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

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).

Evaluating Self-Correcting Vision Agents Through Quantitative and Qualitative Metrics

Recent progress in multimodal foundation models has enabled Vision-Language Agents (VLAs) to decompose complex visual tasks into executable tool-based plans. While recent benchmarks have begun to evaluate iterative self-correction, its quantitative limits and dominant reasoning bottlenecks remain poorly characterized. This work introduces a Diagnostic Micro-Benchmark. Our analysis decouples Task Success Rate (TSR = 62 percent) from Correction Success Rate (CSR = 25 to 33 percent), revealing that initial competence does not predict repair ability. We explicitly quantify the diminishing returns of correction, which saturates after three retries. Our Failure Taxonomy reveals a frequent factor is Semantic Drift (about 28 percent of failures), a loss of contextual state. By isolating this reasoning bottleneck, this benchmark defines a reproducible framework toward stateful, trustworthy multimodal agents.

Project Aletheia: Verifier-Guided Distillation of Backtracking for Small Language Models

Small Language Models (SLMs, under 10B parameters) are attractive for private, on-device deployment, yet they frequently fail on strict constraint-satisfaction problems due to linear, overconfident reasoning traces that do not recover from early mistakes. We introduce Verifier-Guided Distillation, a training protocol that transfers the process of error repair - explicit conflict detection and backtracking - rather than only correct final answers. By training a 7B model on verified reasoning traces that include mistakes and self-corrections, we show that latent verification behavior can emerge in small models, enabling them to occasionally stop, detect contradictions, and revise earlier assumptions.