Context
ADR-001 set a two-part boundary between this record and the Anthology of the SOTA:
- (a) Topic. A work goes wherever a topic can hold it. If an anthology topic can hold it, it goes there and not here.
- (b) Practice. Material that carries an instruction for machine-learning practice belongs in the anthology by definition.
ADR-013 narrowed (a) so that a work could be held in both records when each reads it for a question of its own. Even so, every filing here began with a check of the anthology, and dual holding needed a second question, stated in a curation entry. ADR-005’s flag marked works whose placement was in doubt.
On 2026-10-19 the owner asked for the record’s gap around the rubber hand illusion to be filled: Botvinick & Cohen 1998 and the bodily self-consciousness literature around it. The owner said to assume the anthology does not hold these works, and that the boundary rule should go. The check had cost a clone and a grep for each filing. It had also put nucleation in charge of keeping two records from overlapping, though only the anthology can say what it holds and why.
Decision
- (a) is dropped here. Nucleation files any work the owner wants here, whether or not an anthology topic could hold it. Nothing is checked against the anthology before filing.
- (b) stands. An instruction for machine-learning practice still belongs to the anthology, so the anthology keeps a distinct job. The topic blurbs that send ML practice to the anthology (learning theory, compositionality, databases, learned structure and others) are unchanged.
- Cross-reference is optional. When a filer knows the anthology’s code
for the same work, the LIT cites it (
ANTH-LIT-…). Not knowing it is no defect, and no history note need record a search for it. - Overlap is the anthology’s question. Whether a work filed here is also held there is for the anthology to find out and reconcile, by its own process. Nucleation does not police it.
- The anthology-candidate flag is still applied (ADR-005), after the
topics and never first. It now means that someone thinks the anthology may
want this work as well. It no longer marks a work whose filing here is in
doubt. A transfer, when decided, is still a supersession, as with
LIT-064.
ADR-013 is subsumed. A work may be held in both records without a second question being named. When nucleation reads a work for a question the anthology’s reading does not ask, saying so in the curation entry is still good practice, but it is not a condition of filing.
Alternatives
- Drop (b) as well, and file ML-practice works here too. Rejected by the owner. It would leave the anthology without a job of its own and turn nucleation into a mirror, the alternative ADR-001 rejected.
- Keep (a) and file these works under ADR-013. That would mean a second question and an anthology check for every borderline work. The check is the cost the owner asked to stop paying.
- Retire the anthology-candidate flag. Without (a), the flag no longer records doubt about where a work belongs. The owner kept it, because it still tells the anthology which works here it may want.
Consequences
- CLAUDE.md’s boundary section, the topic vocabulary’s blurb and alert in
luria.yaml, and the README’s opening are rewritten to say what is filed here, not what is excluded from it. - ADR-001, ADR-005 and ADR-013 each take a history note pointing here. Their bodies are unchanged.
- Existing LITs whose history notes record an anthology search keep them. They were true when written.
- The anthology may find works filed here that it also holds. Reconciling them is its business.