公开论文雷达

Large language models (LLMs) have shown proficiency in various software engineering tasks, such as code generation and translation. However, a key limitation in their performance may be their (lack of) understanding of programming-language semantics. Even when explicit semantics are given, it remains unclear whether LLMs apply those rules or lean on priors learned during pre-training instead. We study if LLMs lean on priors or given semantics with a novel task--Program Executability Prediction (PrEx)--that asks models to predict whether a program is semantically valid or invalid (and, if invalid, which formal rule it violates) given the program's syntax and operational semantics. Because PrEx requires both valid and invalid programs, we build a dataset with systematically generated invalid transformations derived from valid programs. We evaluate open-source coding LLMs under two semantic formalisms and two semantic shifts across Human-Written, LLM-Translated, and Fuzzer-Generated program splits. Our findings show that LLMs lean on pre-training priors rather than systematically applying the given rules, performing especially poorly on modified semantics and degrading further as program complexity increases. PrEx is available at https://github.com/EngineeringSoftware/prex.

Data contamination undermines the reliable evaluation of large language models (LLMs) on mathematical problem solving. While rewriting-based evaluation mitigates memorization, existing methods lack guarantees of problem validity and answer correctness. We propose Proof-Verified Benchmark Rewriting (RePro), the first framework to integrate Lean-oriented neural automated theorem provers (ATPs) into benchmark rewriting, which rewrites problems and regenerates answers with correctness ensured by Lean-verified proofs. Experiments on GSM8K and MATH show that RePro's retained rewritten instances achieve 100% well-definedness, feasibility, and answer correctness, while existing methods still produce invalid or incorrect instances. Moreover, several models exhibit accuracy drops on proof-verified rewritten benchmarks, suggesting that their performance is sensitive to surface-level and structural variations and may partly reflect memorization effects. Our source code and data are available at https://github.com/AI4Engi/RePro.

Repository-level software engineering benchmarks have significantly advanced the evaluation of coding agents, but existing benchmarks primarily measure whether generated patches pass functional tests and overlook review-derived acceptance constraints (review constraints) that often influence whether a patch is acceptable in real-world software development. We introduce SWE-Gate, a repository-level benchmark for software engineering agents that explicitly evaluates review constraint compliance alongside functional correctness. SWE-Gate derives review constraints from real pull request review comments and synthesizes repository-level repair instances around these constraints. Each instance provides separate functional and constraint tests, together with non-compliant and gold patches, enabling explicit separation between issue resolution capability and review constraint compliance. We construct SWE-Gate with 303 repository-level repair instances spanning 75 open-source Python repositories across diverse software domains. Experiments with four LLM backends spanning different capability levels under a common coding-agent scaffold reveal a substantial gap between functional success and success under the complete repair specification: among 644 repairs that pass the functional tests, 221 fail to satisfy the provided review constraints. These findings show that functional-only evaluation overestimates agents' ability to satisfy the full requirements of repository-level repair tasks. The replication package including code, data, and experimental results is available at https://github.com/DeepSoftwareAnalytics/SWE-Gate.

Due to their black-box nature, LLMs suffer from limited explain- ability and a lack of determinism. Their usage cost can also rise, particularly with repetitive tasks on large codebases. To mitigate this, we conduct a novel empirical study targeting three domain- specific languages for transformation rules, namely Comby, GritQL, and Ast-Grep. We evaluate three LLMs (GPT-5.4, GPT-oss-120B, and Llama3.1-8B) on six diverse datasets covering four software- evolution tasks: API misuse correction, program repair, API migra- tion, and language version migration. Our results provide evidence that transformation rule synthesis moves beyond proof-of-concept with strong frontier models. GPT-5.4 achieves consistently high rule applicability rates and produces transformations closest to the ground truth across most benchmarks. Smaller and open-weight GPT-oss-120B and Llama3.1-8B models remain effective for simpler, localized changes but struggle with complex migration scenarios. We also observe non-negligible generalizability through the usage of meta-variables and through a high reuse score in the first quartile of many datasets. Finally, when compared to the anti-unification algorithm, LLMs outperform it in correctness, but underperform in rule applicability. Overall, our results show great potential for LLMs to generate sound, correct, generalizable, and reusable rules.

