The claim
The owner, verbatim: “let’s take up the agency question. I think it’s probably worth differentiating between systems where ‘who drives?’ has an answer wrt a particularly level of description or not. I think there’s is probably a spectrum/gradient/distribution of kinds here. for two level systems, we have systems that experience fully top-down causation, systems where there is bidirectional causation, systems where there driving ‘will’ is emergent from the system as a whole, etc.”
How the record grounds it
The kinds are separable by quantities the record already reads:
- The information decomposition (LIT-025). A macro-variable’s synergy splits into downward causation, where it predicts parts’ futures, and causal decoupling, where it predicts its own future. Zero synergy is the aggregate end.
- Proportional causation (THEORY-133). The higher level is the cause, rather than its realizer, exactly when the effect is insensitive to how the higher level is realized.
TERM-051 arranges these as a drive profile. Because both quantities are continuous, the owner’s “spectrum/gradient/distribution” is the natural reading. The named kinds are regions, not classes.
One refinement the record adds. Read strictly, “fully top-down” cannot mean that the higher level drives without its parts, because it is realized by them. The defensible reading is THEORY-133’s: the parts’ detail is irrelevant given the higher level’s state. That is graded, and it can be checked.
What it does not say
- That any particular system sits where intuition puts it. That is measurement, relative to a chosen macro-variable.
- That a driving level is an agent.
Note of 2026-10-30: what holds it back
Kept Proposed by the eighteenth status review. No system in the record has had its drive profile estimated: LIT-025 reports its applications without values and restates a primary the record has not read, so that systems in fact differ, rather than that the formalism allows them to, is not shown. For three or more parts the D and G terms that separate the kinds depend on which axiom the redundancy function drops (CLAIM-343, Active), and the graded reading of “fully top-down” extends THEORY-133, whose criterion is an “exactly when”, without a measure; the claim says neither.