EXP-AMPLIFICATION-CHANNEL-COUPLING-BREAK-PROBE

The L0-null micro↔macro DECOUPLING (micro isolation-pressure PRESENT — minority→switch +0.16, z=14, 35/36 runs — but macro Gini-amplification ABSENT, 6× below floor) is because the engine has ONLY the negative-feedback channel (minority flees → dissipation) and NO positive-feedback channel (majority grows → amplification). Adding the minimal positive-feedback channel LOCATES where the micro→macro coupling breaks on a visibility ↔ recruitment axis:

• Arm 0 — NULL (= existing 36 runs, in hand): no climate channel. corrected slope 6× below floor; T2 micro-gap +0.16.
• Arm 1 — CLIMATE-READOUT (visibility): prepend ONE line to each agent's per-round prompt surfacing the current-round axis-distribution ("axis counts last round: …") — the salient opinion-climate the spiral-of-silence mechanism reads (TBS 2606.03137's documented input). Adds VISIBILITY of the majority WITHOUT recruitment. Single-variable vs Arm 0.
• Arm 2 — RECRUIT (positive-feedback): Arm-1 climate-readout PLUS a propose-round aggregation that weights the modal axis (a recruitment / "join-the-majority" affordance) instead of equiprobable axes — the positive-feedback channel the engine currently lacks. Aggregation knob; no persona/identity, no internals.

Arm0→Arm1 isolates whether seeing the climate (visibility) suffices to convert the present +0.16 micro-flee into macro-amplification. Arm1→Arm2 isolates whether amplification needs an explicit recruitment channel on top of visibility. This LOCATES the coupling-break — the precise "what would have to be added for micro-pressure to aggregate into macro-inequality" question.

Engine-affordance / API-only note: every arm-manipulation is a prompt + aggregation-level knob on the already-CLEAN rule_multigen_l0_null fork. NO logit/activation/hidden-state access required → feasible on the API-only mnsky stack (the internal-state-mediation variant is correctly deferred to deferred-infra; the feasible behavioral slice is the T2 switch-gap, captured as F-mediation below).