Context
ADR-032 made TERM the record’s concept construction. A changed definition
becomes a new version or a superseding TERM, and claims point at the
definitions they are stated in through uses:. On 2026-10-11 the concept
pass found the day’s load-bearing words (“self”, “agent”, “combined form”)
used across some twenty claims with almost no TERM links. The sweep that
followed left the older essay claims’ text alone, because their “agent” is
the paper’s term of art. The owner answered with the rule this decision
records.
Decision
- Cite at the point of use. When an entry’s body invokes a concept as a
TERM defines it, the bare TERM code follows its first use in that body.
luria link --fixspells the target. - Leave other senses unlinked. A word used in another sense stays bare: a reading’s term, the will paper’s supposed “agent” (CLAIM-080), or an ordinary-language use. The absence of a code is then information, not an omission. Where the difference matters, the TERM’s “what it is not” section says how the other sense relates. TERM-044 does this for “agent”.
uses:is narrower than citation. A claim lists inuses:only the TERMs it is stated in. An inline citation that discusses a term, as its history or as a contrast, is a citation and not a use. The sweep’s check found four such cases and kept them out ofuses:(CLAIM-004, CLAIM-055, CLAIM-139, CLAIM-142).- Quotations are never altered. A code for a term that occurs inside a quotation goes outside it, or not at all.
- Terms with several versions are linked by hand. “Fidelity” (eight Active versions), “frame” (five) and “communicative object” (two) need a judgement about which version a passage means. They are linked when an entry is written or revised, not by sweep.
- Adding a citation is not a revision. An inline TERM code changes no
wording, like the targets
luria link --fixwrites, so it takes no version bump. Changing the wording to state a sense does take one, as in CLAIM-169 and CLAIM-172. - The field and the prose are held together by luria, not by sweep.
CLAIM
usesdeclaresexplain: citedinluria.yaml: every TERM in a claim’suses:must be cited in its body, or the lint reports the row asunexplained-relations. A use is written where the claim is stated in the term, as[[uses::TERM-…]];luria link --fixexpands it to a citation and pushes the code up intouses:. Where no point of use can carry the code, the body states why in aref::uses-file:line: the term is named only inside a quotation, the claim applies it without naming it, or the TERM has since been superseded and the claim was stated in the definition current when it was written.
Alternatives
uses:only. The relation exists, but a reader of the prose cannot see which occurrence of a word is the defined one.- Link every occurrence. That is noise. The first use per entry fixes the sense for the rest.
- Rewrite older claims into the new vocabulary. That changes what they said. Citing where they already invoke a defined concept does not.
Consequences
- The first sweep (2026-10-11) used a script that matches distinctive
TERM phrases outside quotations, headings, comments and tables. It
proposed 26 sites. One was a false positive, the line name
“will-organization”, and was excluded. The other 25 were read and
linked. Thirteen
uses:entries followed, after four proposed ones were refused on reading. - The norm is prose until it is a mechanism (DP-3). Version 2
found that half of it already was. The
uses:→prose direction is luria’s existing pushdown (explain:) and pushup ([[field::CODE]]), and needed only a line of configuration. Turning it on reported 70 rows across the claims; each was cleared by a citation at the point of use or by aref::uses-file:statement with its reason. The other half, “a TERM’s title phrase used in an entry that neither cites nor lists it”, has no field to declare and is filed upstream as dmarx/luria#370. Until luria has it, the phrase sweep is re-run by hand when TERMs are added.