Requirements engineering (RE) governs the quality of everything downstream in systems engineering (SE); defective requirements that survive review cycles propagate into design rework, schedule delays, and cost overruns. Because requirements are often written in natural language, recent advances in generative AI have raised expectations that large language models (LLMs) can absorb requirement quality assessment, a task otherwise slow and human expertise-intensive. Yet empirical evidence on whether LLMs can be trusted to do so remains scarce. This study presents the first benchmarking analysis of off-the-shelf LLM performance for requirement quality evaluation. Against an expert-derived ground truth built on INCOSE quality criteria, we evaluate ten models spanning two families (OpenAI and Anthropic) and five generations each, across one hundred independent runs, two requirement sets, and five sampling temperatures. Four contributions follow. First, we quantify a strongly asymmetric error profile: across all models and runs, the best-performing Anthropic model detects a median of only 47% of expert-identified issues while false-flagging 11%. Second, performance degrades significantly where SE judgment is required, as necessity and correctness issues are almost always missed. Third, generational progress is non-monotonic, so newer models cannot be assumed better. Fourth, this error behavior shifts only modestly and non-monotonically across sampling temperatures, indicating characteristic model deficiencies rather than inherent stochasticity. Off-the-shelf LLMs are therefore not yet trustworthy autonomous evaluators. Findings also warrant caution for Agentic AI developers: orchestrating these LLM modules in specialized architectures risks compounding these deficiencies rather than correcting them. Their defensible near-term role is human-in-the-loop decision support.

Coding agents often implement changes before users have fully articulated their requirements, echoing a pattern from requirements engineering: stakeholders cannot express a constraint until part of the system exists to react to. This volatility is associated with schedule and budget overruns in traditional projects, but only at release-cycle granularity. Existing work on coding agents narrows this gap only partway: curated benchmarks fix requirements before implementation by design, and observational studies report pushback frequency without linking arrivals to the code invalidation they cause. We address this using 3,553 eligible SWE-chat sessions, coding post-implementation requirement arrivals along three dimensions and, where repository state can be replayed, linking each arrival to a proxy: deletion or replacement of prior agent-authored lines. A requirement's arrival is followed by roughly twice as much invalidation as matched non-requirement edits, robust to user-turn and net-deletion checks, though not demonstrated as causal. This burden shows no detectable decline over a session and no detected association with operation type once multiplicity is accounted for; several intervals remain wide. A controlled experiment shows delayed disclosure relocates implementation post-reveal, while advance warning produces no detected effect on overwriting. These results establish late requirement emergence as a measurable source of code invalidation.

Automated program repair (APR) research has been constrained to design time. Current techniques localize and fix bugs in RTL or HLS designs before a chip reaches production. Once a hardware vulnerability surfaces post-silicon, the patch must be manually generated: existing automation addresses patch deployment but not patch synthesis. We study the feasibility of extending a dictionary-guided, localize-synthesize-validate APR methodology originally developed for RTL repair to this firmware layer. An automated commit-clustering miner surfaces recurring fix templates across the EDK II (UEFI) firmware repository's full commit history without depending on known CVE identifiers, recovering all three known CVE-fix campaigns and surfacing two additional candidate bug families. Grounded in real fix evidence, we build four independent localizers: missing speculation barriers in C (CVE-2017-5753, Spectre v1), missing bounds checks before array writes in C (a decompression library CVE), missing Return Stack Buffer stuffing in x86 assembly (CVE-2017-5715), and missing integer-overflow guards in Hand-Off Block creation code (surfaced by the miner itself). All four achieve 100% recall; precision ranges from 2.1-15.5% on the C families to 100% on the assembly and HOB families. Root-cause analysis of the C-family false positives attributes 77-90% to two intra-procedural causes, isolating the inter-procedural alias-analysis gap as a measured 15-20% rather than an estimate. A held-out test confirms Spectre v1 localization holds at 100% recall on unseen files; a fifth, independently built dictionary entry (CVE-2018-3630) shows the methodology extends to a new bug signature at low cost; and a naive syntactic baseline recalls at most 14% where our detector recalls 100%. We frame these results within a broader research agenda for a unified hardware-to-firmware correctness lifecycle.

