Beyond the Parameters: ICL to Causal RAG (April 2026)

Free

Comprehensive survey treating context enrichment as a continuum — from in-context learning through RAG, GraphRAG, to CausalRAG; includes claim-audit framework and cross-paper evidence synthesis

FreeFree tier

Type

Open Source

Visit Website GitHub (40 stars)

About Beyond the Parameters: ICL to Causal RAG (April 2026)

This survey provides a unified account of augmentation strategies for large language models along a single axis: the degree of structured context supplied at inference time. It covers in-context learning, prompt engineering, Retrieval-Augmented Generation (RAG), GraphRAG, and CausalRAG. The paper includes a transparent literature-screening protocol, a claim-audit framework, and a structured cross-paper evidence synthesis that distinguishes higher-confidence findings from emerging results. It concludes with a deployment-oriented decision framework and concrete research priorities for trustworthy retrieval-augmented NLP.

Key Features

Unified account of augmentation strategies along a single axis of structured context

Covers in-context learning, prompt engineering, RAG, GraphRAG, and CausalRAG

Transparent literature-screening protocol with reproducible methodology

Claim-audit framework to distinguish higher-confidence findings from emerging results

Structured cross-paper evidence synthesis

Deployment-oriented decision framework for practitioners

Concrete research priorities for trustworthy retrieval-augmented NLP

Pros & Cons

Pros

Comprehensive coverage of the full spectrum from ICL to CausalRAG
Structured and transparent methodology including literature-screening and claim-audit
Includes actionable deployment decision framework for practitioners
Distinguishes high-confidence findings from emerging results
Concise yet informative (7 pages with 4 tables)

Cons

Limited depth due to short length (7 pages) – survey breadth over detail
No empirical experiments or performance benchmarks included
No software implementation or code provided
May not cover latest developments after April 2026
Primarily a survey paper, not a practical tool for direct use

Best For

Academic research on contextual enrichment strategies for LLMsComparative analysis of RAG variants (RAG, GraphRAG, CausalRAG)Guiding deployment decisions for trustworthy retrieval-augmented systemsIdentifying research gaps and future directions in LLM augmentation

FAQ

What is this paper about?

It is a technical survey that provides a unified account of contextual enrichment strategies for large language models, ranging from in-context learning to causal retrieval-augmented generation.

What augmentation methods are covered?

The paper covers in-context learning, prompt engineering, Retrieval-Augmented Generation (RAG), GraphRAG, and CausalRAG.

Who are the authors?

The authors are Prakhar Bansal and Shivangi Agarwal.

What is the claim-audit framework?

It is a framework to systematically distinguish higher-confidence findings from emerging results in the surveyed literature.

What practical guidance does the paper offer?

It includes a deployment-oriented decision framework and concrete research priorities for trustworthy retrieval-augmented NLP.