Advocacy literature

The question put was whether this repository had integrated any of the ALEC or Heritage Foundation literature that pushed for school choice. It had not — no mention of either organization anywhere, the apparent hits being school names in the report-card fixtures, Hamilton County in the DeRolph record, and Ohio’s own EdChoice program rather than EdChoice the organization.

The gap was not accidental, and two structural facts explain it. The actor class enumerates institution_type as legislature | court | executive-agency | commission | legislative-agency and describes an actor as an institution that enacts law, decides cases, administers programs, or publishes the data this corpus is grounded in — an advocacy organization does none of those four. And legislation means enacted: bill designation, numbered General Assembly, signing date, effective date, line-item vetoes. A model bill has none of those. So neither the organizations nor their model statutes had anywhere to land.

One precedent already existed and turned out to decide most of the design. Fordham Institute — Ohio Base Cost Model Commentary is a school-choice-aligned organization’s policy critique, held as a catalog record with its alignment stated in the record — “Fordham is a school-choice-aligned organization and its critique of the base cost model should be read knowing that” — cited from two corpus nodes, with no node of its own, no connector, and no digest. That is a working answer to most of this question, arrived at before the question was asked.

The retrieval constraint shaped the rest. Four documents were read in full before anything was written, because a corpus whose value is one-hop traceability cannot absorb a summary of a document nobody opened. ALEC’s Parental Choice Scholarship Program Act (Universal Eligibility) and Education Savings Account Act came back as full provision text. ALEC’s Index of State Education Freedom (2025) and Heritage’s Education Freedom Report Card (2023) came back as static PDFs and were read directly. Heritage’s prose reports and its interactive report card return HTTP 403 to anything identifying itself as a program — the same failure mode as one of the two already-blocked connectors — so the report card’s category weights are unretrieved and recorded as [open] rather than guessed.

One new class: model-policy. A template statute published by a national organization for adoption by state legislatures, and not law anywhere. A UFO Kind, on the same reasoning legislation is one: identity fixed at the version the publisher finalized, amendment producing a new dated version. Seven properties, of which district_funding_effect and adoption_evidence carry the weight.

Two instances, both ALEC model bills. parental-choice-scholarship-act and education-savings-account-act.

Four catalog records, each stating its publisher’s alignment in the record, on Fordham’s footing — freely available, retrieved by hand, versioned by publication date, no connector, no digest.

No corpus node for any advocacy organization. The publisher is a property here and a named alignment in the catalog record. Fordham is the precedent and it is a good one.

adoption_evidence is a mandatory field, [open] on every node in the class. No source retrieved connects either text to any Ohio bill. Reporting on the Ohio proposal that most resembles the ESA model — H.B. 11 of the 135th General Assembly, the “Backpack Bill”, which did not pass and is therefore not a corpus node — describes it as modeled on seven other states’ enacted programs and does not name ALEC.

The admission test for a node, stated in the class README. Not influence, which this repository cannot measure. A model policy earns a node when it states a mechanism Ohio’s enacted arrangement can be checked against. Both instances qualify on the same field, and it is the field that makes this more than a bibliography: each specifies that the award is subtracted from the resident district’s state aid, and this corpus has already established — negatively, exhaustively, against the department’s own FY2027 calculator, pinned in crates/project/tests/the_voucher_channel_is_absent.rs — that Ohio’s post-FSFP scholarship channel does not work that way. Ohio’s eligibility design converged on the model while its funding mechanism diverged from it, and both halves are verified on both sides.

Edge verbs chosen to survive the provenance being open. parallels and departs-from, never template-for or adopted-as. A relationship that would become false if adoption_evidence is never filled is a relationship asserting more than is known.

The corpus can now hold the advocacy literature without importing its positions, and the strongest thing it gained is a falsifiable comparison rather than a reading list: the model bills specify a deduct from the resident district, Ohio’s original EdChoice did that, and Ohio’s post-FSFP arrangement verifiably does not.

A pre-existing contradiction in program/edchoice-expansion surfaced and was resolved. Its description called the FSFP mechanism question “the highest-priority factual gap in this node” and marked it [open], while its own mechanism property recorded the answer as [verified] with a test pinning it. The property was right and the description was stale. Two new nodes were about to be built on the verified side of a claim the same node called open, which is how the contradiction was noticed at all.

The class starts with a known asymmetry that is not an omission. Heritage publishes no model bills, so its literature is catalog records cited from the Ohio nodes they bear on and has no instance in the class.

Two things are now on the record about what advocacy scorecards measure, both verbatim from the documents. ALEC’s index has no spending category at all — its 100 points score program availability, nothing else. Heritage’s return-on-investment category, where Ohio ranks 47th, treats lower per-pupil spending as better, in its own words to another state: “Iowa can improve its ROI ranking by lowering per-pupil spending”. Ohio’s 47th is therefore substantially a statement that Ohio spends a lot, not an efficiency finding, and a reader taking the four category ranks as four independent assessments would have that one backwards. Heritage’s per-pupil denominator also includes capital expenditure and debt interest, which every per-pupil operating figure in this repository excludes by declared scope, so the two are not substitutable.

Highest-value next read, recorded as a gap. ALEC publishes a third variant, Parental Choice Scholarship Program Act (Universal Eligibility, Means-Tested Scholarship Amount), whose title describes Ohio’s enacted EdChoice expansion exactly. Its text is unretrieved. open

The one method that could move adoption_evidence is a textual diff of a model against the enacted Ohio section, which regime-diff already does for successive Ohio statutes. Nothing points it at a model bill yet, and until something does, the field stays open rather than being filled by inference.

Widen actor to admit advocacy organizations. Rejected. It looked smallest — one enum value — but the class description would have had to be rewritten too, since an advocacy organization does not enact, decide, administer, or publish the data this corpus rests on. The cost is worse than the edit: government actors are authorities whose publications are primary sources, and merging the two makes the graph unable to distinguish “the department says X” from “an advocacy group says X” at the class level. In a corpus whose README opens on provenance weight, that is the wrong thing to lose.

A second new class for advocacy organizations. Rejected as redundant once Fordham’s precedent was read properly. An organization’s position in this graph is the publisher of a source, which a catalog record already encodes.

Put the model bills in legislation with an enacted: false property. Rejected outright. It inverts the class’s meaning to accommodate two documents and invites exactly the error the separation prevents — a reader taking a template for a statute.

Catalog records only, no class at all. The genuinely smallest option, and rejected because it cannot express the finding. The deduct comparison is a mechanism claim linking a model provision to a verified Ohio fact, and a catalog record has nowhere to put an edge.

Omit adoption_evidence when the answer is nothing. Rejected. A class that could record a resemblance but had nowhere to record the absence of provenance would read as an accusation.