Overview

COVE is a research project investigating task-adaptive, code-based verification of large-language-model outputs. The central motivation is simple: generated answers should be examined with explicit verification mechanisms where possible rather than treated as correct by default.

The public preprint introduces the research direction. For the exact methodology, evaluation, results, and scope, consult the archived publication.

Research question

How can verification be made more proactive, task-sensitive, and operationally useful within LLM-powered systems?

This page is an accessible research overview, not a substitute for the paper. The archived preprint is the source of record for technical claims.

Authorship & publication

Anjaney Shrivastava and Dharmala Hrudai Aditya.

First made public on October 25, 2026 via Zenodo. This is a preprint and should not be described as peer-reviewed unless a later peer-reviewed version is published.

Cite the work

For a reliable bibliographic record, use the citation information provided on Zenodo. DOI: 10.5281/zenodo.23636512.