The four kinds of parameter

Decision the-open-item-audit produced a four-kind taxonomy of Ohio funding parameters — legislated, delegated, measured, uncodified — and explicitly declined to build it into the ontology:

Introduce the four-kind taxonomy as an ontology change — a kind field on parameter. Deferred, not rejected. It is the right shape and it would touch every parameter node; doing it in the same pass that discovered the distinction would mean committing to a classification derived from nine examples.

It also deferred a second notation change, separating “unknown” from “unentered” in the [open] marker, “for the same reason — it is a change to how the whole corpus is written, and this pass found the need rather than the design.”

Both deferrals were correct and both were about to be forgotten. Neither had a marker anywhere in the corpus; they existed only inside the alternatives block of one decision file, which is the least-read field of the least-read class of document in the repository.

The nine original examples were all weights and prices inside a single regime in a single biennium. The five parameter nodes are a different sample: one spans seventy years, one was retired by a regime change, one is set by an appropriation rather than by any formula.

Add kind to parameter.ont.yml as a text property, not an enum, and classify all five nodes.

Add [unentered] to the corpus’s claim vocabulary as a narrowing of [open] documented in .yidam/corpus/README.md rather than as a fourth confidence level in the prelude. The prelude’s three tags grade how well a claim is supported; this one says whether there is a claim at all, which is a different axis and not the vendored prelude’s business.

Count both, in a generated claim-audit block, so the inventory the audit produced by hand is produced by edfund-connect index from now on.

Text, not an enum, and the first node classified is why. base-cost-per-pupil was legislated before the Fair School Funding Plan — $5,283.00 in FY2006, a figure in an act — and is measured under it: R.C. 3317.011 states no base cost, it states a build-up from reported teacher salary and insurance inputs, and $8,241.61 falls out. The kind is a property of a parameter in a regime, not of a parameter. An enum would have forced a choice between two correct answers.

This also sharpens what the cost-input-refresh scenario is. The lever is which reference year is read; the thing it moves is a measurement. Read as legislated, the refresh looks like a policy that sets a number, when it is a policy that chooses a measurement.

One parameter the taxonomy cannot classify, and saying so is the useful output. appropriation-proration-factor is certainly uncodified — no section of R.C. 3317 sets it. It is either measured (the appropriation divided by the entitlement, nobody deciding) or delegated (the department choosing the factor, or which entitlements a line is prorated against), and the calculator cannot tell them apart: a factor being computable does not establish it was computed. The node had this as a vague [open]; the taxonomy turns it into a specific question with a named source, the appropriation act’s language, which is an lsc-budget retrieval rather than a dew-foundation one.

A fifth mode, found on the first parameter with a long span. local-share-charge-off-millage is legislated — R.C. 3317.022, moved three times by act between 1991 and 1997. But from FY2002 through FY2007 the General Assembly could not have changed it: DeRolph III ¶62 held it at 23 mills “irrespective of the language of R.C. 3317.012(D)(4) suggesting such a methodology”. The taxonomy asks what would have to happen for a value to change, and for six years the answer was a change in the law of the case. Not added as a category — one instance does not earn one — but recorded, because a taxonomy built from one biennium meeting a seventy-year parameter and finding a hole on the first attempt is the kind of thing that should be written down before the second attempt.

The notation split, measured. 153 unresolved marks: 128 live questions and 25 empty fields. The retagged clusters are series_path (8), established (7), typology (5), appropriation_line (3) and irn (2) — every one a value that exists in a source nobody has opened, not a question anybody has to think about.

vetoes (6) was on the audit’s empty-field list and stayed [open]. Its text reads “line-item vetoes exercised; education funding effects not yet assessed”, and assessing an effect is analysis. A distinction that inflates itself is not worth having.

Two fields split across both columns, which is the sign the line is drawn somewhere real rather than per-field: Eastland-Fairfield’s typology is genuinely open — JVSDs sit outside the department’s typology entirely and how they should be grouped for peer comparison is unresolved — while its five neighbors are missing a code that exists and is published.

An enum for kind, validated by the schema. Rejected on the first node classified, as above. The regime-boundary case is not an edge case in a corpus whose stated span is 1851 to the present; it is the normal case for any parameter that outlives a regime, and there are more of those coming than there are parameters currently in the corpus.

Add [unentered] to the prelude’s claim tags. Rejected. .vendor/prelude/ is vendored from upstream and a local edit would either be lost on the next sync or fork the vocabulary every yidam repository shares. The tag is also genuinely domain-shaped: it exists because this corpus deliberately holds exemplar nodes with metadata fields that bulk extraction never filled. A consumer that knows only the three prelude tags reads [unentered] as an unfilled field and loses nothing, which is the property that makes a local narrowing safe.

Retag the vetoes cluster too, and match the audit’s classification. Rejected. The audit grouped by “was this field ever populated” and the distinction being drawn is “does closing it require thought”. Those agree on four clusters and disagree on this one, and following the audit where it was reasoning about a different question would have put six analysis tasks in the typing column.

Do the classification and skip the generated count. Rejected, and this is the part with the most leverage. The audit’s own finding was that the [open] markers “are a record of what somebody did not get to, written at a moment, and never revisited.” A classification committed once has exactly that property. The count is now derived on every edfund-connect index run, so the next drift is a diff rather than a discovery.