CS Paper Radar

What the Author Wrote, What the Agent Reads: Silent Cue Loss in Enterprise Document Ingestion

Yikai Zhao, Saurabh Pandey, Pradeep Kumar Misra

cs.CL cs.AIcs.IR ·submitted 2026-10-21 16:51 UTC · v1
benchmark

Abstract

Enterprise agents act on documents. Before an agent can follow a policy, answer a compliance question or file a ticket, some ingestion layer must turn a Word file or PDF into tokens - and that layer is where an agent deployed in the wild silently loses information. We show that one layer is lost by every mainstream converter - the meaning an author encodes in how text looks - and that the loss makes agents confidently wrong rather than visibly incomplete. We build a task from 76 real policy documents: recover the requirements the author singled out, scored against the file's own OOXML properties by a deterministic matcher with no LLM judge. Formatting-blind ingestion collapses from micro F1 0.690 to 0.150 (docling) and 0.075 (plain text), a gap that replicates across six answering models from two vendors - five Claude models spanning the Opus, Sonnet and Haiku tiers, and OpenAI's GPT-5.6. An agent that cannot see the cue guesses, and docling's false-positive rate on requirements the author did not flag is four times ours. A span-level leave-one-out localises the loss to two cues - highlight (+0.302) and colour (+0.262), both Holm-adjusted p < 1e-4 - while bold, italic, monospace and strikethrough each sit within +/-0.003 of zero. The same asymmetry appears in PDF. Colour is also the cue no mainstream converter preserves and no parsing benchmark scores, and it is common: 32.8% of 6,460 documents carry a deliberate colour or highlight, 10.3% on an obligation. A pre-registered test shows that typing a cue for the agent adds +0.085 beyond merely preserving it (Holm-adjusted p = 0.049). We validate the labels against rendered page images through an independent modality (kappa = 0.72) and bound what wording alone can achieve (AUC 0.664). For agent builders the practical finding is that ingestion, not reasoning, is where this reliability is won or lost, and that the fix is cheap and measurable.

Comment
28 pages, 5 figures

Closest papers

Nearest neighbours by meaning across the whole index.

Agents That Edit Documents: Measuring Agentic PDF Forgery Against a Non-Agentic Control

cs.AI cs.CR ·2026-09-20 ·Simiao Ren, Ankit Raj, Tommy Duong, Yuxin Zhang, Dennis Ng +4 more
similarity 0.68
Abstract · 15 pages, 12 figures, 2 tables

AI agents that carry a multi-step computer task through on their own became ordinary tools in the past year, and the same autonomy is available to anyone whose task is harmful. We ask what that means for a relying party -- an insurer, a lender, an auditor -- whose evidence is a filed PDF. AgentForge-Bench measures how reliably an off-the-shelf coding agent, driving one of seven open-weight models with a shell and the stock Python PDF stack, alters one dollar amount, date or address in a real filed financial document from a single sentence of intent, graded by rules rather than by a model. Across 1,750 cells, 1,419 (81.1%) satisfy the verifier, and 808 (46.2%) also survive every stricter filter: visible, localized, typeface-matched, original value gone document-wide. A deterministic script with no model in it solves 98 of the 125 documents; the agents solve 124, and none the script solves alone. Agents misreport 41% of their wrong edits as done, no model refused, and the cheapest verified forgery costs 2.4 cents. The raw rate overstates the threat by about a factor of two; the strict rate is still large.

0.43

Structure for Reading, Prose for Writing: Asymmetric Structural Conditioning in Multi-Agent Document Authoring

cs.AI cs.IR ·2026-08-21 ·Cheng Yu, Nikhil Mathew, Zhengjie Wang
similarity 0.65
Abstract · 10 pages, 3 figures

Multi-agent pipelines that author formal documents must both read a requester's forms and write against them. We report a deployed tender-response system, running an open-weights model under sovereignty constraints, and evaluate it against human-written bids the same organisation actually submitted. On a blind comparison where the system had no worked example available, an LLM judge rated its answers at least as good as the human-submitted answer on $40$ of $55$ ground-truth sections, better on $4$, missing on none, and flagged one unsupported claim in total. Classifying every gap the judge identified shows that $68\%$ were content absent from the system's own sources -- knowledge the human author held and the pipeline was never given -- so only $6$ of the $15$ adverse verdicts involve a deficiency the system could have avoided. A divergence from ground truth is more often an information-availability result than a writing-quality one, and evaluations that do not separate the two understate such systems. Against this backdrop we report a conditioning asymmetry. It is well established that rendering documents as structural markup rather than flat prose improves extraction, and we reproduce that on three reading tasks. The benefit does not transfer to conditioning: converting a bid's \emph{instruction} material from prose to nested XML dropped answer quality from $74\%$ to $48\%$ under a paired comparison. We further find that naming a forbidden construction concentrates rather than removes it -- $96\%$ of surviving defects fall in the two forms the prompt explicitly names -- and that coupling a stochastic annotation to a deterministic windowing function moves the extracted requirement count from $68$ to $51$ on a byte-identical file. Structure belongs where the model reads; prose and self-applied tests belong where it writes.

