Agregador

arXiv:2610.28810v1 Announce Type: cross Abstract: The widespread adoption of generative artificial intelligence in recent years has brought a parallel rise in prompt injection and jailbreak attacks. By embedding malicious instructions, these exploits manipulate language models to bypass intended behavior and operational boundaries. In traditional chatbots, such attacks are confined to a single session because state memory is not preserved across conversations. However, persistent memory features in modern conversational interfaces enable cross-session attack propagation, establishing a threat vector known as memory injection. This work explores this emerging vulnerability in leading LLM-as-a-service platforms, highlighting its real-world severity and the need for countermeasures. While existing literature focuses on violations tailored to specific platform features, we present a general threat model. We demonstrate that all NIST attack goals are collectively achievable across ChatGPT Instant 5.3, Claude Sonnet 4.6, and Gemini 3 Flash under this model. We also show that memory injections can exploit multiple attack vectors, including personal memories, projects, skills, organization instructions, and Model Context Protocol (MCP)-connected resources. Our findings reveal that all evaluated services can be manipulated to persistently spread disinformation, generate harmful content, or simulate offline unresponsiveness, even in fresh sessions. In the particular case of Claude, shared-memory features allow an attacker to compromise multiple users within an organization and also exfiltrate user prompts. To address these vulnerabilities, we derive defense mechanisms from our proposed threat model and foundational cybersecurity principles. Our analysis indicates that nearly half of these mitigations remain unimplemented across the target services. Datasets and attack artifacts are available at https://github.com/PedroBarcha/Memory-injection.

arXiv:2610.28803v1 Announce Type: cross Abstract: While statistical learning has made remarkable progress in recent decades, driven by large datasets and data-driven protocols that have transformed scientific practice, it still faces a key limitation: it struggles to generalize beyond the data it has seen, often failing under distribution shift. Causality offers a promising solution: causal models focus on the underlying mechanisms of change rather than surface-level associations, enabling robust reasoning under external interventions. This allows systems not only to learn from data, but also to reason about hypothetical scenarios, adapt to new environments, and inform better decisions. This thesis explores how these approaches can inform each other, in two parts. The first part examines algorithmic recourse, that is, identifying which changes an individual can make to overturn an unfavorable decision. We propose an alternative perspective to the established framework, arguing for a more realistic cost function, a reformulation of the problem in causal terms, and the inclusion of a temporal dimension in policy evaluation. The second part focuses on dynamical systems, extending the notions of causal interventions and effects to discrete and continuous stochastic processes. We show that the equilibria of these processes, and how they change following an intervention, can be linked to Structural Causal Models (SCMs), viewed as compressed representations of the underlying temporal dynamics. We argue for the relevance of such temporal causal models, since in real-world scenarios actions take time to produce their effects. The common goal of both parts is practical: the utility, robustness, and trustworthiness of these models depend critically on their ability to realistically represent the underlying phenomena.

arXiv:2610.28797v1 Announce Type: cross Abstract: We introduce CALLIOPE, a top-down framework for discovering topic hierarchies in large corpora of unstructured data (e.g., LLM conversations), while supporting differential privacy (DP). Unlike traditional and even recent LLM-augmented approaches that ultimately rely on clustering vector embeddings, our framework leverages LLM capabilities directly for the entire pipeline. Experiments demonstrate that this approach provides better utility over existing (embedding-based) methods in the DP setting, especially in the high privacy regime. We evaluate the quality of CALLIOPE outputs using a variety of traditionally studied clustering metrics, as well as some novel utility metrics that we introduce.

arXiv:2610.28788v1 Announce Type: cross Abstract: Flow-matching policies adjust inference cost through the number of solver steps, but reinforcement learning at a single fixed solver budget can hide degradation at cheaper deployment budgets and even erase useful one-step behavior present in a pretrained policy. We introduce Action-Ratio-Clip Flow (ArcFlow), a deployment-aware algorithm for training one flow policy across a specified range of solver budgets. ArcFlow combines action-aligned supervision, budget exposure, and worst-budget checkpoint selection. For reconstruction objectives, its bridge aggregates temporal probes into one surrogate ratio and one PPO-style clip per action, with a boundary residual exactly equal to deployed one-step reconstruction error. Cycling exposes the shared field to the deployment range, and selection optimizes its performance floor. For native velocity objectives, multi-budget endpoint supervision applies the same principle without replacing the native update. Both realizations preserve the architecture and deployment-time cost. Across DM Control, manipulation, and locomotion, ArcFlow attains the highest mean one-step return among five flow-policy baselines on 25 of 30 tasks under each method's reported training settings. Component studies identify complementary improvements from boundary supervision, the fixed-grid bridge, budget exposure, and deployment selection. These results establish deployment-budget-aware training as an effective route to strong low-step control with one policy serving multiple inference costs.

