ADR-040: A formal-apparatus line, held as a second line by the general results that a substantive line borrows, so that setting that line aside does not take them with it

Context

A line exists “so that a line can be set aside as a whole and what depends on it can be found” (ADR-032). Setting a line aside is a status change on its members. The argument page (docs/argument-lines.md, chains over rests_on with line as the invariant) shows that this no longer works for pragmatic-transport.

These are the counts from record/claims.d on 2026-10-12. Superseded and Rejected claims are included only in the line totals.

• pragmatic-transport is most of the argument. It holds 188 of 262 claims and 113 of the 153 rests_on edges. One component of 99 claims, 83 of them in pragmatic-transport, fills most of the page. 128 claims have no rests_on edge at all and do not appear on the page.
• The other lines rest on it for general results. Of the 11 rests_on edges that cross lines, 9 end in pragmatic-transport, and most of those end in results that do not mention translation:
  • CLAIM-198 (will-organization) and CLAIM-196 (distributed-agency) rest on CLAIM-197, that sections over disjoint contexts always glue.
  • CLAIM-189, 190 and 191 (nucleating-argument) rest on CLAIM-056 (error propagation), 081 (retention against drift), 142 (observational equivalence is relative to probes) and 149 (a training probe cannot certify).
  • Only CLAIM-189 → CLAIM-034 (fidelity is one goal among several) and CLAIM-191 → CLAIM-002 (a text’s identity across retellings) depend on the account of translation itself. Both edges are genuine: the record is a chain of retellings.
• So setting the line aside takes too much. Retracting pragmatic-transport today would retire CLAIM-197, a lemma about sheaves that holds whatever the account of translation concludes. With it would go one premise each of CLAIM-198 and CLAIM-196, and four of the premises of nucleating-argument.
• distributed-agency shows the same pattern, smaller. Its chain CLAIM-152 → CLAIM-154 / CLAIM-164 is about individuation in general: a sufficiency partition relative to a target, the threshold at which a unit is worth naming, and what a spectral gap makes definite. It is borrowed by the line’s own claims about combined forms.

The schema allows a fix that keeps every claim where it was filed. luria reads a vocabulary field as a list when the scheme declares it many: true (luria/contract.py, values_of). A list in a field without it is a lint finding: “holds N values, but the CLAIM scheme declares it one lines value — set many = true there if it should hold several”. The chains already treat a scalar as a set of one, and they head a chain with every value its members hold in common (luria/invariants.py, shared; luria/chains.py, _sections). The restriction is this record’s configuration, not luria’s.

Decision

Approved by the owner on 2026-10-12 (“approved, go ahead with the yaml change and group A”). Items 1, 2 and the group A values are made in the same contribution as this decision. The group B splits followed on 2026-10-13, at the owner’s request (“go ahead with the group B splits”). Group C was decided the same day, as recommended (“move forward as recommended”).

1. Allow a claim more than one line. Add many: true to the line field of the CLAIM scheme in luria.yaml. It stays required, and its vocabulary stays closed. The other schemes keep one line until a case asks otherwise.
2. Add the line formal-apparatus, blurbed: results that hold whatever any substantive line concludes: gluing and contextuality, comparison of experiments and probes, identifiability, conservation, and thresholds for naming a unit. Mathematics and method that the other lines borrow, held as a second line so that setting a substantive line aside does not retire them.
3. Set-aside rule for claims with two lines. Setting a line aside retires the members that hold no other line still standing. A claim that also holds formal-apparatus stays standing. It is retired only on its own merits, by refutation or supersession, never with a line.
4. The test for the second value. Would the claim keep its truth value if every thesis of every substantive line were retracted? And does it rest on no claim that would not? Both must hold. The second condition is the one that matters in practice: a claim that holds formal-apparatus while resting on a substantive claim would survive the retraction of its own premise.
5. The claims, sorted by that test (below). Only group A takes the value. Group B is split, and group C is the owner’s call.

