ADR-001: Nucleation holds what the anthology's topics cannot: LIT, NOTE and THEORY, with a seeded closed vocabulary

Context

The anthology’s reading-time triage of 2026-09-25 (ANTH-ADR-052 governs its topics) found that 47% of the reading time in the ten-minutes-or-more group went to 34 works no anthology topic could hold. The heaviest clusters were quantum foundations with category theory, network science, and complex systems. The owner kept ANTH-ADR-052’s decision not to add topics for them. The owner also decided that “out of scope there” should not mean “unrecorded”, and named this repository as the catchall.

Decision

• Schemes: LIT, NOTE and THEORY, copied from the anthology’s luria.yaml, with ADR, DP, a changelog and a curation journal. The copy is deliberate. A work that turns out to belong in the anthology, or an anthology note that turns out to belong here, moves as a file and needs no reshaping.
• No SOTA practice scheme, and no consensus vocabulary. The practice scheme exists to record what someone should do in ML. Material that carries such an instruction belongs in the anthology by definition. THEORY loses its explains: relation, which pointed at practices.
• A closed topic vocabulary, seeded from the triage’s clusters: quantum-foundations, mathematics, network-science, complex-systems, probabilistic-modeling, neuroscience, natural-sciences, operations-research, society-and-governance. Closed for the anthology’s reason: every tag is one somebody chose and blurbed. The alert says so. A work that needs a word the list lacks gets the word added by decision, not the nearest wrong one.
• The boundary. A work goes here if no anthology topic can hold it. If one can, it goes there. The test is the anthology’s own (its ANTH-ADR-059): the topics follow the content.
• Cross-reference, not duplication. An ANTH remote lets a document here cite the anthology (ANTH-LIT-285), and codes are not shared. Nothing is filed in both.

Alternatives

• Add the topics to the anthology. ANTH-ADR-052 counted the candidates and rejected them. The owner kept that decision, and this record is the alternative it leaves room for.
• A plain markdown catalogue. Lower ceremony, but no lint, no relations, no generated indexes, and no path for a work to move between the records.
• A full mirror including SOTA. It would sit mostly empty, and it would invite filing ML practice in the wrong place.

Consequences

• The seeded vocabulary is provisional in the way every closed list is. The first works filed will test it, and a misfit is a finding about the list.
• Do not run luria remotes --refresh for ANTH. Under luria 0.31.0, discovery reads the anthology’s luria.yaml and indexes only its first scheme (practices). After that, every LIT, THEORY and ADR citation reads as “NO SUCH DOCUMENT”. Without a lockfile, the code-only filename convention resolves ANTH-LIT-285 to the right file. That is a luria limitation, and it is worth an upstream issue.
• The depth of each filing is left to the curation log to state per batch. NOTE’s Read and Skimmed statuses already say how deeply something was read, so no new mechanism is needed.