Visible but Not Avoided: State-Dependent Re-Proposal of Rejected Candidates in Catalog-Based LLM Optimization Loops

Studies repeated rejected edits while optimizing a screenshot-only GUI agent, contrasting static probes with reached loop states and testing how rejection wording changes subsequent edit proposals.

AI-assisted editorial review against a primary source. This is a catalog annotation, not an independent reproduction or endorsement.

What is this resource?

Resource type

Research papers

Environment

unspecified

Topics

evaluation, grounding, training-free

Record publication date

2026-10-11

Evidence retrieved

2026-10-20T15:31:34.620763Z

Resource ID

arxiv:2610.15738

Recorded paper version

arXiv v1

Version updated: 2026-10-11T16:18:11Z.

This identifies the retained arXiv metadata snapshot; newly observed changes require separate review.

Authors

Jianlin Chen, Wenhui Chen, Chi Man Vong

Cite this paper

Download BibTeX · Download CSL JSON

Citations preserve the selected source edition, its authors and recorded date precision. Unrecorded venues and identifiers are omitted.

What licenses are recorded?

Code, training data, weights, and documents can have different terms. A blank declaration means unknown, not unrestricted use.

Code

Not recorded

Data

Not recorded

Document

CC-BY-4.0 (The retained version-specific arXiv HTML infobox explicitly declares License: CC BY 4.0 for this document.)

Weights

Not recorded

License review notes

• Document: Publisher document declaration only; no code, data or weights license inferred. Source 1

Explore related topics

Scope and limitations

The downstream test uses one screenshot-only agent, nine synthetic apps and 60 cached development items, with TAP-tag or text-answer outputs; this is not demonstrated live-app completion. Two catalogs reword the same 40 edits. Re-proposal effects do not establish internal causal mechanisms or task-success gains. Main serving providers were unpinned; rationale labels came from LLM annotators. No independent reproduction.

Primary evidence and verification

Source retrieval is separate from experiment reproduction. Use the original publication or publisher page when citing claims about the resource.

1. arxiv api

   Retrieved 2026-10-20T13:48:10Z. Supports: name, authors, published_at, updated_at, identifiers.arxiv, details.arxiv_snapshot.

   Content fingerprint

   6d8c0463b6d50320246021f0394d0e39c9fd839f1ef1f742c8cdfa231872b059
2. editorial primary review

   Retrieved 2026-10-20T15:31:34.620763Z. Supports: name, kind, summary, environments, topics, scope, notes, licenses.document.

   Content fingerprint

   7ec947ceb462b5fd23a4a67df64e931e427224ef721b681d3fc92c1d4379449f

Suggest a correction

Open a source-backed correction. Include this record ID and the original source. Read our editorial policy and coverage boundaries.