The case
A97: “Condition A, Shared complicity: Two friends sit together with an empty wallet between them. Both are amused by a weekend of overspending. Their expressions convey mutual understanding, not condemnation. Condition B, Normative authority: Two people sit together with an empty wallet between them. One sternly admonishes the other for overspending. Their expressions convey judgment and embarrassment. Condition C, Underspecified relation: Two people sit together with an empty wallet between them after a weekend of overspending.” Observation contexts: objects and actions; affect and alignment; role and authority; interactional meaning; plausible continuation.
What it can show
That “These prompts overlap substantially in concrete scene elements but prescribe different interpersonal configurations”, so object-level evaluation cannot tell a faithful image from one that reverses the relation (CLAIM-013, CLAIM-123). Manuscript §7 keeps A and B: “two friends sharing amused complicity and … an adviser criticizing someone may describe nearly identical visible objects”. Without condition C there is no measure of which relation a generator supplies when none is specified.
Note of 2026-10-10: a C₂ model, and the laboratory
A198 §III formalized this scene as a role swap. Two participants and “an interaction concerning overspending”, with basis e₁ = a→b and e₂ = b→a, where the arrow “means only ‘occupies the active evaluative position relative to.’” The group C₂ swaps the participants, and the character is (2, 0). That is the character of any free action of C₂ on two points, so it carries no communicative content. The contrast between complicity and reprimand sits in the stabilizers: the swap moves the one-way configuration and fixes the mutual one (CLAIM-132). The model, like the case, has no counterpart of condition C.
Its conditions A and B are the spending row of the topic × act laboratory (CASE-040).