Context: Accurate alignment between pull requests (PRs) and corresponding issues is crucial for efficient software development and maintaining code quality, as misalignments can reduce traceability, hinder defect localization, and decrease maintainability. Objective: This study aims to improve automated PR-issue alignment classification by leveraging fine-tuned large language models (LLMs) across multiple alignment categories, and conducts interpretability analysis to investigate the effects of PR-issue fields on the predictions of fine-tuned LLMs. Method: Our methodology consists of dataset preparation, LLM fine-tuning, and interpretability analysis. We first extended an existing dataset and applied data augmentation to address class imbalance. GPT-4o was then fine-tuned via instruction tuning, and open-source LLMs including CodeLlama-7B, CodeQwen1.5-7B, StableCode-3B, CodeGemma-7B, and Deepseek-Coder-6.7B were fine-tuned using classification-specific heads. Interpretability analysis using Shapley Additive Explanations (SHAP) was conducted to examine the influence of PR-issue fields on predictions for the best-performing open-source LLM. Results: Fine-tuned LLMs outperformed baseline models, achieving average improvements of 6.15% in accuracy and F1-micro, 14.69% in F1-macro, and 6.15% in recall. CodeLlama-7B emerged as the best-performing fine-tuned LLM overall, while interpretability analysis revealed that code diffs together with issue body and PR body contents exert the greatest influence on predictions. Conclusions: Fine-tuning substantially enhances PR-issue alignment classification, improving both accuracy and efficiency. Interpretability analysis provides actionable insights into the dataset features driving alignment decisions, deepening understanding of how LLMs reason over software artifacts.

Code generation aims to automatically generate source code from task requirements and has attracted significant attention with the rapid advancement of large language models (LLMs). Despite remarkable progress, LLMs often struggle to generate correct code for complex software engineering tasks because task descriptions are frequently incomplete, ambiguous, or lack critical contextual information. Existing approaches primarily improve the capabilities of coding agents through more sophisticated tools, skills, and workflows, while largely overlooking the quality of the task requirements themselves. To address this limitation, we draw inspiration from software requirements engineering and propose WiseSpec, a novel requirements-driven agent framework for repository-level code generation. WiseSpec automatically constructs structured and information-rich requirements, assesses their quality through execution-based evaluation, and iteratively refines them to better guide code generation. Experimental results show that WiseSpec consistently outperforms all baselines, achieving an average improvement of 13.17% in %Resolved.

This paper presents preliminary findings from a multivocal literature review investigating how quantum software testing is characterized across academic and practitioner-oriented sources. Our study integrated peer-reviewed studies with gray literature, including blogs, tutorials, forums, technical reports, documentation pages, and company webpages. Our results indicate a rapidly evolving but fragmented ecosystem involving classical adapted testing approaches, quantum-specific techniques, statistical validation methods, simulators, debugging environments, and verification frameworks. The reviewed material also revealed recurring challenges related to scalability limitations, hardware noise, probabilistic execution, limited observability, and immature tooling ecosystems. These findings provide an initial characterization of how research and practice currently discuss quantum software testing challenges, techniques, and tooling.

Large language models have shown promising capabilities in code refactoring, but existing approaches remain limited to method-level tasks. In this paper, we envision LLM-based refactoring as a continuous component of software maintenance rather than a tool invoked only for occasional manual refactoring. Under this vision, AI agents continuously monitor, evaluate, and improve codebases against explicit and evolving notions of software quality. We present a roadmap organized around five dimensions: the multi-objective optimization problem, quality definition and evaluation, multi-timescale integration of heterogeneous signals, architecture and design pattern, and trust in autonomous refactoring. We further identify integration into continuous delivery pipelines and cost considerations as cross-cutting concerns. For each dimension, we analyze the underlying challenges and pose open research questions. These dimensions define a research agenda for advancing autonomous refactoring from isolated code improvements to system-level quality maintenance.

