Agent-to-agent (A2A) alpha discovery is slowed in contemporary LLM multi‑agent systems by repeated feedback cycles between mining and evaluation agents, where hand‑offs consist of free‑form natural‑language messages that lack a stable contract and cannot be replayed. We restructure this communication as a structured A2A protocol of typed, causally addressable, unicast records, so the committed stream forms a causal trajectory.
Along that trajectory a single predictor equipped with four typed heads forecasts the accumulated guidance the two miners would receive several cycles ahead; a transactional verify‑leap controller then commits a multi‑cycle speculative outcome only when it passes a four‑level gate, otherwise it rolls back to the exact prior state.
The contribution lies in the structure itself, not in accuracy. A controlled ablation shows that an equal‑information free‑text channel attains the same predictor hit rate, indicating that typing provides a state that can be schema‑checked, replayed deterministically, and prevented by construction from leaking a forecast to an evaluator—auditability by construction rather than an empirically stress‑tested guarantee.
On a CSI~1000 out‑of‑sample holdout, our single run is the only one among eight methods (seven baselines plus ours) to maintain a positive median annualized return and factor‑level Sharpe, although the median excess return remains negative for every method. The development‑selected top‑20 portfolios achieve a $0.71$ median holdout Sharpe.
We report these single‑run results descriptively, gross of costs, and are explicit about their limits throughout; in particular we do not isolate the effect of the leap machinery from the inherited search substrate, which we leave to future work.
Review: The paper highlights the promise of structured communication for auditable multi‑agent systems, yet further investigation is needed to quantify its true impact on overall performance.