ProCreations's picture
repair claims 3-6: falsifications from the paper's own tables, seeded runs for 5-6
988e5c8 verified
|
Raw
History Blame Contribute Delete
810 Bytes
metadata
title: Reproduction  Scalable Table Data for Long-Context Reasoning
emoji: 📐
colorFrom: blue
colorTo: indigo
sdk: static
app_file: index.html
pinned: false
tags:
  - trackio
  - trackio-logbook
  - open-experiment
  - icml2026-repro
  - paper-Dx3Z5QGAZB
  - arxiv:2603.21719

Scalable Table Data for Long-Context Reasoning

Six claim-matched checks: exact mutual-information enumeration and seeded sampling for the two theoretical claims, plus exact released-measurement reconstruction for every numerical benchmark and ablation claim.

PYTHONDONTWRITEBYTECODE=1 PYTHONHASHSEED=0 PYTHONWARNINGS=error python3 -W error reproduce.py --out replay
PYTHONDONTWRITEBYTECODE=1 PYTHONWARNINGS=error python3 -W error validate_evidence.py
PYTHONDONTWRITEBYTECODE=1 python3 verify_manifest.py