Most vision-language-action (VLA) models -- OpenVLA, $π_0$, RT-2, RDT-1B -- are monolithic: they emit raw motor commands or short action chunks without organizing behavior into reusable abstractions, so they degrade on long-horizon tasks and resist interpretation. Existing skill-discovery methods sidestep the core question of when two action sequences are behaviorally equivalent, either clustering contrastive embeddings or delegating the judgment to a language model uncalibrated to the robot's dynamics. We introduce REFACTOR-VLA, a wake/sleep system for learning reusable skills. Its sleep phase clusters motor-program fragments under a Behavioral-Equivalence Kernel (BEK) computed from rollouts of a learned latent world model $M_φ$; its wake phase emits typed lambda terms over a Hindley--Milner-inspired vocabulary, consumed by a library-conditioned rectified-flow action decoder. Abstractions are admitted only if they pass Minimum Description Length and return-preservation gates. On LIBERO we report two findings. First, enlarging the world model from 188M to 430M parameters worsened performance on 4 of 4 suites, so capacity alone does not help. Second, the training objective matters far more: adding an auxiliary supervised contrastive (InfoNCE) loss during world-model warmup substantially improves sleep-phase clustering, giving Normalized Mutual Information at $n=3$ seeds of $0.462 \pm 0.021$ (object), $0.867 \pm 0.025$ (spatial), $0.915 \pm 0.013$ (goal) and $0.754 \pm 0.010$ (LIBERO-10), and beating the strongest published baseline on all 4 suites by a mean $Δ= +0.184$. Across providers ($n=12$) the 95% bootstrap confidence interval for mean pairwise NMI is $[0.683, 0.729]$ (mean $0.705$). The sleep phase also yields the first real-LIBERO task-language library: the decoder uses 2 of 3 admitted abstractions and rewrites all 256 sampled demonstrations.

Large language models (LLMs) are increasingly evolving from conversational assistants into agents capable of operating external digital environments. Graphical user interface (GUI) agents play an important role in this transition, as many real-world workflows remain accessible only through user-facing software interfaces. However, despite recent progress on general computer-use benchmarks, domain-specific professional standard operating procedures (SOPs) remain challenging for GUI agents because they often involve implicit domain knowledge, software-specific conventions, and task-level verification requirements. We introduce OmegaUse-SOP, a human-in-the-loop SOP Engineering system for transforming human demonstrations of professional computer use into reusable SOP skills for GUI agents. Analogous to prompt engineering, SOP Engineering iteratively refines demonstrations, execution rules, and domain knowledge to convert professional SOPs into reusable GUI-agent skills. OmegaUse-SOP consists of four modules: Observe, Reason, Configure, and Execute. Together, these modules record expert operations as multimodal GUI traces, abstract low-level events into semantic step-level instructions, incorporate domain rules and task-specific parameters, and execute the resulting skills in live GUI environments through step-wise grounding, action generation, and verification. To demonstrate its effectiveness, we collaborate with a power-sector client and test OmegaUse-SOP on photovoltaic simulation workflows in PVsyst 7.2. The results suggest that OmegaUse-SOP can improve GUI-agent reliability on professional SOP tasks, highlighting a practical path toward deploying GUI agents in domain-specific professional software environments.

End-to-end GUI testing is essential for verifying web applications, yet existing evaluations rely on predefined checklists and are confined to the data and frameworks of web generation benchmarks, leaving the bug-discovery ability of vision-language models (VLMs) systematically untested. We introduce \textbf{C}ode-driven \textbf{A}gentic \textbf{T}esting (CAT), a paradigm in which the agent writes Playwright code to drive the browser, gathers feedback, and autonomously explores web applications to uncover bugs. We instantiate CAT with CATJudge, an agentic framework that unifies Browser-Use and Computer-Use tools within a single environment and CATTest, a benchmark of 102 AI-generated web applications with carefully annotated bugs, built through close human-AI collaboration to feature complex interactions and subtle defects. Experiments with mainstream VLMs show that all evaluated models perform poorly, revealing a clear gap between current VLM capabilities and the demands of real-world testing in AI web development. We release our code and data at https://github.com/SleepyWithoutCoffee/CATJudge.