arXiv:2610.28779v1 Announce Type: cross Abstract: World models let agents compare imagined outcomes before acting, but an agent must still decide which alternatives to explore and when a proposed plan should replace its learned policy. We introduce SDI (Sibling Disagreement in Imagination), which connects these decisions through a shared actor-based reference. SDI compares action sequences imagined from the same model state, using repeated rollouts of the actor's own sequence to measure variation under fixed behavior. This reference scales both the spread across candidate returns, which provides an exploration signal, and a proposed plan's predicted advantage, which governs action acceptance. The agent executes a planned action when its advantage exceeds the reference variability by a fixed factor; otherwise, it follows the actor. SDI reuses the base agent's world model, actor, and critic without requiring an additional disagreement ensemble. Exploration and planning guide training data collection, while evaluation uses the learned actor alone. Across Meta-World, Adroit, and DM-Control, SDI achieves a higher observed task-average score than every compared baseline in each suite and matches or exceeds the strongest baseline's performance on 15 of 18 tasks. Component ablations and robustness analyses further examine the roles of exploration, planning exposure, and action acceptance. Paired evaluations from identical simulator states show that the gate improves both recovery from actor failures and preservation of successful behavior relative to ungated planning, while substantially reducing return degradation. Together, these results support comparing imagined alternatives against a shared actor reference as a unified approach to exploring and planning with world models.

arXiv:2610.28770v1 Announce Type: cross Abstract: RLVR improves language-model reasoning, but commonly relies on independent stochastic rollouts for exploration, which can repeatedly sample similar failed strategies. Prior work has seen success with approaches that guide exploration by introducing dependence into the sampling process by including incorrect solutions in the context for subsequent attempts. Still, these methods introduce the additional question of how to assign credit to each attempt. We propose to view the credit assignment through the lens of Stochastic Shortest Paths, a framework in classical reinforcement learning that seeks to minimize the expected cumulative cost to reach a terminal state (in our case, a verified successful response). We show that, under suitable simplifying assumptions, this perspective naturally leads to a simple modification to GRPO, which we call Retry-Aware Policy Optimization (ReAPO). ReAPO allocates additional generation to failed trajectories, allowing successful retries to be reinforced without rewarding preceding failures, and assigns credit in a theoretically principled way to each attempt. Across Qwen3 and OLMo-2 models from 0.6B to 4B and eight mathematical and scientific reasoning benchmarks, ReAPO improves first-attempt performance over GRPO as well as natural recent baseline algorithms, while additional ablations demonstrate the utility of both serial sampling and ReAPO's credit assignment. On Qwen3-0.6B, ReAPO also outperforms GRPO with twice the group size or twice the training steps, reaching their peak accuracy with 39% and 31% fewer generated tokens, respectively. Taken together, our results suggest that ReAPO is a simple yet effective approach for improving LM reasoning performance through improved exploration and retry-aware credit assignment.

arXiv:2610.28767v1 Announce Type: cross Abstract: Current evaluation of LLMs on benchmarks typically aggregates performance over individual examples into a single scalar score. This approach is overly coarse and ignores the diverse scope of skills covered by benchmark datasets. To address this limitation and better exploit benchmark data, we propose SkillEval, a new evaluation framework that performs cross-model, skill-level analysis. The framework integrates advances from two areas: (1) it leverages the meta-cognitive capabilities of LLMs to automatically discover interpretable skills from questions, and (2) it applies cognitive diagnosis models (CDMs) from psychometrics to estimate per-skill abilities of LLMs. We conduct experiments on 3,811 LLMs evaluated on 9,523 items across five benchmarks spanning diverse domains (MATH, BBH, GPQA, MuSR, IFEval). Compared to the traditional average-based evaluation paradigm, SkillEval produces reliable, more calibrated and informative yet concise model profiles that more accurately predict model behavior on unseen items. We also demonstrate the utility of SkillEval in two downstream tasks: (1) LLM routing, which dynamically assigns each query to a capable but efficient model, reducing inference cost by over 60% while achieving improved performance, and (2) Computerized Adaptive Testing (CAT), where an unseen model can be efficiently evaluated using a trained CDM. Furthermore, we show that these skill profiles provide useful insights, for example, how benchmarks share skills and how instruction-tuned models differ from their base models.

