# LAR — Layered Agentic Retrieval

> A publisher-side architecture for agent-native web surfaces. This site is itself a LAR surface, published as a minimal demonstration of the pattern it documents. Working proposal at v0.1, not a settled standard, offered to the commons to be forked, renamed, reshaped, or absorbed into something better.

## Start here

- [/.well-known/lar.json](https://lar.md/.well-known/lar.json): the canonical manifest for this surface. Declares identity, context, operations and policies. This is the entry point; everything else is reachable from it.
- [Institutional identity](https://lar.md/lar/about.md): who publishes this surface, what LAR claims and expressly does not claim, and the open questions.

## The working paper

- [Layered Agentic Retrieval — full text, Markdown](https://zenodo.org/records/21622425/files/layered-agentic-retrieval.md): the paper itself, plain Markdown, no rendering pipeline required. The Appendix A.1 manifest schema in it parses as valid JSON.
- [Layered Agentic Retrieval — citable record](https://doi.org/10.5281/zenodo.21622424): open access, PDF and Markdown, CC BY-SA 4.0. Concept DOI, always resolves to the latest version. Text identical to the SSRN version (`10.2139/ssrn.6801118`).

## Specification and reference examples

- [Repository](https://github.com/lar-spec/lar): schema, two worked examples, changelog, contribution process. Apache-2.0.
- [JSON Schema (draft-07)](https://raw.githubusercontent.com/lar-spec/lar/main/schemas/lar.schema.json): the canonical `lar.json` structure.
- [Archived release](https://doi.org/10.5281/zenodo.20544009): citable snapshot of the specification repository.

## Discovery

This surface declares itself three ways, which is what the working paper's §8 proposes: a `LAR:` directive in [robots.txt](https://lar.md/robots.txt) alongside `Sitemap:`, a `<link rel="lar">` in the page head, and this file. Each reaches a different class of agent — crawler-class, HTML-rendering, and llms.txt-aware — and none is sufficient alone.