Grounded Revision vs. Prior Injection: Probing Retrieval-Augmented Patent Claim Amendment

2026-09-28

AI Summary

This study investigates whether retrieval-augmented generation (RAG) provides substantive grounding or merely injects templates in patent claim amendment. Using USPTO examination data, we compare random and structure-matched retrieval strategies, constructing a deterministic five-channel metric system and a four-quadrant taxonomy that require no LLM-assisted evaluation. Experiments are conducted using an XML-aligned corpus and pre-registered invocation techniques. Our findings reveal that mainstream models do not exhibit classic prior injection behavior; retrieval effects remain weak and directionally inconsistent, exposing model-specific locality differences. This work provides empirical evidence for delineating the operational boundaries of RAG in specialized text editing.

Abstract

Retrieval-augmented generation is widely used in professional writing, yet whether retrieval grounds revision or merely injects templates is rarely tested where "correct" has a definable meaning. Patent claim amendment supplies that signal: the examiner names the attacked limitation and cites prior art, providing per-case ground truth. We release three artifacts: (i) a corpus of 7,385 USPTO prosecution cases with XML-aligned pre/post claims, rejection, and cited prior art; (ii) a seven-probe battery comparing random and structural-match retrieval as two policies under a fixed prompt scaffold; (iii) a deterministic five-channel metric (C1-C3 and C5 in main, C4 supplementary) requiring no LLM evaluation. Across 9,600 pre-registered calls on four frontier LLMs (Claude Sonnet 4, Claude Haiku 4.5, GPT-5.4, GPT-4o-mini), no tested model exhibits detectable classical prior-injection behavior; retrieval effects are small and direction-inconsistent between random and structural retrieval, and the null is unchanged under a dense (semantic) retriever, across retrieval depths k in {1,3,5,10}, and under a paraphrase-sensitive grounding metric. Revision locality reveals a model-specific difference that the template channel misses. The four-cell taxonomy, which we treat as exploratory, leaves the prior-injector cell unoccupied.

Problem

• Retrieval-Augmented Generation
• Patent Claim Amendment
• Prior Injection
• Grounded Revision

Innovation

• Retrieval-Augmented Generation
• Patent Claim Amendment
• Deterministic Evaluation Metrics
• Prior Injection Probing
• Revision Locality