arXiv:2610.28759v1 Announce Type: cross Abstract: The Gabriel graph is a proximity graph widely used in computational geometry and geometric machine learning, underlying large-margin classifiers and, as we show in this work, density-aware clustering methods. Its construction, however, has cubic time complexity $O(N^3)$ in the naive formulation, which severely limits its use on datasets beyond a few thousand samples. This paper presents a set of GPU parallelization strategies, implemented in CUDA, that make Gabriel graph construction practical at scale. We introduce (i) an implicit closed-form mapping from a linear thread index to the corresponding vertex pair $(i,j)$, which removes the need for an explicit $O(N^2)$ adjacency structure and lets every thread locate its pair in constant time and constant memory; (ii) a two-pass, bitmask-based, atomically-compacted edge-list construction that produces a dense edge array proportional only to the number of edges actually found; and (iii) a thread-coarsening strategy in which each thread evaluates several candidate pairs, trading launch and scheduling overhead against load balance. We evaluate these strategies on the acceleration of a Gabriel-graph-based large-margin classifier (NN-clas) and on a novel Gabriel-graph clustering algorithm that identifies connected components on a residual graph and performs GPU-parallel Union-Find with hierarchical edge-based merging. On 14 public benchmark datasets, the GPU implementation reduces Gabriel graph construction time by two to nearly four orders of magnitude relative to a sequential CPU baseline (up to 8484x), without loss of classification accuracy, and the proposed clustering method is statistically competitive with DBSCAN on the topological validity index Dunn while eliminating the need for an explicit density radius parameter.

arXiv:2610.28716v1 Announce Type: cross Abstract: Effective decision-making in human-centric environments such as Intelligent Tutoring Systems (ITSs) and clinical healthcare requires policies that are adaptive, data-efficient, and ethically deployable. Apprenticeship Learning (AL) infers policies directly from expert demonstrations by recovering the underlying reward structure, enabling strictly offline learning without environment interaction. This dissertation develops a progression of offline AL frameworks organized around two challenges: reward modeling and demonstration quality. For reward modeling, EM-EDM clusters demonstrations by latent reward functions and induces cluster-specific policies, and THEMES partitions trajectories into temporally coherent sub-trajectories guided by a hierarchically learned reward regulator; across one educational and three clinical datasets, THEMES outperforms flat and hierarchical AL baselines. For demonstration quality, HALIDE integrates continuous quality signals into hierarchical policy induction, enabling principled reuse of mixed-quality data under distribution shift. CCHAL further attributes quality to individual decisions via Bayesian credit assignment, dual-arm criticality scoring that contrasts expert and non-expert behavior, and a Bradley-Terry contrastive IRL objective over failure trajectories, outperforming baselines including HALIDE on three clinical EHR datasets and an ITS dataset. Finally, a real-world classroom evaluation of StuEDM, an EDM-induced policy that preserves student agency and intervenes only on disagreement, shows significant learning gains over an Expert baseline and a higher pretest-adjusted posttest score than a CQL policy trained on a far larger corpus, while overriding students less than half as often. Together, this work establishes offline AL as a practical, data-efficient, and ethically viable paradigm for human-centric environments.

arXiv:2610.28709v1 Announce Type: cross Abstract: Evaluating the energy transmission ratio of an X-ray beam through a polymer composite is a challenging task in radiation shielding studies. The reference solution with Monte Carlo transport simulation is highly accurate but computationally expensive: each of our reference simulations is capped at $4\times10^9$ photon histories and a 20,000 s (over 5 hours) time limit, while closed-form analytical methods fail to represent such complex scenarios. We present an artificial neural network model that maps the unordered multi-set of molecular constituents, their mass fractions, and four bulk physical descriptors directly to the transmission ratio $I(t)/I_0$ in 1-3 milliseconds. The model is a Set Transformer, which makes it permutation invariant by construction and accepts mixtures of different sizes up to 11 constituents within a fixed-size padded set. Each constituent is embedded with a chemistry-pretrained transformer language model, ChemBERTa. In our proposed method, training data are produced by an automated PENELOPE/penEasy pipeline using the radiation source at 30, 80, 100, and 120 kV tube voltages over polyethylene-based shielding composite slabs of thicknesses spanning 0.15-2.5 mm. After preprocessing, the dataset contains $n = 4697$ simulated configurations on which a leakage-free, jointly stratified 10-fold cross-validation is performed, attaining mean squared error MSE = $(1.12 \pm 0.56)\times10^{-4}$, normalized-MSE percentage NMSE% = 0.0294% $\pm$ 0.0147%, and $R^2 = 0.9988$. The corresponding per-sample percentage errors give an alpha-trimmed ($\alpha = 0.10$) absolute percentage error (APE) of 1.66% and a median APE of 1.15%. Our model's inference is several orders of magnitude faster than the underlying Monte Carlo simulation (sub-millisecond per query vs. several hours for the reference run) with effective accuracy in predicting $I(t)/I_0$.