A. Add formal-apparatus (24 claims, none resting on any claim)

From pragmatic-transport:

• CLAIM-037: formal contextuality is failure of global extension, not context dependence.
• CLAIM-040: invariance under transformation yields conservation (Noether).
• CLAIM-041: order effects and noncommuting operations do not by themselves establish quantum contextuality.
• CLAIM-043: equal entropy does not entail equal relational structure.
• CLAIM-044: ambiguity and contextuality are different phenomena.
• CLAIM-069: Contextuality-by-Default and multimaximal couplings.
• CLAIM-070: a global kernel carries a global extension to a global extension.
• CLAIM-093: an invariance need not give a Noether law.
• CLAIM-094: a harmonic observable conserves only its mean.
• CLAIM-107: harmonicity as a Noether-type conservation criterion.
• CLAIM-121: kernels that commute with restriction preserve overlap-consistency.
• CLAIM-131: a compatible family of the event sheaf always glues.
• CLAIM-132: characters classify representations up to isomorphism, in characteristic zero.
• CLAIM-141: causal-loop consistency and sheaf contextuality behave oppositely under distributions.
• CLAIM-144: inertial frames differ in coordinates, not in a failure to glue.
• CLAIM-148: a matrix factorization is identified only up to a gauge.
• CLAIM-197: sections over disjoint contexts always glue. CLAIM-198 and CLAIM-196 rest on it.
• CLAIM-209: probes recover only the distinctions their evidence identifies.
• CLAIM-245: context-wise recoding and per-content signalling under Contextuality-by-Default.
• CLAIM-260: approximate transports compose into a category only under a triangle inequality.

From distributed-agency:

• CLAIM-152: individuation by sufficiency is relative to its target.
• CLAIM-154: reify when d′² exceeds the log variance ratio.
• CLAIM-164: a spectral gap makes definite a subspace, not a direction.
• CLAIM-188: the Gaussian Occam term misprices singular models. This is the counter to CLAIM-154, and it travels with it.

Four of these state the general result in the vocabulary of communication: CLAIM-044 (“an ambiguous utterance”), 107 (“a pragmatic observable”), 131 (“a communicative object”) and 209 (“communicative decisions”). Moving them would have meant restating their titles. Adding a value keeps them in pragmatic-transport, where that wording belongs, so no title changes.

B. Split, do not tag (7 claims)

These are general in content, but each is filed resting on a thesis about translation, directly or through another member of this group:

• CLAIM-056 (error propagation) rests on CLAIM-106 (fidelity is graded).
• CLAIM-081 (retention against drift) rests on CLAIM-056.
• CLAIM-142 (observational equivalence is relative to probes) rests on CLAIM-046 and CLAIM-011.
• CLAIM-143 (disagreement is not incompatibility until the correspondence is fixed) rests on CLAIM-011, 004 and 035.
• CLAIM-149 (a training probe cannot certify) rests on CLAIM-011.
• CLAIM-193 (a linear probe family sees only normalized kernels) rests on CLAIM-142.
• CLAIM-126 (what model stitching certifies) rests on CLAIM-050.

Tagging them would break the test above. For each one, a general claim is filed in formal-apparatus alone and rests on nothing. The existing claim keeps its line and gains rests_on the general one. Claims in other lines that cite the general content re-point to the new claim:

• CLAIM-189 and CLAIM-191 to the general forms of 081 and 056;
• CLAIM-190 to the general forms of 142 and 149.

Done on 2026-10-13. The general forms are CLAIM-267 (from 056), CLAIM-270 (081), CLAIM-264 (142), CLAIM-265 (143), CLAIM-266 (149), CLAIM-269 (193) and CLAIM-268 (126). Each holds only formal-apparatus, rests on no claim, and is Proposed like its original. CLAIM-189, 190 and 191 re-point to them. The originals keep their bodies and add a dated note.

