Three immunoregulatory signatures define non-productive HIV infection in CD4+ T memory stem cells

The persistent HIV reservoir constitutes the main obstacle to curing HIV/AIDS disease. Our understanding of how non-productive HIV infections are established in primary human CD4+

Dispersal, adaptation and persistence of H5N1 in the sub-Antarctic and Antarctica

High pathogenicity avian influenza virus (HPAIV) H5N1 reached the sub-Antarctic and Antarctica in 2023, subsequently spreading to remote locations within this region where it had

ApeA cleaves genomic RNA to defend against RNA phage infection

To protect themselves against bacteriophage infection, bacteria encode a vast diversity of antiphage defense systems. However, the mechanisms of action of most of these systems

FASTERCC: Accelerating Flux Consistency Testing and Context-Specific Reconstruction for Large-Scale Metabolic Network Models

The increase in size of metabolic network models especially with the advent of single-cell data calls for scalable reconstruction and analysis tools. Such models, often

Acceleration and Velocity Dissociate Temporal Phases of Postural Control in Rhesus Macaques

Maintaining balance requires the nervous system to transform sensory signals about unexpected postural perturbations into precisely timed motor commands. Although human studies have established that

When the Specification Emerges: Benchmarking Faithfulness Loss in Long-Horizon Coding Agents

March 19, 2026

arXiv:2603.17104v1 Announce Type: cross
Abstract: Current coding-agent benchmarks usually pro- vide the full task specification upfront. Real research coding often does not: the intended system is progressively disclosed through in- teraction, requiring the agent to track durable design commitments across a long session. We introduce a benchmark for this setting and study faithfulne Ss Loss U nder eM ergent s Pecification (SLUMP), defined as the reduc- tion in final implementation faithfulness un- der emergent specification relative to a single- shot specification control. The benchmark con- tains 20 recent ML papers (10 ICML 2025, 10 NeurIPS 2025), 371 atomic verifiable compo- nents, and interaction scripts of approximately 60 coding requests that progressively disclose the target design without revealing the paper itself. Final repositories are scored with a five-level component-faithfulness rubric and accompanied by an exposure audit to verify that scored components are recoverable from the visible interaction. Evaluated on Claude Code and Codex, the single-shot specification control achieves higher overall implementation fidelity on 16/20 and 14/20 papers, respectively. Structural integration degrades under emergent specification on both platforms, while seman- tic faithfulness loss is substantial on Claude Code and small on Codex. As a mitigation case study, we introduce ProjectGuard, an exter- nal project-state layer for specification tracking. On Claude Code, ProjectGuard recovers 90% of the faithfulness gap, increases fully faith- ful components from 118 to 181, and reduces severe failures from 72 to 49. These results identify specification tracking as a distinct eval- uation target for long-horizon coding agents.

Subscribe for Updates

Copyright 2025 dijee Intelligence Ltd. dijee Intelligence Ltd. is a private limited company registered in England and Wales at Media House, Sopers Road, Cuffley, Hertfordshire, EN6 4RY, UK registration number 16808844