arXiv:2610.28699v1 Announce Type: cross Abstract: Reinforcement learning (RL) enables adaptive and autonomous decision-making in embedded, edge, and heterogeneous computing systems. Field-programmable gate arrays (FPGAs) and integrated circuits (ICs) offer low latency, energy efficiency, and architectural flexibility for RL deployment, while introducing challenges in resource constraints, implementation complexity, and hardware-algorithm co-design. This review surveys more than 100 studies on RL implementations across FPGA and IC platforms, emphasizing architectures, implementation strategies, performance evaluation, and applications. The literature is organized by RL family, including model-based Markov decision processes, classical value-based methods, deep RL, actor-critic, policy-gradient, and multi-agent approaches. We further classify studies by abstraction level, hardware realization strategy, and application domain, examining how RL algorithms map onto reconfigurable and fixed hardware. Unlike surveys restricted to particular algorithms or platforms, this work provides a unified analysis of FPGA and IC implementations, covering acceleration, reconfigurable architectures, energy-aware control, and system-level optimization. Comparative assessments consider resource utilization, latency, throughput, power consumption, and reported performance gains. Finally, we identify recurring architectural patterns, practical limitations, open challenges, and research directions for efficient RL-driven hardware systems. The survey offers a consolidated reference for researchers developing and evaluating RL solutions on specialized hardware.

arXiv:2610.28693v1 Announce Type: cross Abstract: Photoplethysmography (PPG) enables low-cost and non-invasive heart rate (HR) monitoring but remains sensitive to waveform distortion, physiological variability, and measurement interference. Temporal waveform characteristics and frequency-domain periodicity provide complementary information for HR estimation; however, effectively exploring and integrating these representations within a compact model remains challenging. This paper proposes TF-IG, a parameter-efficient interaction-gated time-frequency learning framework for PPG-based HR estimation. TF-IG employs a dual-branch architecture in which a temporal branch captures pulse morphology via dilated depthwise-separable convolutions, while a frequency branch extracts cardiac periodicity information from bounded log-spectrogram representations. To combine the two domains, we introduce an interaction-gated fusion module that explicitly models feature agreement and discrepancy, while also learning feature-wise weights to balance temporal and spectral information. A parameter-shared regression head further maps the temporal, spectral, and fused representations into the final HR prediction space. Experiments on the benchmark BIDMC PPG and Respiration Dataset using a subject-exclusive training, validation, and test protocol show that TF-IG achieves a mean absolute error (MAE) of 1.227 BPM, a root mean square error (RMSE) of 2.060 BPM, a Pearson correlation coefficient of 0.9934, and 92.28 percent of estimates within 3 BPM, while remaining parameter-efficient with only 84.6K parameters. Ablation results further demonstrate the complementary contributions of the temporal and frequency branches, interaction-gated fusion, and shared regression head. These results indicate that explicitly modeling cross-domain interactions provides an effective accuracy-complexity trade-off for lightweight PPG-based HR estimation.

arXiv:2610.28679v1 Announce Type: cross Abstract: Agents need reliable ways to validate their outputs, both when completing tasks and when filtering synthetic training data. We study verifier generation: from the instruction and environment available for solving a task, an agent writes an executable verifier that must accept every valid solution and reject invalid ones. We introduce VerifierBench, a benchmark of 222 tasks from six agent benchmarks spanning tool-call, state-based, and code verification, with 4,993 hidden test cases that include alternative valid solutions and attacks; Verifier4Verifier (V4V), a six-check audit, screens its source verifiers and test suites. Across 19 models in a common agent harness, normalized balanced accuracy ranges from 10.6 to 62.1, with generated verifiers consistently rejecting invalid solutions more reliably than they accept valid ones. This apparent strength can mask overly restrictive verification: most false rejections in the analyzed records come from verifiers that accept no tested solution to a task. Case studies reveal unsupported assumptions about solution formats, tool use, and code changes that exclude valid solutions. These errors survive self-testing because agents test solutions that share the verifier's mistaken assumptions, or submit verifiers without confirming that any correct solution passes. These findings suggest testing both whether a verifier accepts a correct solution and whether it accepts other solutions permitted by the task. The project page and dataset are publicly available.

