Source Research Paper
73 words
4 sentences
NLP Salience & Cue Spans
(Red: Problem | Blue: Method | Green: Outcome | Amber: Metric)
Distilled Research Artifact
68% compression
24s saved
Executive TL;DR
Inkling-Small Distillation
A multi-agent open-weights framework autonomously reproduced 84.2% of benchmarks across 120 ICML 2026 papers with 3.4x less human labor.
Core Problem
Problem: Empirical ML benchmark reproduction is labor-intensive and difficult to scale across major conference publications.
Technical Methodology
Method: Multi-agent verification framework inspecting code repositories and running discrepancy diagnostics.
Key Quantitative Finding
Outcome: Achieved 84.2% autonomous reproduction rate on 120 papers with a 3.4x reduction in human verification time.
Extracted Quantitative Anchors
Success Rate:84.2%
Speedup / Ratio:3.4x
Paper Sample:120 papers
Venue:ICML 2026