Context
A line names a strand of the record’s own argument, so that it can be set aside as a whole (ADR-032). The vocabulary is closed, which means a line is one somebody named and blurbed. In the third synthesis pass (workbench of 2026-10-12), 102 Active or Proposed THEORYs still had no edge to any CLAIM or QUESTION. Most of them sat in two clusters that no line asked about:
- About eighteen readings in consciousness science and neuroscience: report
confounds, deficit arguments, crows reporting without laminar cortex,
overflow, and the case of AI. They ask what could show that a system has
experience. The nearest line,
organizational-affect, is set aside, and it is about a functional characterization (CLAIM-039) that explicitly brackets experience. - About twenty-five readings in learning dynamics, loss landscapes and
representation learning. They ask why trained networks have the structure
they have.
pragmatic-transportuses a few of them for identifiability (CLAIM-193, CLAIM-187). It is about what survives translation, not about training.
The assistant had at first treated the closed vocabulary as a reason not to propose lines. The owner pointed out that closedness only requires a blurb.
Decision
Two lines are added to luria.yaml:
evidence-for-experience: what could show that a system, organized, non-biological or without a cortex, has experience and not only access or function, and what each kind of evidence (report, deficit, architecture, behaviour) can and cannot establish.learned-structure: why trained networks come to have the structure they have, which features, symmetries and connected solution sets training reaches and why. The line covers understanding, not instruction for practice, which belongs to the anthology.
Their first members are the questions filed to place the unlinked THEORYs.
Alternatives considered
- Force the clusters onto existing lines. Rejected. Questions on
experience filed under
organizational-affectwould contradict that line’s bracketing of experience. Questions on training dynamics filed underpragmatic-transportwould stretch “what survives translation” until it named nothing. - Leave the readings unplaced. This was the assistant’s first recommendation for some of them. It costs nothing to file. But a reading no question asks for is one no later session will find by following the argument.
- One broader line for both clusters. Rejected. The clusters share no question. A line that holds both could be set aside only as a block of unrelated material.
Consequences
- The record now has eight lines. Neither new line has a claim yet. Their questions are open, and the readings answer them in part.
learned-structuresits near the anthology’s subject matter. A question or claim on it that would tell a practitioner what to do belongs in the anthology instead, by the boundary rule in CLAUDE.md.- A cluster that fits no line is now a prompt to propose one, with a blurb, for the owner to approve. It is not a reason to force a fit.