As large language models (LLMs) become increasingly capable, the long-term value of AI systems depends not only on solving individual requests, but also on transforming experience and accumulated knowledge into durable, reusable competence. We introduce SimSkill, a self-evolving agent built around the Simulation of Urban MObility (SUMO) traffic simulator. SimSkill identifies capability gaps, generates and solves environment-grounded tasks, verifies solutions through an action--critic loop, and consolidates experience into episodic, procedural, and semantic memory without updating the backbone model. Through autonomous exploration, it builds a reusable library spanning the traffic-simulation workflow. We evaluate SimSkill on two held-out benchmarks with three backbone LLMs and independent artifact-based verification. SimSkill improves verified completion by up to 25 percentage points, while ablations show complementary contributions from procedural and semantic memory. Its benefits remain backbone- and budget-dependent: memory does not improve every model or uniformly reduce inference cost. More broadly, SimSkill illustrates a design paradigm in which natural language preserves and composes computational capabilities, while executable tools and code provide precise and reproducible execution. All code and experimental data are publicly available at https://github.com/qiliuchn/SimSkill-V1.

An agent that inherits six one-line memories may pull at most one archived source record before acting; a directive written into the store can steer that choice: a pointer to the record, a criterion that identifies it, or both. Across twelve registered studies on one instrument lineage (14,760 attempts) we measured where the request goes under each form. On six direct-provider models a length-matched criterion exceeded a bare id by +35.0 points [+31.2, +38.8] (Study D); the contrast failed its registered superiority rule on a nine-model OpenRouter-served panel (Study E). Appending the id cancelled the criterion on three Claude models (Opus 5: 40/40 to 0/40; Study F-x); six byte-matched edits gave each exact string its own effect (Study G), and a re-run at eighty runs per cell left fifteen of thirty replication contrasts within the margin, fifteen unresolved and none beyond (Study G'). A ratification line (+96.0 points on Opus 5) and a budget of two credits restored the target on all three (Study J); across five criterion strings the suffix's cancellation held for four of the five wordings on Opus 5 and all five wordings on Fable 5.1 (Study H2); in a second store every model followed the criterion (Study H1). Continued into a decision, the criterion moved the choice toward the current record (+100.0 points, Opus 5) and away from it on Fable 5.1 (Study I). A one-character plan pointer's effect (+78.0 points; Study B, after a correction of its first repository report) returned the same verdict under a prospectively registered re-run (+81.7 points; Study B'). All results are descriptive effects of exact edits on fixed panels with registered intervals and no mechanism claim.

Distributed LLM-agent teams can read the latest shared facts and still act on an obsolete plan. A planner may derive an action from requirement $r_3$, another agent may commit $r_4$, and an executor may receive $r_4$ without replacing the plan derived from $r_3$. We call this \emph{stale-plan execution}: state freshness does not establish that the plan authorizing an action remains valid. We introduce PlanFence, a dependency-scoped action-validation protocol. Plans cite the exact public records they used, and an executor validates only the records that can affect the pending external action, replanning once or blocking when validation is incomplete. In 30 controlled live workflows with a post-plan revision, a freshness-only executor acts on the obsolete plan in every task, whereas PlanFence completes all tasks without an invalid action. Controlled replay reveals two conditional boundaries: proactive synchronization yields lower coordination stall at low churn, while PlanFence avoids repeated update-path coordination as churn grows and avoids validating unrelated state as the shared keyspace grows. These are controlled safety and systems-cost results, not general task-accuracy gains.

