Memory-backed agents need provenance that can survive leaked or migrated snapshots, where logs, visible outputs, and trusted metadata may be absent. We propose MEMMARK, a state-evolution attribution watermark that embeds an owner-controlled signal into latent memory-write decisions. At each internal LLM call, MEMMARK samples among admissible candidates using keyed, distribution-preserving selection, and records cryptographic commitments with signed session anchors and reveal evidence. This makes attribution depend on reproducible backend behavior rather than mutable provenance fields. Across A-Mem and Graphiti on LOCOMO, with three LLM backbones, MEMMARK preserves memory utility: Overall F1 retains 99.6 percent of the unwatermarked baseline, while BLEU-1 changes by +0.2 percent. It also provides usable carrier capacity, with 1.16, 1.14, and 1.26 bits of mean entropy for update-target, link-target, and semantic-realization decisions. In the snapshot-only R3 setting, MEMMARK recovers the full 40-bit payload from final snapshots, while wrong-key verification remains near chance. Under nine memory-lifecycle attacks, verification distinguishes tampering, evidence deletion, and partial payload recovery. These results show that robust snapshot-only attribution is feasible for long-term agent memory without surviving traces, trusted metadata, or utility degradation.

With the growing adoption of Large Language Model (LLM) agents in persistent, real-world roles, they naturally encounter continuous streams of tasks and inevitable failures. A key limitation, however, is their inability to systematically learn from these mistakes, forcing them to repeat identical errors in similar contexts. Unlike prior training-free methods that primarily store raw instance-level experience or focus on retrieving successful trajectories, we propose Mistake Notebook Learning (MNL), a novel memory framework that enables agents to self-curate generalizable guidance from batch-clustered failures. This mechanism allows agents to distill shared error patterns into structured “mistake notes”, updating an external memory only when batch performance improves to ensure stability. To further amplify adaptability, we integrate MNL with test-time scaling, leveraging aggregated failure patterns to actively steer the search process away from known pitfalls. Experiments on mathematical reasoning, Text-to-SQL, and interactive agent benchmarks show that MNL achieves competitive performance compared to existing memory mechanisms in both effectiveness and efficiency. These findings position structured mistake abstraction as a critical lever for robust agent evolution, enabling continuous improvement without the cost of parameter updates.

Sales dialogues require multi-turn, goal-directed persuasion under asymmetric incentives, which makes them a challenging setting for large language models (LLMs). Yet existing dialogue benchmarks rarely measure deal progression and outcomes. We introduce SalesLLM benchmark, a bilingual (ZH/EN) benchmark derived from realistic applications covering Financial Services and Consumer Goods, built from 30,074 scripted configurations and 1,805 curated multi-turn scenarios with controllable difficulty and personas. We propose a fully automatic evaluation pipeline that combines (i) an LLM judge for sales-process progress, and (ii) fine-tuned BERT classifiers for end-of-dialogue buying intent. To improve simulation fidelity, we train a user model, CustomerLM, with SFT and DPO on 8,000+ crowdworker-involved sales conversations, reducing role inversion from 17.44% (GPT-4o) to 8.8%. SalesLLM benchmark scores correlate strongly with human ratings (average Pearson r = 0.86; inter-annotator agreement Krippendorff’s α = 0.86, 500 annotations per model). Across 15 mainstream LLMs, in Chinese the strongest models are competitive with typical (junior-to-intermediate) human salespeople — not with sales experts — while weaker ones fall below this baseline, and cross-lingual consistency remains poor. SalesLLM benchmark serves as a scalable benchmark for outcome-oriented sales agents. Our code and data are released at https://github.com/Bairong-Xdynamics/Benchmarking-LLM-Realistic-Selling-Skill

LLMs are increasingly used to arbitrate among multiple candidate outputs and compress them into a single final result. The final answer may not be incorrect, yet information compression can still make it suboptimal. We define this problem as source-resolution distortion (SRD), the irreversible loss or incorrect mapping of task-relevant source-side structure in the final output during arbitration.

We model pairwise arbitration as a compression process; we then present a source-bounded SRD measurement framework that aligns the reference structure fixed before arbitration with the final output and uses six metrics to measure its preservation and incorrect mapping. To reduce SRD, we further propose A²R (Auditable Arbitration for Resolution), a reference resolution method based on a decision ledger. Motivated by anti-aliasing in signal processing, A²R extracts query-relevant source units and records the local resolution relations among them in a decision ledger before using it to constrain final generation.

Across 520 pairwise cases, compared with Checklist Prompt, A²R-Decision lowers CER from 0.23 to 0.16 and raises Prov-H from 0.45 to 0.67. Its answer-preference scores against three structured baselines range from 51% to 53%, showing that similarly preferred answers can still differ markedly in source recoverability. A 200-case three-candidate extension preserves the system ordering.