Yunze Han | Searcharxiv

arXiv2026

Nothing Breaks: No Single Peer Can Soundly Gate Post-Quantum Delivery

Post-quantum protection is delivered to a peer, not declared in a file: whether a session is quantum-resistant is a relation between a server's configuration and the clients that reach it. We show that no single peer can soundly gate that relation. Shipped SSH clients are not ordered: two of their post-quantum capability classes are minimal and incomparable, so a check pinned to either misses the other family's withdrawal. A peer taking both fares no better: it falls back and misses both, or, where classical outranks one family, catches just that one. No case flags both. Nothing above the wire carries the relation either. An artifact-side instrument cannot encode it, because a peer population is not one of its inputs; and across seven configurations on two protocols we find that not one of the five scalars deployed auditors expose to automation moves, while unrelated degradation moves the ones that discriminate at all: the auditors do compute the delivered algorithm, and discard it at the interface automation reads. Nothing else catches the loss either, because nothing breaks: removing a hybrid key exchange starts the daemon, validates the configuration, passes the tests and serves the client, and the adversary it defends against does not exist yet, so no functional signal can carry the loss even in principle. We then show that agents make that state reachable at scale, driving a validated downgrade in 40 of 40 episodes from ordinary engineering prose, against 0 of 40 on a matched neutral document.

cs.CR↗

arXiv2026

A Systematic Evaluation of Trajectory Data Curation for LoRA Fine-Tuning of Code Agents

Supervised fine-tuning (SFT) of open-weight LLMs on expert agent trajectories has emerged as a prominent approach to building capable code agents without reliance on proprietary models. A central yet underexplored question is how trajectory quality and quantity jointly shape model performance. We present a systematic empirical study of trajectory data filtering for LoRA fine-tuning of Qwen2.5-Coder-7B-Instruct on the SWE-trajectory dataset (67,074 trajectories, of which 32,161 are resolved). We propose a two-axis quality scoring framework -- Efficiency and Style -- and evaluate it through 16 controlled experiments spanning strategy, scale, and ablation analyses. Since 7B-scale models attain near-zero SWE-bench resolve rates, we adopt cross-entropy (CE) loss on held-out trajectories as the primary metric, validated via first-action generation: CE loss and ROUGE-L are perfectly rank-correlated (Spearman $\rho$ = -1.00), with limited-sample evidence supporting but not conclusively establishing this proxy. Our results reveal a scale-dependent quality-quantity trade-off: at small scales, doubling the dataset (500 to 1,000) yields ~12.7% CE-loss reduction whereas the TopQ-Random gap stays <1% (Mann-Whitney p > 0.10); at 2,000 trajectories this same gap widens to 3.6% (p = 0.016). Ablation further identifies error-retry rate as the dominant sub-dimension, performing comparably to the full composite ($\Delta$ < 0.2%). Together, these findings establish trajectory-level quality scoring as a viable but scale-sensitive lever for code-agent SFT and offer a proxy-validated evaluation protocol for the regime where end-to-end resolve rate is statistically infeasible.

cs.AI↗