Context
The record’s own argument (ADR-032) has three lines that each turned on whether a quantity was needed:
- Will-organization. The owner asked whether freedom of will comes in “gradations (perhaps even gradients/spectra?)”. The answer filed was a partial order by inclusion of what an agent can viably reach, with incomparable pairs, and a gradient only of a declared measure (CLAIM-354).
- Frustration. The owner’s correction, U37 of the exchange of 2026-10-17: “We can do quite a lot with access to an ordering, even if it doesn’t give us a true measure.” Frustration became a complex of jointly satisfiable demands whose relief is enlargement, with reorganizations that relieve different conflicts incomparable (CLAIM-347).
- Valence. The owner reframed a common currency as a negotiation admitting an ordering (CLAIM-440). Two reviews of the pull request that filed it found the measures confusing ordinal and cardinal structure: a Hodge gradient that reverses a comparison the wins settle, a netted record that erases alternation, a baseline valid only for complete tournaments, an approximate fit read as a Lyapunov function, and a sign read as determinate where the record fixes only a class (CLAIM-434, parts 4 to 11). The fix in each case was the ordinal statement first (TERM-077, TERM-078, TERM-076).
Nothing recorded the method as one. Each line found it separately, and the valence line found it by being reviewed.
Decision
For entries of the record’s own argument (CLAIM, TERM, QUESTION, CASE, ARG):
- Ordinal first. A comparison is stated as an ordering, a preorder or a partial order when that suffices for what the entry claims. A scalar is introduced only when the claim needs one.
- State what a measure is cardinal in. A numerical measure names what its differences mean (margins as intensities, costs, rates) and is checked against the ordinal facts it should respect. Where it can reverse one, as a least-squares fit can, the entry says so.
- Incomparability is a result. Where two things are incomparable under the declared order, the entry says they are, and does not force a ranking.
- Report the class. Where a record fixes only a class of orderings, potentials or representations, conclusions drawn from it are stated for the class, set-valued where its members disagree, and narrowed only by an intervention or further observation named in the entry.
- Scope. The rule binds the record’s own argument. Readings (LIT, NOTE, THEORY) report what their sources say, cardinal or not.
Alternatives considered
- Leave it to each claim. The status quo. It worked where the owner intervened and failed where nobody did: the valence line’s first measures were cardinal by default and needed two reviews to correct.
- Scale first, ordinal as a fallback. Most quantitative literatures default this way, and it makes measures easy to compute. It is the default that produced the errors above, because a scale supplies answers to comparisons the data never made.
- A principle (DP) instead of a decision. The record’s design principles concern what the record is for; this is a working rule about how its argument is stated, which is a decision’s business.
Consequences
- Existing entries are not swept. An entry is brought under the rule when it is next revised; CLAIM-434’s parts 4 to 11 and the three potential terms already meet it.
- A review can cite this decision against a measure introduced without its cardinality stated.
- It costs some determinacy: entries will more often say “partially identified” or “incomparable” than give a number. That is the intended cost.