, because most statutory_basis fields were [open] and waiting on it.

Both clauses of the blocker are true. The first is a statement about the absence of a convenience, and it was read for the life of the corpus as a statement about the absence of the data. The Revised Code is server-rendered: a plain curl returns the operative text, the effective date, and the act that last amended the section.

This is the third stale blocker this repository has found by opening a file instead of reading its own note about the file — after the NCES crosswalk that had been “blocked since genesis” and turned out to be one column in a cataloged directory, and tax-abstract, whose blocker outlived the block in the other direction. The pattern is worth naming: a recorded blocker is a claim with a date on it, and this corpus has been treating them as facts.

Wire ohio-laws for the current text of named sections and nothing more.

Fourteen sources, one per section, chosen as exactly the sections some node’s statutory_basis points at. Named rather than crawled: a crawl of Chapter 3317 would pull three hundred sections the corpus has no use for and would churn the digest manifest on every unrelated amendment.

Add connect::html — a text extractor, not an HTML parser, because nothing here wants a tree. It drops <script> and <style> with their contents, treats block tags as line breaks and inline tags as transparent, and resolves the entities the site emits. The inline distinction is not cosmetic: a cross-reference is section <a>3317.011</a> of the Revised Code, and breaking on every tag splits one citation across three lines and makes it unsearchable.

Commit the extract as a record-oriented text fixture rather than CSV. Statutes are full of commas and write_csv does not quote.

Do not attempt a parameter series. The version archive is rendered per version rather than published as data, and it begins at 1 July 2014 — after the charge-off was already gone. The corpus takes the current text and its effective date and says so.

Every weight in the formula now has an independent check. All fourteen multiples match the statute to the last digit: six special education, three English learner, five career-technical.

That is a different check from the one already passing, and the difference matters. The weights were transcribed from the department’s spreadsheet and confirmed by reproducing the department’s own published amounts — which establishes that the department is self-consistent. A spreadsheet can be internally perfect and not be the law. Nothing in this repository could tell the difference until now.

Four open questions answered, because the statute publishes reasons and the calculator publishes quantities:

• the special education categories are named in R.C. 3317.013, in weight order, by cross-reference to Chapter 3323. The node said they were “in rule rather than in the calculator”; they are in statute;
• the English learner taper is keyed to time enrolled in United States schools and to attaining a proficient score — a schedule rather than a discretion, which is evidence against the reclassification-incentive reading and not proof of anything;
• the career-technical categories are program areas, not severities, so the drop from 0.5905 to 0.2154 is a gap between kinds of program rather than grades of one;
• the squaring in the disadvantaged-pupil index is R.C. 3317.02(I)(1)(a), which the DPIA node had recorded as “not located in statute here”.

One citation was simply wrong. The DPIA node cited R.C. 3317.029. There is no such section; the program is R.C. 3317.022(A)(4).

And one thing turned out not to be law at all. DPIA’s 65/35 blend of economically disadvantaged and directly certified ADM appears in no section. R.C. 3317.03(B)(21) delegates the definition to the department, constrained only by “a student shall not be categorically excluded … based on anything other than family income.” So the $422, the squaring and the statewide denominator are legislated to the digit and the blend is changeable without a bill — a distinction of durability the node had been presenting as uniform.

A correction to work committed hours earlier. The twenty-mill floor’s last untested candidate — that fixed-sum levies sit outside the reduction — had been tested against Table SD-1 that morning and called “disconfirmed as a general rule” on the strength of 155 districts landing on exactly 20.0000. R.C. 319.301(A)(1) says the reductions “do not apply to … taxes levied at whatever rate is required to produce a specified amount of tax money, including a tax levied under section 5705.199” — the emergency levy section. The exclusion is explicit law. The data showed only that it is not what puts those 155 districts on the floor. An absence of an effect in one group is not an absence of the mechanism, and the correction is recorded rather than quietly swapped because the wrong reading was the more natural one.

Also: R.C. 319.301 never says “twenty mills”. It says two per cent of the taxable value, with two-tenths of one per cent for joint vocational districts. The corpus’s figures are right and the name is a colloquialism a reader will not find in the statute.

A vintage risk that is new in kind. R.C. 319.301 is effective 20 March 2026 under two bills of the 136th General Assembly — Ohio changed its property tax reduction law weeks ago. The other thirteen sections read 30 September 2025, H.B. 96. A budget rewrites these at the same URLs, and the weight-verification test is what turns a silent divergence into a failing build.

Crawl Chapter 3317. Rejected: three hundred sections the corpus has no use for, a digest manifest that churns on every unrelated amendment, and no way to tell which sections anything depends on. The named list is self-documenting — it is the corpus’s own citations.

Use a third-party mirror with an API. Rejected. The General Assembly’s own site is the primary source; a mirror adds a party between the corpus and the law, and the thing being verified is precisely fidelity to the law.

Write a real HTML parser. Rejected as more than the task needs. A statute’s meaning is in its own numbering — (A), (1), (a) — which is text and survives tag-stripping. A tree would buy nothing and would be a much larger surface to keep correct.

Build the version series anyway, by fetching each rendered version page. Not rejected on principle, but it would not deliver what the corpus wants from it. The archive starts in 2014 and every genuinely open series question here — the charge-off rate era, when each proration factor was last set — is older than that. It stays with lsc-budget.