Reproduction Beyond Benchmarks: ConstBERT and ColBERT-v2 Across Backends and Query Distributions

TL;DR

This paper evaluates the robustness of ConstBERT and ColBERT-v2 across different query types and backends, revealing architectural limitations in multi-vector retrieval models that affect reproducibility and performance.

Contribution

It demonstrates that architectural constraints, not just training data or tuning, limit multi-vector retrieval models' effectiveness across diverse queries and backends.

Findings

ConstBERT reproduces within 0.05% MRR@10 on MS-MARCO.

Performance drops 86-97% on long, narrative queries.

Backend parameters and fine-tuning can cause significant performance gaps.

Abstract

Reproducibility must validate architectural robustness, not just numerical accuracy. We evaluate ColBERT-v2 and ConstBERT across five dimensions, finding that while ConstBERT reproduces within 0.05% MRR@10 on MS-MARCO, both models show a drop of 86-97% on long, narrative queries (TREC ToT 2025). Ablations prove this failure is architectural: performance plateaus at 20 words because the MaxSim operator's uniform token weighting cannot distinguish signal from filler noise. Furthermore, undocumented backend parameters create an 8-point gap due to ConstBERT's sparse centroid coverage, and fine-tuning with 3x more data actually degrades performance by up to 29%. We conclude that architectural constraints in multi-vector retrieval cannot be overcome by adaptation alone. Code: https://github.com/utshabkg/multi-vector-reproducibility.