Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PCN-Rec: Agentic Proof-Carrying Negotiation for Reliable Governance-Constrained Recommendation

Aradhya Dixit, Shreem Dixit

TL;DR

PCN-Rec addresses governance-constrained recommendation by decoupling reasoning from enforcement through a proof-carrying negotiation pipeline. An LLM-based mediator proposes a Top-N slate along with a machine-checkable certificate, which a deterministic verifier uses to ensure constraints such as head/tail exposure and genre diversity are satisfied; a deterministic repair path handles violations and yields auditable traces. Empirical results on MovieLens-100K show near-perfect governance compliance for feasible users with minimal NDCG@10 loss (≈0.403 vs 0.424, Δ≈-0.022, p<0.05) compared with a one-shot baseline, demonstrating the practical viability of auditable, constraint-satisfying recommendations. The approach provides a scalable, transparent framework that can be extended to richer constraint sets and adaptively adjust the candidate window.

Abstract

Modern LLM-based recommenders can generate compelling ranked lists, but they struggle to reliably satisfy governance constraints such as minimum long-tail exposure or diversity requirements. We present PCN-Rec, a proof-carrying negotiation pipeline that separates natural-language reasoning from deterministic enforcement. A base recommender (MF/CF) produces a candidate window of size W, which is negotiated by two agents: a User Advocate optimizing relevance and a Policy Agent enforcing constraints. A mediator LLM synthesizes a top-N slate together with a structured certificate (JSON) describing the claimed constraint satisfaction. A deterministic verifier recomputes all constraints from the slate and accepts only verifier-checked certificates; if verification fails, a deterministic constrained-greedy repair produces a compliant slate for re-verification, yielding an auditable trace. On MovieLens-100K with governance constraints, PCN-Rec achieves a 98.55% pass rate on feasible users (n = 551, W = 80) versus a one-shot single-LLM baseline without verification/repair, while preserving utility with only a 0.021 absolute drop in NDCG@10 (0.403 vs. 0.424); differences are statistically significant (p < 0.05).

PCN-Rec: Agentic Proof-Carrying Negotiation for Reliable Governance-Constrained Recommendation

TL;DR

PCN-Rec addresses governance-constrained recommendation by decoupling reasoning from enforcement through a proof-carrying negotiation pipeline. An LLM-based mediator proposes a Top-N slate along with a machine-checkable certificate, which a deterministic verifier uses to ensure constraints such as head/tail exposure and genre diversity are satisfied; a deterministic repair path handles violations and yields auditable traces. Empirical results on MovieLens-100K show near-perfect governance compliance for feasible users with minimal NDCG@10 loss (≈0.403 vs 0.424, Δ≈-0.022, p<0.05) compared with a one-shot baseline, demonstrating the practical viability of auditable, constraint-satisfying recommendations. The approach provides a scalable, transparent framework that can be extended to richer constraint sets and adaptively adjust the candidate window.

Abstract

Modern LLM-based recommenders can generate compelling ranked lists, but they struggle to reliably satisfy governance constraints such as minimum long-tail exposure or diversity requirements. We present PCN-Rec, a proof-carrying negotiation pipeline that separates natural-language reasoning from deterministic enforcement. A base recommender (MF/CF) produces a candidate window of size W, which is negotiated by two agents: a User Advocate optimizing relevance and a Policy Agent enforcing constraints. A mediator LLM synthesizes a top-N slate together with a structured certificate (JSON) describing the claimed constraint satisfaction. A deterministic verifier recomputes all constraints from the slate and accepts only verifier-checked certificates; if verification fails, a deterministic constrained-greedy repair produces a compliant slate for re-verification, yielding an auditable trace. On MovieLens-100K with governance constraints, PCN-Rec achieves a 98.55% pass rate on feasible users (n = 551, W = 80) versus a one-shot single-LLM baseline without verification/repair, while preserving utility with only a 0.021 absolute drop in NDCG@10 (0.403 vs. 0.424); differences are statistically significant (p < 0.05).
Paper Structure (30 sections, 8 equations, 2 figures, 1 table, 1 algorithm)

This paper contains 30 sections, 8 equations, 2 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Method: Proof-Carrying Negotiation for Recommendation
  3. PCN-Rec Framework
  4. Candidate Window.
  5. Negotiation Agents.
  6. Mediator (LLM) + Certificate.
  7. Deterministic Verification.
  8. Deterministic Repair (Fail-Safe).
  9. Feasibility vs. Method Failure
  10. Feasibility Analysis and Empirical Results
  11. Problem Setup and Constraints
  12. Head/Tail exposure.
  13. Genre diversity.
  14. Feasibility vs. Method Failure
  15. Proof-Carrying Verification Metric
  16. ...and 15 more sections

Figures (2)

  • Figure 1: PCN-Rec proof-carrying negotiation pipeline. A base recommender produces a candidate window ($W$). Two agents negotiate (User Advocate for relevance; Policy Agent for governance), and a mediator LLM proposes a Top-$N$ slate plus a structured certificate. A deterministic verifier recomputes constraints from the slate and accepts only verifier-checked certificates; failures trigger deterministic constrained-greedy repair followed by re-verification, producing an auditable trace.
  • Figure 2: Constraint feasibility vs. candidate window size $W$. The feasibility rate (fraction of users for whom at least one compliant slate exists within $C_W(u)$) increases with $W$, while tail-shortage decreases. We select $W{=}80$ as an operating point.