Contexts are Never Long Enough: Structured Reasoning for Scalable Question Answering over Long Document Sets
A new QA framework called SLIDERS handles document sets that outgrow any context window by storing extracted facts in a relational database and reasoning over them with SQL.
Script: GPT-5.4 Voice: Murf.AI Gen2
Transcript
Justy Long-document QA keeps hitting the same wall: the context window runs out.
Justy This paper says the real fix isn’t a bigger prompt. It’s structured reasoning over a persistent database of extracted facts.
Justy The system is called SLIDERS. It pulls salient information from many documents into relational tables, then reasons with SQL.
Justy The essential insight is the reconciliation step. [pause] Before querying, it repairs duplicate, inconsistent, and incomplete records.
Justy It does that using provenance, extraction rationales, and metadata, so local extractions become globally coherent.
Justy That matters because chunking documents only moves the problem downstream. You still need a clean way to combine evidence at scale.
Justy The results are strong too, including gains on collections reaching 3.9 million and 36 million tokens.
Justy Today, read arXiv 2604.22294 and focus on the reconciliation stage, not just the benchmark scores.
Justy Expect more QA systems to look like databases with reasoning, not bigger text windows.