arXiv:2610.28676v1 Announce Type: cross Abstract: The rapid evolution of digital circuit complexity demands innovative approaches to hardware design automation. While Large Language Models (LLMs) have influenced software development, their application to Hardware Description Language (HDL) generation remains largely unexplored, particularly for VHDL. This paper presents a comprehensive study for AI-assisted digital circuit design, addressing gaps in current Electronic Design Automation workflows. The work introduces a VHDL dataset of 18,500+ syntax-verified examples from open-source repositories and converted Verilog designs. Five 7B-parameter LLMs were fine-tuned using parameter-efficient QLoRA technique, enabling deployment in resource-constrained environments. To evaluate model efficacy, a rigorous three-level pipeline was established: (1) syntax and synthesis validation, (2) functional simulation with testbenches, and (3) FPGA prototyping on Zynq SoCs. Results show fine-tuned models achieve up to 92% syntax correctness and 73% functional accuracy on common design tasks, with DeepSeek-Coder demonstrating the highest performance gains. Generated designs were successfully deployed on real hardware, confirming not only syntactic but operational viability. This work demonstrates that LLMs, when tailored with domain-specific data and lightweight fine-tuning can contribute meaningfully to the future of generative EDA workflows.

arXiv:2610.28667v1 Announce Type: cross Abstract: Coding agents can now implement a specification, write and run tests, repair failures, and declare the result ready for release. When the same agentic system produces both the software and the evidence of its correctness, a passing test suite is no longer independent evidence: visible tests become an optimization target, code and tests may share one misinterpretation of the specification, and the builder's own tests may simply be incomplete. This paper argues that agent-produced software requires an assurance plane that is structurally independent of the build plane. We characterize four gaps between stakeholder intent and production (requirement, conformance, evaluation, and model gaps) and identify which of them verification can close. We propose a framework that separates building from verifying, introduces a Verification Contract stating what must be demonstrated before release, and organizes verification in a specification-first phase without access to the implementation and an implementation-aware phase based on mutation analysis, property-based testing, and fuzzing, with an assurance report that leaves the release decision to a human. We evaluate the framework in a pre-registered study on three stateful services, two of them written by third-party models, and an EvalPlus pilot. The builder's programs were correct, but its own tests missed subtle interactions: a verifier in a fresh context detected more of the 46 seeded faults than the builder's tests (Holm-corrected p = 0.006) and surfaced the ambiguities of the specification, at the price of rare false oracles. A verifier from another provider did not add detection.

arXiv:2610.28666v1 Announce Type: cross Abstract: Reviewer lottery error describes error due to a small set of peer reviews attending to only some of the issues with a paper that a larger pool of eligible reviewers would raise. We show this error can be reduced by surfacing a checklist of criticisms learned from a historical review corpus. Our approach generates a tree of checklists representing coarsenings of the atomic review claim space, then selects the coarsening that maximizes a theoretically-motivated measure of correlated agreement among reviewers. We realize this approach with an LLM-assisted pipeline and test it on NeurIPS 2021 and ICLR 2022 review data. In simulated review with LLM reviewers, the learned checklist reduces the mean squared error (MSE) of a three reviewer committee by 58% on NeurIPS and 21% on ICLR, and on both venues the checklist with the highest correlated agreement has the lowest error for committees of three or more. In a human study with 20 participants, the learned NeurIPS checklist lowers the MSE of a paper ranking, measured against the NeurIPS 2021 order, from 0.48 to 0.22.

arXiv:2610.28663v1 Announce Type: cross Abstract: A knowledge graph needs a stable address for its evidence even where the literature offers competing classifications and agrees on none. Yet an address chosen for retrieval readily passes for the only classification available. This happens whenever records gathered under one name concern units that differ in boundaries, rank, or the criteria by which they are described or counted. In Sinitic dialectology the same localities recur in schemes whose units differ in membership and rank, and a unit may change its affiliation when the evidence for its diagnosis changes. Two tone counts, each quoted faithfully, may still count different things. The taxonomy of an entire field thus comes to rest on the unstable frame of a literature that contradicts itself. Neither verifying a quotation nor searching for conflicting values detects these differences. I distinguish the working address that gathers candidate evidence, the category that the source defines and the decision that admits a passage to a particular comparison, and argue that such decisions, together with the dependence among records, belong to the provenance of a comparison. On four re-runnable cases, the accompanying PaRAGraph Casebook shows where such differences arise among claims extracted by a language model and verified before entering the graph. The platform now stores the corresponding decisions as objects for one dimension of comparability, the address of a record: the placement of a locality in each source's scheme, the decision that a name in a given publication denotes a unit rather than a place, and the origin of every address. The working taxonomy can then be revised as new sources yield new claims, while the path from each comparison to its evidence stays fixed. Keywords: knowledge graphs; source attribution; classification; claim extraction; comparative concepts; Sinitic dialectology; Yue

