# How this site is made

> Every page is built from the projects' source code at a pinned commit, not from vendor marketing.

1. **Catalog.** One YAML file lists the categories and the GitHub repositories in each.
2. **Pinned clones.** Each repository is cloned and its commit SHA recorded; every link points at that SHA.
3. **Machine wikis.** Two open-source DeepWiki generators — deepwiki-open (RAG over embeddings) and OpenDeepWiki (agentic) — run on every clone with DeepSeek V4 Flash. Their output is raw research material, never published verbatim.
4. **Question answering.** A code-reading agent (Claude Code driving DeepSeek V4 Flash in an isolated container, clone mounted read-only) answers each category's "how is X implemented?" questions with file/line citations, or says *not applicable* / *insufficient evidence*.
5. **Citation check.** Every citation is resolved against the pinned commit (file exists, line range in bounds). Invalid ones are dropped; an answer left with none is re-asked, and not published if it still has none.
6. **Editorial pass.** A stronger model writes overviews and cross-project verdicts from the wikis, answers and code. Where they disagree with the code, the code wins; worker mistakes get an *editor's note*.

## What to trust

Citation checks prove the cited code exists at the pinned commit, not that the prose is right. Follow the citation links.

## Inclusion rules

Vendor satellite repositories (SDKs, example collections, n8n nodes) are skipped when the main repository is listed. Each project has one primary category.

Canonical page: https://llms-technical-reviews.com/about/