Abstract
A recommendation platform serves signals such as rankings, trending indicators, popularity metrics, price distributions, and market-level statistics to autonomous agents. A differential privacy layer generates agent-specific responses by applying calibrated (e, d)-differential privacy noise independently per agent, optionally with per-agent consistency within an epoch, so different agents querying the same signal receive different values. The layer maintains per-agent privacy budgets with signal-type-specific query costs and composition-aware accounting, and can rate-limit or return maximally noised coordination-relevant signals upon budget exhaustion. The layer may also randomize signal freshness by scheduling per-agent update delays within signal-type-specific windows. A coordination-risk monitor computes a segment risk score from behavioral correlation, price variance, and concentration and adaptively increases or decreases noise (optionally with hysteresis). Noise allocations may be optimized across signal types to balance utility loss and coordination disruption.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "Coordination-Resistant Differential Privacy Layer for Recommendation Platform Signals", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10730