Multi-agent LLM systems commonly use an orchestrator to decompose a task for a team of workers and then improve through textual reflection. Despite strong empirical results, these systems lack a unified account of coordination, memory improvement, and the role of external verification. We model orchestrator-worker interaction as a bilevel coordination game: under bounded coupling, the workers' local-update game is an approximate potential game whose equilibrium slack is controlled by decomposition quality. We then analyse reflection as stochastic movement over semantic memory states. For free-form reflection, we derive a finite-time upper bound, prove worst-case tightness, and give a positive lower bound under a falsifiable persistent-harm condition. We further prove an information-theoretic impossibility result: no gate that observes only the generated transcript can improve uniformly over text-indistinguishable environments, whereas an environment-grounded gate can. Motivated by this separation, we introduce Stochastic Reflective Memory Ascent (SRMA), which accepts a candidate memory only after a grounded evaluation risk strictly decreases. Under calibration and non-degenerate corrective mass, SRMA converges exactly, geometrically or polynomially; matching constructions show that both rate regimes are order-tight. We also provide confidence gating for stochastic evaluation and re-anchoring guarantees for piecewise-stationary environments. Experiments instantiate these objects with environment-grounded metrics and test the predicted coordination and drift laws. On 500 SWE-bench instances, the complete Kimi-based system resolves 72.2% versus a 70.8% public mini-SWE-agent reference. Code: https://github.com/YihangChen9/Bilevel-Coordinated-Reflection

Long-running LLM agents rely on persistent memory to carry state across interactions, including permissions, restrictions, and revocations. When memory misrepresents this evolving authorization state, the agent's own records can grant authority that the underlying history never permitted, resulting in misaligned behavior without any external attacks. We term this failure endogenous authorization laundering, where spurious permissions written into memory lead to unauthorized actions as their provenance is washed away. We then introduce EAL-Bench, which measures how accurately persistent memory preserves evolving authorization state and whether errors propagate to downstream unauthorized actions. We evaluate five LLMs as memory writers and two as executors across procurement, cybersecurity, and finance. We find that under incremental memory updates, writers create false authority for up to 50.2% of unauthorized requests; once false authority is present, executors act on it in 98.6% of trials. Two safeguards, requiring stored permissions to be backed by valid source events, and tracking permission changes through bounded event sourcing, substantially reduce laundering, but both also reject more legitimate actions, exposing a safety-utility tradeoff. Persistent memory is therefore not merely a performance component, but a part of an LLM agent's effective authorization policy.

As enterprise platforms transition to conversational reasoning interfaces, the stateless nature of LLM APIs creates an architectural gap. While statelessness enables horizontal scalability for AI providers, it forces client applications to manage the entire burden of conversational state and semantic memory. The work identifies the Hydration Proxy Pattern, an architecture that decouples session persistence from the reasoning engine. The framework ensures platform sovereignty over conversational data while enabling secure, multi-stage semantic grounding. We further propose the Context Stabilization Mandate to resolve the tradeoff between sovereign state management and KV caching.

The browser has become a first-class database host: applications increasingly want to store, query, and reason over structured data entirely on the client - for privacy, offline operation, local-first collaboration, and, most recently, as durable memory for in-browser AI agents. One way to get SQL in the browser, compiling PostgreSQL to WebAssembly (PGlite), inherits PostgreSQL's process model: a single backend connection that executes one statement at a time and blocks. That model cannot express concurrent transactions, and it leaves richer capabilities - graph queries, database branching - to whatever the compiled server happens to include. We present zeta-lite, the browser form factor of the Zeta database engine: a WebAssembly build that compiles the same Zeta server down to a 2.87 MB gzipped artifact. Zeta-lite keeps the engine's log-centric asynchronous MVCC core, which yields two capabilities no other in-browser SQL engine provides. First, overlapping snapshot-isolated transactions on a single thread: multiple transactions hold distinct read/commit timestamps and interleave, with snapshot-isolation conflict detection between them. Second, copy-on-write database branching - whole-database fork, merge, and rebase - is unique in a browser SQL database and rare even in servers. On top of these, zeta-lite exposes a feature-complete PostgreSQL surface (joins, CTEs, window functions, JSONB with GIN indexes, full-text search, HNSW vector search, SQL/PGQ graph queries, multi-database) and snapshot-to-OPFS durability. Across Chrome, Firefox, and a native reference runtime, zeta-lite sustains 268k-315k point reads/s and holds a mixed read/write workload flat over millions of operations. This small, fully-featured, concurrent SQL database is an especially good fit for agentic memory - where cheap branchable state lets an agent explore, inspect, and commit or discard speculative work.