0.20

Who Verifies the Tools? How Warped Formatters Mislead Agents and Their Verifiers

cs.AI cs.SE ·2026-10-23 ·William Jeffries
publishedcode/datahot direction similarity 0.64
Abstract · Accepted as a poster at the Who Verifies the Agents? workshop at NeurIPS 2026 (non-archival). 12 pages, 3 figures. Code and data: https://gi…

Agents and verifiers often rely on the text a tool prints, and that text can misrepresent what the tool actually did, even when the tool's computation is correct. We saw this happen in a production RAG research system used by the Pilsudski Institute in New York, where tool output formatters misled an agent into giving wrong or incomplete information to users. Each tool did its job correctly, but its output omitted search limits, result caps, or identifier namespaces. We also recreated an unlabeled estimate caught in code review before release. We studied whether these formatters misled agents and verifiers and how much the repaired wording helps. We fed tool results to agents using two formatter versions: one with the key omissions and one with all missing details clearly stated. That difference alone changed what the agents claimed and what the verifiers accepted. Unsupported agent claims dropped substantially for three of the four failure types. In one test, verifiers went from accepting incorrect answers 90 out of 90 times to rejecting them 90 out of 90 times. A control showed the verifiers could catch the errors once they were given the missing facts and asked to judge correctness. We include the audit prompt we used with Codex and Claude Code in an exploratory demonstration, written with prior knowledge of our own system's defects. When pointed at an old version of our own system, it rediscovered the defects that were still live and surfaced twelve more we had not catalogued. When pointed at OpenClaw and Hermes, two of the best-known open-source agentic systems, it found warped formatters in both. We opened PRs to fix both, and the OpenClaw fix has already been merged. Practitioners can run the same prompt on their own systems today.

0.51

Quantifying Overclaiming Propensity in Frontier LLM Agents

cs.SE cs.AIcs.LG ·2026-09-17 ·Nolan Smyth, Yorguin-Jose Mantilla-Ramos, Pascal Jr Tikeng Notsawo, Saskia Helbling, Alberto Tosato +4 more
highly cited similarity 0.63
Abstract · 28 pages, 7 figures, 8 tables

Frontier coding agents are increasingly trusted to work autonomously for long periods of time, yet what they actually did is often hard to tell from their final response. We quantify the propensity of such agents to overclaim task completion, which may mislead the user. We operationalize overclaiming as a final response that reports work that the agent's own transcript shows it did not do, for example, claiming to have read a file it never opened. This criterion requires no inference about intent and does not depend on whether the delivered work is correct; it asks only whether the reported work was done. We introduce OverclaimBench, an evaluation suite of five file-review scenarios with transcript-based coverage measurements and registered planted defects. We evaluate eight proprietary frontier models in their own production command-line interfaces and four open-weight models under a single fixed harness, and find that 1) agents fail to read every file they were asked to review in 67.9% of runs; 2) among these incomplete runs, agents are misleading 80.4% of the time (59-96% per model), either falsely claiming a complete review or leaving the gap undisclosed; 3) requiring delegation to subagents increases coverage, but a large majority of reviews that remain incomplete are still misleading; and 4) agents that falsely claim a complete review miss planted defects at about 1.8 times the rate of agents that read every file, showing that claims of completion can conceal substantive failures. Together, these results show that agents' final responses are not reliable accounts of their actions.

0.48

Beyond the Document: Compiled Context and Governed Agents for Any Work Product

cs.SE cs.AIcs.IR ·2026-11-09 ·James Rhodes, George Kang
similarity 0.63
Abstract · 8 pages, 1 figure, 4-package held-out evaluation; 12,000 evaluated responses

A document preserves the visible result of professional work but often discards the research, calculations, assumptions, evidence, and authorization boundaries behind it. We study a different unit of delivery: a Portable Agent Format (PAF) package that carries a work product together with compiled context, evidence, provenance, and rules for subsequent interaction. To test whether this context remains useful beyond the process that produced it, we evaluate four new packages, created after the hosted and offline readers were fixed: an investment proposal, fund research, financial-statement analysis, and an engineering incident review. Two packages were produced by specialized agents and two through a general-purpose assistant's packaging skill. Across 400 frozen questions and 12,000 evaluated responses, a hosted governed reader succeeds on 96% of questions versus 86% with instructions plus raw-file retrieval (Holm-adjusted p < 10^-7). On the 260 answerable questions per condition, success is 94% versus 79%; the gap varies by package. Source-material chat and instructions alone reach 48% and 74% overall. A separate 0.8-billion-parameter offline reader succeeds on 69% overall, but 54% on answerable questions. Its -1 response rate is 2.9%, combining materially wrong and unauthorized answers, versus 56% with the source and instructions alone. Ablations suggest advantages for the integrated selector and package-adapted writer, without isolating representation from training effects. The same run exposes a failure: offline retrieval resolves none of the 40 historical-record questions in one financial-data package. These findings support treating compiled knowledge as part of the delivered work product, while leaving independent standards interoperability and end-to-end governance conformance unproven.

