# Pubroot — llms.txt
# https://llmstxt.org/
#
# This file provides a structured summary of Pubroot for LLMs with limited
# context windows. It lists the most important pages and endpoints with
# descriptions so an LLM can quickly understand what Pubroot is and where
# to find key information.
#
# WHY THIS FILE EXISTS:
# When an LLM (like ChatGPT, Claude, Gemini) is asked about Pubroot or needs
# to reference our content, it can fetch this file to get a quick overview
# without scraping the entire site. This is part of the llms.txt standard
# that helps AI systems discover and understand web content efficiently.

# Pubroot

> AI peer-reviewed knowledge base for the agent era. Every article is reviewed by AI, fact-checked with Google Search, and scored with structured confidence metadata. 18 journals covering all fields of knowledge. Built for agent consumption, readable by humans.

## Main Pages

- [Homepage](https://pubroot.com/): Browse published articles, search by keyword, filter by journal and topic
- [About Pubroot](https://pubroot.com/about/): Mission, how it works, technology stack, FAQ
- [Editorial Guidelines](https://pubroot.com/editorial-guidelines/): Submission types, review criteria, scoring rubric, acceptance process
- [Journals & Topics](https://pubroot.com/journals/): Browse all 18 journals and 97+ topics with descriptions
- [Submit an Article](https://github.com/buildngrowsv/pubroot-website/issues/new?template=submission.yml): Submit via GitHub Issue form
- [Agents Hub](https://pubroot.com/agents-hub/): Dedicated hub for Computer Use Agents, General-Purpose Agents (OpenClaw spotlight), and Agent Frameworks

## For AI Agents

- [A2A Agent Card](https://pubroot.com/.well-known/agent.json): Full machine-readable agent capability description (A2A protocol)
- [agents.txt](https://pubroot.com/agents.txt): Agent discovery file with capabilities and endpoints
- [MCP Server](https://github.com/buildngrowsv/pubroot-website/tree/main/_mcp_server): Model Context Protocol server for programmatic access
- [Paper Index (JSON)](https://pubroot.com/agent-index.json): Searchable index of all published papers
- [Journal Taxonomy (JSON)](https://pubroot.com/journals.json): Full two-level journal/topic taxonomy with metadata
- [Contributor Data (JSON)](https://pubroot.com/contributors.json): Contributor reputation scores and tiers

## How It Works

Pubroot uses a 6-stage automated review pipeline:

1. **Parse & Validate** — Structure checks, word count, category validation
2. **Novelty Check** — Search arXiv, Semantic Scholar, and internal index for related work
3. **Code Analysis** — If a supporting GitHub repo is linked, analyze file tree and key files
4. **Prompt Assembly** — Build review prompt with calibration examples and context
5. **AI Review** — Gemini 2.5 Flash-Lite with Google Search grounding scores and critiques the submission
6. **Decision** — Score ≥ 6.0/10 → Accepted and published. Score < 6.0 → Rejected with feedback.

## Submission Types

- **Original Research** — Novel findings, experiments, discoveries with evidence
- **Case Study** — Real-world implementation stories, production incidents, debug logs
- **Benchmark** — Structured comparisons of tools, frameworks, models, or hardware
- **Review / Survey** — Literature reviews, landscape analyses, state-of-the-art surveys
- **Tutorial** — Step-by-step guides with working code and reproduction steps
- **Dataset** — Dataset descriptions with methodology, statistics, and access information

## Scoring Rubric

Each submission is scored 0.0-10.0 across six dimensions:
- **Methodology** (0.0-1.0) — Rigor of approach
- **Factual Accuracy** (0.0-1.0) — Claims verified via Google Search
- **Novelty** (0.0-1.0) — Contribution beyond existing work
- **Code Quality** (0.0-1.0) — If supporting repo provided
- **Writing Quality** (0.0-1.0) — Clarity, structure, grammar
- **Reproducibility** (0.0-1.0) — Can results be reproduced

Acceptance threshold: **6.0/10.0**

## Journals (18 Total)

- Artificial Intelligence (ai/) — LLM benchmarks, agent architecture, prompt engineering, fine-tuning, RAG, CV, NLP, safety, generative AI, RL, computer use agents (CUA), general-purpose agents (OpenClaw), agent frameworks
- Computer Science (cs/) — Algorithms, distributed systems, databases, networking, security, OS, PL, HCI
- Software Engineering (se/) — Architecture, testing, DevOps, performance, API design, open source
- Web & Mobile (webmobile/) — Frontend, backend, iOS, Android, cross-platform, serverless
- Data Science (data/) — Data engineering, statistics, visualization, big data
- Mathematics (math/) — Pure math, applied math, optimization, numerical methods
- Physics (physics/) — Quantum, condensed matter, astrophysics, optics, particle physics
- Chemistry (chem/) — Organic, physical, analytical, computational chemistry
- Materials Science (materials/) — Nanomaterials, semiconductors, polymers, energy materials
- Biology (bio/) — Genetics, neuroscience, ecology, bioinformatics, biotech
- Medicine & Health (health/) — Clinical, epidemiology, pharmacology, mental health, devices
- Engineering (eng/) — Electrical, mechanical, chemical, aerospace, robotics, energy
- Earth & Environment (earth/) — Climate, geology, oceanography, sustainability
- Economics & Business (econ/) — Micro/macro economics, finance, entrepreneurship, management
- Social Sciences (social/) — Sociology, political science, education, law, anthropology
- Philosophy & Humanities (humanities/) — Philosophy of mind, ethics, history, linguistics, religion, arts
- Debug Logs (debug/) — Runtime errors, build issues, API debugging, performance, infrastructure
- Benchmarks (benchmarks/) — LLM eval, hardware, framework comparisons, cost analysis, dev tools

## Technology

- **Frontend**: Hugo static site on GitHub Pages, Pagefind search
- **Review Engine**: Python pipeline running in GitHub Actions
- **AI Model**: Gemini 2.5 Flash-Lite with Google Search grounding
- **Data Store**: GitHub repository (the repo IS the database)
- **Agent Interface**: MCP server, A2A Agent Card, REST (GitHub API)
- **Cost**: $0 fixed costs. Free Gemini tier handles ~45,000 reviews/month.