MutMem V1 introduced retention-preserving, cryptographically authorized mutation for persistent agent memory but did not provide a complete portable verification contract or clean-install reproduction path. MutMem V2 closes that publication gap without introducing a second memory engine. It specifies exact canonical bytes, domain-separated object and bundle commitments, mandatory recall-evidence membership and ordering, external trust anchors, identity epochs, revocation, authorization, request receipts, ordered disclosure, and three mutation terminal types. The released protocol contains 18 versioned object schemas, 39 recall vectors, 15 mutation vectors, and 37 closed recall failure reasons. Independent Node and Python implementations agree on verdict and primary reason for all 72 structural and cryptographic terminals; a production-conformance corpus agrees on 42/42 cases across 28 required classes. A clean Node v26.8.1 installation reaches first-boot, restart, and scheduler readiness with no experimental memories. A separately scoped 120-unit Canary experiment supports only explicit-marker traversal. Every public table regenerates from a self-hashed aggregate, and an independent verifier reconstructs the statistics and claim boundaries. Historical V1 empirical results remain historical. MutMem V2 supports claims about portable integrity, authorization, traceability, conformance, and reproducibility under stated assumptions; it does not establish semantic truth, universal robustness, or independent replication.

Clinical LLM assistants must reason over multi-visit patient trajectories, yet whether the compact history representations used to scale them---retrieval, structured timelines, LLM summaries, agentic memory---preserve the longitudinal signal clinical reasoning needs has not been measured. We introduce ClinTraceBench: 385 MIMIC-IV-derived verified dialogues with event-ID provenance, a nine-task taxonomy (T1--T9), and L0--L4 deterministic + L5 human-audit validation (98.92\% agreement). We evaluate eight history representation strategies---a no-context floor, \textit{last-visit-only}, \textit{full-context}, BGE-M3 \textit{dense-retrieval}, two compression schemes, and two agentic-memory systems (\textit{Mem0}, \textit{A-Mem})---across four backbones (DeepSeek-V3, GPT-4o-mini, Haiku~4.5, Sonnet~4.6) on 6{,}271 questions: 32 cells, 200{,}672 predictions. Four findings: (SP4) a controlled T3 injection probe isolates compression-induced \textit{relation} loss---with the attribution sentence present \textit{before} construction, \textit{Mem0}, \textit{A-Mem} and \textit{llm-summary} still recover only 0--5.3\% of the injected positives; (SP1) compressed strategies pay an aggregation tax on multi-visit trends and cross-patient comparisons; (SP2) the blind-to-full gap spans $+29.8$~pp (GPT-4o-mini) to $+62.7$~pp (Haiku); (SP3) abstention scales non-monotonically with context length. On the Pareto frontier Haiku dominates Sonnet under \textit{full-context} (\$25.76 vs.\ \$106.21), inverting the ``biggest backbone wins'' heuristic.

LLM agents that cache recovery suggestions from API errors can skip re-derivation in later episodes, spending fewer tokens and fewer model calls on constraints they have already learned. Server-side data drift turns those cached fixes into silent failures, and the usual remedy, re-deriving on every episode, gives the savings back. We introduce invalidation contracts, a protocol layer that attaches version stamps and cacheability hints to every recovery suggestion so the client can evict stale entries without trial and error, and keep the rest. The contract decomposes realized savings into two independent factors: validity, the fraction of cached suggestions that remain correct after a drift event, and compliance, the fraction the planner applies on the first attempt. Validity depends only on the protocol and is vendor-independent. Compliance depends on the planner model: identical wire bytes yield 100% first-try compliance on Claude Haiku 4.5 and 11% or below on Claude Sonnet 5, which exhibits input-schema conservatism, refusing fixes that add fields the original request did not contain. We evaluate across seven models, three serving paths, two domains, and approximately 9,400 episodes. Row-level invalidation raises compliance by 0 to 66.7 percentage points across the seven models, 55.6 to 66.7 on three, and recovers 29-33% of baseline token cost on four of seven models, while table-level invalidation destroys co-located entries and drops post-drift first-try rates to 0% on five of seven. Eviction precision is 1.00 at row granularity on every model under the row-level oracle of Section 4.1. The contract adds 15% to response payload. Version-stamp validity is deterministic by construction and produced identical results across every model and serving path, with zero contract failures in the entire evaluation.