The design
A4: “Set up multiple agents exposed to the same structured norms, categories, or interpretive training. Some populations interact; others acquire similar dispositions independently. Vary whether they possess any mechanism for collective decision-making or coordinated intervention.” Then estimate independently their shared patterns of interpretation, reciprocal influence, adaptive coordination, and response to joint perturbation.
What is measured
Whether the interacting populations show a relational organization the independently trained ones lack, in their response to a perturbation of some members, and whether that organization is present with no mechanism for collective decision.
What the outcomes would show, and what they would not
- CLAIM-418’s separation. A perspectival organization present where no collective agent is would show perspectival and agential boundaries coming apart in a constructed system.
- The risk the design must meet first. A4: “a shared perspective will simply mean correlated responses or socially learned norms. The study only helps the theory if it specifies, ahead of time, what additional relational organization distinguishes a perspective from those simpler phenomena.” The independently trained populations are the control for correlation; the defeater of CLAIM-417 is the control for the rest.
- Against the category. A4: “If it cannot make that distinction, the result should count against the theoretical usefulness of the category.” The record should keep that outcome possible.
- Related evidence. Honeybee swarms reach collective decisions with no central controller by cross-inhibition of recruitment (Seeley et al., LIT-1406); A5: “collective decision-making is not automatically collective interpretation.”
Note of 2026-10-29: the swarm’s stop signals, read
Seeley et al., LIT-1406, cited above under “Related evidence”, is now read (NOTE-1411). It supports the citation: nest-site scouts end rival scouts’ dances with stop signals, and no arbiter settles the choice. Two precisions.
- Targeted only while choosing. In the decision phase the signals go mostly to dancers for the other site (213 contra against 24 ipsi); once the swarm has chosen, they go to all dancers in proportion. The targeting result rests on two swarms in one island setup.
- Deadlock-breaking is a model result. That cross inhibition is what lets a swarm facing two equal sites pick one comes from the ODE model in the paper’s supplement, which was not read, and holds only above a critical rate of signalling. Whether the swarms observed were above it is not shown.
A5’s caution stands: the paper shows a collective decision, not a collective interpretation. The design is unchanged.