ADR-013: A work may be held in both records when each reads it for its own question

Context

ADR-001 set the boundary between the two records:

• a work goes wherever a topic can hold it;
• “nothing is filed in both”;
• a work that belongs on the other side moves there, as LIT-064 did.

That rule assumed a work has one question worth asking of it. Some works have two, one for each record:

• The anthology’s question is what a paper implies for ML practice, or what it explains about ML systems.
• Nucleation’s question is what the same paper means inside a line of mathematics, logic or retrieval theory that the anthology does not hold.

On 2026-09-27 the owner asked for exactly this case. Xiong’s The Lattice Representation Hypothesis of Large Language Models is held in the anthology (ANTH-LIT-460), read there for concept geometry. The owner wants it re-read here against van Rijsbergen’s The Geometry of Information Retrieval (LIT-262): formal concept analysis, Galois connections and lattices of classes are that book’s second chapter.

Decision

A work may hold a LIT in each record when each record reads it for a question of its own.

• Each record’s entry is its own reading. Each has its own NOTE, status, tags and relations, filed by that record’s rules. A work can be Active in one and Proposed in the other; the statuses answer different questions.
• The entries name each other. The nucleation LIT cites the anthology entry by code (ANTH-LIT-…), in its body and in its first history note. The anthology names nucleation works in prose, as it already does.
• Disagreements are reported, not fixed across the boundary. If a nucleation reading finds the anthology’s entry wrong, the nucleation NOTE says so. The fix is then made in the anthology by that record’s process, not edited from here.
• The default is unchanged. A work still goes to one record. A second entry needs a reason: a question the first record’s reading does not ask. The curation entry that files the second entry states that reason.
• Moving is still moving. When a work simply belongs on the other side, it moves and the old entry is Superseded, as LIT-064 was. Dual holding is for two readings, not for hedging about where a work belongs.

ADR-001’s “nothing is filed in both” is narrowed to “nothing is filed in both without its own reading”.

Alternatives

• Keep one entry and cite across. This is the old rule. It fails when the second question needs a full reading, with its own NOTE, relations to this record’s works, and THEORY sources, that the other record’s LIT cannot carry.
• Move the work. That loses the first reading’s standing in its own record, where it is doing work.
• Mirror everything. Rejected in ADR-001 for the reason it gave: it would invite filing in the wrong place, and two readings of one work are only worth having when they ask different things.