---
title: Reproduction — Multi-task Linear Regression without Eigenvalue Lower Bounds
emoji: 🛡️
colorFrom: purple
colorTo: green
sdk: static
pinned: false
tags:
  - icml2026-repro
  - paper-D5Ijcnz1L9
---

# Reproduction bundle

Claim-native deterministic audit of all six registered claims for OpenReview
`D5Ijcnz1L9` and arXiv `2605.17126v1`. The registered matrix-weighted MTLR
estimator is executed directly on linear and logistic experiments at the
paper's native synthetic dimensions.

```bash
python3 -m pip install -r requirements.txt
PYTHONDONTWRITEBYTECODE=1 PYTHONWARNINGS=error python3 reproduce.py
PYTHONDONTWRITEBYTECODE=1 PYTHONWARNINGS=error python3 validate_evidence.py
```

Finite native-scale runs and exact covariance objects accompany the pinned
high-probability theorems; they are not replacement universal proofs.