0.29

Fidelity Is Not Enough: Dispatch-Level Instrumentation for Agentic Datasheet Extraction

cs.CL cs.AI ·2026-08-28 ·Qing Ye, Meng-Hsuan Lin
top venuebenchmark similarity 0.62
Abstract · Accepted at EMNLP 2026 Industry Track. 7 pages + appendices

One model passed our fidelity check without ever opening the datasheet. We found it while qualifying models for an internal extraction service: a structured-output constraint had silently disabled tool use, and the model answered anyway, with fabricated source text. Only the per-tool trace exposed it. Fidelity -- whether an extracted value matches the source -- is the standard measure for agentic document extraction, and it scores that run a success. We therefore log every tool call in an agentic benchmark of 25 hand-curated claims over three components, with 12 more on a fourth, 37 in all. From that dispatch record we build two instruments: a rule-based failure-attribution classifier, and a silent-failure detector whose two rules check only which tools were called, never the extracted value. The detector raises no flag on 207 clean fidelity-passing extractions across three model families, and recovers all 50 planted faults that withhold exactly the tools its rules check. The two results are not symmetric: the first bounds the false-positive rate, the second is recall by construction, and detection power against runs that call their tools and still answer wrongly is unmeasured. A second, independent oracle, a causal chamber that tests whether the datasheet's claims hold under physical measurement, is intentionally partial: it confirms only what the apparatus can exercise, a verifiable envelope of 2 of those 37 claims, and we give a taxonomy of why the rest are not physically gradable. Under a controlled perturbation, fidelity passes throughout while the chamber verdict flips exactly at the measurement uncertainty. Across three deployed model stacks (one destabilised by its serving stack, not by any capability gap) the tool layer buys portability and observability rather than accuracy, and earns its premium only once a document outgrows the context window.

0.37

mimeo: Compiling Public Expert Corpora into Agent Skills and Testing What Transfers

cs.AI ·2026-08-31 ·Timothy Kassis
code/data similarity 0.62
Abstract · 42 pages, 6 figures. Toolkit and expert profiles: https://github.com/K-Dense-AI/mimeo

Giving an agent a file about a named expert can supply hard-to-find material, produce a recognizable persona, or change what the agent decides. These are different claims. We test each one. mimeo is an open-source tool that finds a person's public work, checks each extracted quotation against the cached source text, and writes a file an agent can load. Eight logged builds averaged 38 model calls; the check rejects 13.2% of extracted quotations. We tested four expert files with one coding-agent harness. Knowledge access was clearest: mimeo answered all 20 obscure, quotation-heavy questions; no closed-book condition answered more than 10. Keyword search (BM25) over the same pages answered 15-17, a gap this sample cannot resolve. Grounding showed one clear benefit: personas written from model memory misstated a documented position on 1-4 of 20 answers under every grader; the plain agent and mimeo never did. Every persona was easy to spot on short open prompts, and adding task material lowered identification by 18-23 points. mimeo was no more identifiable than a from-memory profile. Judgment transfer remained unresolved because both tests hit their ceiling: every condition found 94-97% of the problems planted in engineering tasks and scored 94-100% on 16 new application scenarios. An AI-judged "sounds like the expert" score changed with the judge: two of four preferred answers based on a model's stereotype, while two found no difference on the same text. That is a caution against relying on a single AI judge. The evidence supports mimeo as a compact, inspectable reference on a person, not as a demonstrated transfer of their judgment. Toolkit and expert profiles: https://github.com/K-Dense-AI/mimeo

0.51

The Interface Is Downstream: Designing the Terms of Human-Agent Collaboration

cs.HC ·2026-09-23 ·Hector Ouilhet Olmos
similarity 0.61
Abstract

Before an agent responds or acts, much of the experience has already been designed. Memory and retrieval shape what it notices. Evidence rules shape what it may claim. Permissions shape what it can do. Learning rules shape what it carries into the next encounter. The argument comes from Alicia, a personal agent I've built and used since January 2026. A fine-tuning pilot produced no defensible model-performance result. It exposed a provenance failure: Alicia repeated an interpretation from a retrieved synthesis, cited a source note credited by that synthesis, and left the synthesis out of the visible chain. In a model-blind review of thirty-three citations, one reviewer judged that the retrieved intermediary supplied the claim in sixteen relayed citations and part of it in four. Five of twelve citations to directly retrieved targets lacked support in the target excerpt supplied for review. These judgments remain unadjudicated, and the packet is not public. I call the shared setting a humorphic environment: a persistent computational setting that translates a human practice into software. The first Humorphism paper translated partnership. This paper translates the studio, the room where practice happens. The failure prompted an audit of attention, evidence, action, and learning, with a review artifact and available recourse for each. The test is whether the person can inspect and contest what shaped the teammate's behavior. The output is downstream. Correction, consent, and learning carry the collaborative interface back upstream.

0.11