Abstract
Systems and methods are disclosed for stabilizing agent-facing recommendation policies subject to performative feedback. Interaction data is collected from an exposed population receiving a deployed recommendation policy and a holdout population receiving baseline recommendations. A first predictor trained on exposed data and a second predictor trained on holdout data are compared to compute a performative effect magnitude, including a divergence between predicted distributions. Counterfactual demand estimation produces item-level recommendation dependence scores and contamination ratios separating organic and induced demand. A self-referential loop detector evaluates directionality using correlation and Granger causality and may apply Fourier analysis to detect feedback resonance. Stability is assessed using a spectral-radius criterion of an estimated performative Jacobian. A correction controller applies damped policy updates and control-theoretic corrections, including proportional-derivative terms, optional model predictive control, diversity drag, organic preference recovery, and market-shape monitoring constraints.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "Performative Stability Controller for Agent-Facing Recommendation Systems", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10726