<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhong-xia-2026-tracefork">
    <titleInfo>
        <title>TraceFork: A System for Inspecting GUI-Agent Trajectory Forks</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Yuzhang</namePart>
        <namePart type="family">Zhong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Stephen</namePart>
        <namePart type="family">Xia</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Sweta</namePart>
            <namePart type="family">Agrawal</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Alberto</namePart>
            <namePart type="family">Abad</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-461-3</identifier>
    </relatedItem>
    <abstract>Autonomous GUI and web agents frequently generate long, multi-step execution traces with rich environmental states, making manual inspection and fault localization challenging. Existing visualization tools typically render isolated single-run trajectories, obscuring fine-grained behavioral discrepancies across parallel runs on the same task. We introduce TRACEFORK, an interactive system designed to align, compare, and inspect paired same-task trajectories produced by AI agents or human demonstrators. TRACEFORK ingests execution traces from standardized benchmarks (e.g., WebArena, VisualWebArena) and custom user logs via an extensible normalization schema. Using a multi-modal dynamic programming alignment engine, the system collapses shared prefixes onto a single trunk, isolates behavioral divergence points into distinct branches, and identifies downstream reconvergence via conservative rejoin cues. Across a library of 1,722 tasks and 1,387 aligned trajectory pairs (33,337 steps), TRACEFORK achieves a median alignment latency of 45.5 ms per pair (p95 320.4 ms). On a manually annotated pilot subset, it achieves an F1 score of 0.910 for divergence localization, offering an open-source, deterministic diagnostic workbench for the agent research community.</abstract>
    <identifier type="citekey">zhong-xia-2026-tracefork</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-demos.15/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>160</start>
            <end>166</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T TraceFork: A System for Inspecting GUI-Agent Trajectory Forks
%A Zhong, Yuzhang
%A Xia, Stephen
%Y Agrawal, Sweta
%Y Abad, Alberto
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-461-3
%F zhong-xia-2026-tracefork
%X Autonomous GUI and web agents frequently generate long, multi-step execution traces with rich environmental states, making manual inspection and fault localization challenging. Existing visualization tools typically render isolated single-run trajectories, obscuring fine-grained behavioral discrepancies across parallel runs on the same task. We introduce TRACEFORK, an interactive system designed to align, compare, and inspect paired same-task trajectories produced by AI agents or human demonstrators. TRACEFORK ingests execution traces from standardized benchmarks (e.g., WebArena, VisualWebArena) and custom user logs via an extensible normalization schema. Using a multi-modal dynamic programming alignment engine, the system collapses shared prefixes onto a single trunk, isolates behavioral divergence points into distinct branches, and identifies downstream reconvergence via conservative rejoin cues. Across a library of 1,722 tasks and 1,387 aligned trajectory pairs (33,337 steps), TRACEFORK achieves a median alignment latency of 45.5 ms per pair (p95 320.4 ms). On a manually annotated pilot subset, it achieves an F1 score of 0.910 for divergence localization, offering an open-source, deterministic diagnostic workbench for the agent research community.
%U https://aclanthology.org/2026.emnlp-demos.15/
%P 160-166

Markdown (Informal)

[TraceFork: A System for Inspecting GUI-Agent Trajectory Forks](https://aclanthology.org/2026.emnlp-demos.15/) (Zhong & Xia, EMNLP 2026)

• TraceFork: A System for Inspecting GUI-Agent Trajectory Forks (Zhong & Xia, EMNLP 2026)

ACL

• Yuzhang Zhong and Stephen Xia. 2026. TraceFork: A System for Inspecting GUI-Agent Trajectory Forks. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 160–166, Budapest, Hungary. Association for Computational Linguistics.