The Shelf · Signal Library
No. 202

ReproBridge: refined artifacts raise agent reproduction success from 76.4% to 92.6%

StrongEvidence grade
Nov 6, 2026Shelved
Agentic AISection
sha256:b0e60f5bf1b4…Fingerprint

instead of leaving coding agents to guess missing details from raw paper code, ReproBridge builds a standardized artifact (refined codebase, environment setup, launchers, reproduction guide) and has a separate Auditor check execution evidence. Across 1,029 tasks from 36 papers and four models, refined artifacts lifted mean success to 92.6%, cut downstream API spend 19.7%, and narrowed the cross-model spread from 21.6 to 3.9 points.

Read the source — arXivBrowse the Shelf

Cite as — Sawdust Shelf v2026-11-06.1 · No. 202 · fingerprint sha256:b0e60f5bf1b4…
Share image rendered from the living shelf — the number, the grade and the fingerprint travel with the card.