From 3c7dfd48b1c79220260d6302632a9e1b73fec3a5 Mon Sep 17 00:00:00 2001 From: Aditya Datta Date: Sun, 19 Jul 2026 15:55:35 +0530 Subject: [PATCH] docs: add authoritative further reading --- README.md | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) diff --git a/README.md b/README.md index acb9a83..ef689f8 100644 --- a/README.md +++ b/README.md @@ -151,6 +151,24 @@ Optional extensions to deterministic flows (e.g., ChainWeaver) or governance lay See [`docs/non-goals.md`](docs/non-goals.md) for scope boundaries that keep the lab focused on offline routing evaluation rather than live runtime ownership. +## Further reading + +These primary sources provide background for the patterns used or discussed in the +lab. Inclusion does not imply that this repository implements or validates every +method or framework below, and results from its synthetic offline data do not +establish production safety or governance. + +- [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629) + — the reasoning-and-action pattern behind many tool-using agents. +- [Model Context Protocol specification](https://modelcontextprotocol.io/specification/2025-06-18) + — the protocol contract for exposing tools, resources, and prompts to agents. +- [Doubly Robust Off-policy Value Evaluation for Reinforcement Learning](https://proceedings.mlr.press/v48/jiang16.html) + — a foundational reference for doubly robust off-policy evaluation. +- [Effective context engineering for AI agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents) + — practical guidance on selecting and maintaining useful agent context. +- [NIST AI Risk Management Framework 1.0](https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10) + — a voluntary framework for incorporating trustworthiness into AI risk management. + ## Keywords Agent evaluation · offline evaluation · off-policy evaluation · counterfactual