arXiv:2610.28658v1 Announce Type: cross Abstract: How much can we trust an agent to find the paper we need? As researchers, we search the literature from what we have read toward what we have not. We thus describe what we want in our own terms, which a paper may show only implicitly; our requirements grow as we read; and what we seek may lie in an empty region no paper yet covers. We introduce CompLit, a literature search benchmark spanning all eight arXiv domains, with one diagnostic setting for each of these needs. The strongest agents, Codex (GPT-5.6-Sol) and Claude Code (Opus 4.8), answer 86-87% of Standard queries, yet fall to 66-68% when requirements are implicit (implicit), lose another 6 and 15pp, respectively, as they accumulate (cumulative), and abstain on at most 53% of the queries no paper satisfies (unmet). No agent leads in every domain, and all are less accurate on papers from before 2019 or after 2023. Tracing their search, we find that agents often reach the right paper, or the closest one, yet misjudge it. Doing well on CompLit is a step toward agents that we can trust to find the paper we need, or to tell us that none exists.

arXiv:2610.28656v1 Announce Type: cross Abstract: Future-work statements offer a record of the questions researchers choose to leave open. Using them to describe a field, however, requires separating thematic change from changes in document coverage, topic assignment, and scholarly writing. FutureLens examines these measurement issues in a recovered snapshot of 5,655 ACL papers published during 2012--2022, including 4,005 with supplied future-work reference text. We compare non-negative matrix factorization (NMF) and latent Dirichlet allocation (LDA), aggregate topic presence at the paper level, and examine normalized prevalence, exploratory trend statistics, assignment stability, and agreement between input-derived text and references. Across available seeds, NMF has higher mean normalized pointwise mutual information (0.1387 versus 0.1280) and higher pairwise assignment stability, while LDA has greater keyword diversity (0.8900 versus 0.7750). Seed-zero results show increasing prevalence of broad model-training and question-related vocabulary, alongside substantial citation and document-structure content. Mean input/reference topic-set overlap is 0.5191 for NMF and 0.4924 for LDA; temporal rank agreement varies widely and does not imply agreement in prevalence levels. These findings support a descriptive account of the available corpus, with explicit qualifications for missing references, incomplete runs, and unverified reference provenance. The contribution is an empirical measurement study of longitudinal scholarly signals, not a new topic-modeling algorithm or a forecasting system.

arXiv:2610.28655v1 Announce Type: cross Abstract: Publishers and funders increasingly ask authors to disclose how AI tools contributed to a piece of research. In computational work the repository is where an author or an auditor would look for the answer, and we ask what it can actually supply. We audit two single-investigator repositories by the same author in the same field, each carrying a different class of agent residue: a prescriptive trace of standing rule files and a narrative trace of a running agent scratchpad, in repositories of 466 and 103 commits. The two are legible in opposite ways. 85.6% of the lines ever committed to the prescriptive trace are visible at the current revision, and a standing rule states a conclusion and discards the episode behind it; the narrative trace records episodes and shows 21.8%. Counted by revision the gap is far smaller: 15 of the prescriptive trace's 41 revisions leave no visible line, against 18 of the narrative trace's 41. The Co-Authored-By trailer, the field an automated audit reads, is one of three AI trailer keys Project A carries and covers 66 of the 121 commits carrying any. In Project B it has low sensitivity against an explicit within-repository proxy: 39 commits revise a file maintained to hold an agent's working notes and 1 (2.6%) carries the trailer. Project A's repository-level rate is unstable in time too, stepping from 2.7% to 17.7% across a boundary we did not specify in advance. Repository channels disagree, and neither rate can be read as an involvement rate; we do not isolate the cause, and say so. We give a field-by-field account of which provenance claims such traces support, which they cannot, and which they answer confidently and wrongly, construct an auditable evidence ledger, and derive the requirements this places on any system meant to record research provenance as it happens.