C. The owner’s call (7 claims)

Decided on 2026-10-13, as the assistant recommended:

• Tagged formal-apparatus: CLAIM-011, CLAIM-079 and CLAIM-123.
• Split: CLAIM-187, into CLAIM-263 in learned-structure. That is the protocol for trained networks, and the first claim in that line. CLAIM-187 keeps the carry-over to communicative kinds, and its premise re-points from CLAIM-142 to CLAIM-264.
• Left as they are: CLAIM-059, CLAIM-213 and CLAIM-237. 059’s general content is in CLAIM-267. 213 is an unadopted position, not a result. 237’s lemma is needed by no other line.

The candidates as first sorted:

• CLAIM-011 (an uninformative correspondence makes intertwining vacuous). It is general, but it is the premise most of pragmatic-transport rests on. Tagging it would make the line’s own foundation outlive the line.
• CLAIM-059 (three regimes of repeated reconstruction). The dynamics are general, but the claim is framed for the line’s research programme.
• CLAIM-079 (stability does not entail viability, will-organization). It is a general claim about dynamical systems, and nucleating-argument’s CLAIM-195 rests on it. It is also a thesis of the will paper.
• CLAIM-123 (conditional generation can meet each condition and fail their conjunction). It may belong to learned-structure rather than here.
• CLAIM-187 (comparing computation by quotienting the symmetry group). ADR-039 named it, with CLAIM-193, as nearer learned-structure.
• CLAIM-213 (a relational kind is a hypothesis about invariance). It is general philosophy of science, filed as advice from an interlocutor.
• CLAIM-237 (the first open problem of directed transport is not yet posed). The result is about menus of experiments, but it is stated as the line’s own open problem.

Alternatives considered

• Move the claims instead of adding a value. This was the assistant’s first draft. Rejected, at the owner’s suggestion:
  • It takes general results out of the chains that use them. The pragmatic-transport page would lose the lemmas its arguments run through.
  • It forces title changes on the four claims worded for communication.
  • It makes line say where a claim ended up rather than what it is about.
• many: true with no new line. For example, CLAIM-197 would hold pragmatic-transport, will-organization and distributed-agency. Rejected:
  • A claim’s lines would then record who borrows it, and grow with every new borrower.
  • Retraction would depend on how many lines happen to cite a claim, not on what the claim is.
• Leave the lines as they are. Rejected: the set-aside that ADR-032 makes the purpose of a line would retire theorems along with the theses that used them.
• Split pragmatic-transport into several lines. For example, the account of translation, its formal apparatus, and the audit of its manuscript (CLAIM-202, 211, 216–244 and others). The formal part of that is this decision. The audit is left where it is: if the line is set aside, its audit should go with it.

Consequences

• What a retraction takes. Retracting pragmatic-transport would retire 144 of its 164 standing claims, not all of them. CLAIM-198 and CLAIM-196 keep their premise. After the group B splits, nucleating-argument loses only CLAIM-034 and CLAIM-002. Those are its real dependence on the account of retelling.
• The argument page will not show fewer crossings. A step crosses when the two claims share no line, and CLAIM-198 (will-organization) still shares none with CLAIM-197 (pragmatic-transport, formal-apparatus). The crossing report keeps showing every borrowing. What changes is that a borrowing from formal-apparatus can no longer be lost to a retraction. If that noise proves unwanted, the report could treat a step into formal-apparatus as a non-crossing. That would be a change to luria and is not proposed here.
• Claims list under both lines. The claims index groups by line (axis: line), so a claim with two values appears under both headings. This is how luria already renders plural fields.
• Side observation. CLAIM-160 and CLAIM-171 (what data written for a future reader means) are filed in pragmatic-transport but bear on nucleating-argument’s subject. With plural lines they could hold both. That is not proposed here.
• Order of work, once approved:
  1. the luria.yaml change (field and line);
  2. the group A values;
  3. the group B splits;
  4. the owner’s calls on group C.