# Record Grouping Controls Evidence Weight in Language Models

arXiv: https://arxiv.org/abs/2609.08698 · HTML: https://mlsanity.com/paper/2609.08698/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 25 references that are in the map.
- First posted on arXiv 8 Sep 2026.

## Builds on (most cited first, 22 of 22)

- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Mistral 7B](https://mlsanity.com/paper/2310.06825/) (2023)
- [Phi-4 Technical Report](https://mlsanity.com/paper/2412.08905/) (2024)
- [Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs](https://mlsanity.com/paper/2503.01743/) (2025)
- [The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions](https://mlsanity.com/paper/2404.13208/) (2024)
- [Entity-Based Knowledge Conflicts in Question Answering](https://mlsanity.com/paper/2109.05052/) (2021)
- [Tug-of-War Between Knowledge: Exploring and Resolving Knowledge Conflicts in Retrieval-Augmented Language Models](https://mlsanity.com/paper/2402.14409/) (2024)
- [Context-faithful Prompting for Large Language Models](https://mlsanity.com/paper/2303.11315/) (2023)
- [What Evidence Do Language Models Find Convincing?](https://mlsanity.com/paper/2402.11782/) (2024)
- [Seeing Things from a Different Angle: Discovering Diverse Perspectives about Claims](https://mlsanity.com/paper/1906.03538/) (2019)
- [GEO: Generative Engine Optimization](https://mlsanity.com/paper/2311.09735/) (2023)
- [Resolving Conflicting Evidence in Automated Fact-Checking: A Study on Retrieval-Augmented LLMs](https://mlsanity.com/paper/2505.17762/) (2025)
- [Task Matters: Knowledge Requirements Shape LLM Responses to Context-Memory Conflict](https://mlsanity.com/paper/2506.06485/) (2025)
- [Whose Facts Win? LLM Source Preferences under Knowledge Conflicts](https://mlsanity.com/paper/2601.03746/) (2026)
- [Auditing Provenance Sensitivity in LLM Agent Action Selection](https://mlsanity.com/paper/2607.20827/) (2026)
- [How Much Can We Trust LLM Search Agents? Measuring Endorsement Vulnerability to Web Content Manipulation](https://mlsanity.com/paper/2606.16821/) (2026)
- [Rational Synthesizers or Heuristic Followers? Analyzing LLMs in RAG-based Question-Answering](https://mlsanity.com/paper/2601.06189/) (2026)
- [How Adversarial Environments Mislead Agentic AI?](https://mlsanity.com/paper/2604.18874/) (2026)
- [How retriever redundancy and diversity impact RAG effectiveness](https://mlsanity.com/paper/2608.13956/) (2026)
- [Establishing Knowledge Preference in Language Models](https://mlsanity.com/paper/2407.13048/) (2024)
- [Mergeable Model-Side Aggregation States for Long-Context Language Models](https://mlsanity.com/paper/2607.26448/) (2026)

## Abstract (arXiv)

Retrieved records are presentation units; a supplied partition determines which records enter a language model as one evidential contribution. We characterize the invariant group-content state that removes within-group copies while retaining complementary canonical content, show that equal group counts can encode different evidence states, and derive a sharp content-aware partition-error bound. Given a supplied partition, our pre-generation representation deduplicates and aggregates content within groups and bounds each group's contribution. Across 104,402 trials and 6 public checkpoints, a central natural-text intervention finds that content-fixed false splits add 10.27-32.66 percentage points and false merges remove 9.13-31.79 points; a matched six-slot control retains the positive direction in all 16 cells. In a new 48-item controlled campaign panel, changing the supplied partition produces measurable, checkpoint-dependent decision shifts across all four models, and the balanced mirror design exposes substantial order interactions. Together, the theory and experiments establish the supplied partition as a controllable pre-generation representation variable and characterize its checkpoint-dependent behavioral effects.