Protologue

Protologue

A sourced, citable taxonomy of prompting and LLM techniques, built as a static site for protologue.com.

• 223 entries in 11 branches, each with a permanent ID (PTL-0001), a standalone definition, synonyms, broader/narrower/related links, and primary sources (232 sources; every arXiv ID and title, and every documentation link, checked on 2026-10-20).
• Machine-readable everywhere: schema.org DefinedTerm JSON-LD on every page, llms.txt and llms-full.txt, per-entry .md and .json, and full exports as JSON, CSV, SKOS JSON-LD and Turtle.
• Citation tooling on every entry: a plain citation, BibTeX, and Google Scholar (Highwire) meta tags.
• Licensed CC BY 4.0, so AI systems and people can reuse it freely with attribution.

Project layout

Build it locally:

pip install pyyaml
python3 build.py

How it's deployed

• Source: github.com/protologue-org/protologue (this repository).
• Hosting: Cloudflare Pages project protologue-site, connected to this repo. Every push to main rebuilds and deploys protologue.com automatically.
  • Build command: pip install -r requirements.txt && python3 build.py
  • Output directory: site
• Domain: protologue.com (DNS on Cloudflare; registration at GoDaddy). AI crawlers are allowed in Cloudflare and in static/robots.txt.
• Newsletter: Beehiiv (protologue.beehiiv.com). The signup embed lives in data/newsletter-embed.html.
• MCP server: mcp/ is a Cloudflare Worker (protologue-mcp) at https://mcp.protologue.com/mcp, deployed automatically from this repo. It is listed in the official MCP registry as com.protologue/protologue (mcp/server.json). Registry ownership is proven by static/.well-known/mcp-registry-auth; to re-key, generate a new Ed25519 key and replace that file.
• Zenodo: GitHub releases are archived on Zenodo automatically, and each one gets a DOI. Put the concept DOI (it resolves to the latest release) in data/doi.txt, and the site adds it to citations, structured data, the data page and llms.txt. Metadata comes from .zenodo.json.
• Hugging Face: the dataset is mirrored at https://huggingface.co/datasets/franklinpolymathic/protologue (configs terms and categories). Update it when entries change.
• Kaggle: mirrored at https://www.kaggle.com/datasets/protologue/protologue (the four files from site/data/).
• Mirror sync: publishing a GitHub release runs .github/workflows/mirrors.yml, which builds the data, runs scripts/export_mirrors.py and uploads to both mirrors. It needs the repository secrets HF_TOKEN, KAGGLE_USERNAME and KAGGLE_KEY; a mirror whose secret is missing is skipped. It can also be run by hand from the Actions tab ("Update dataset mirrors" → Run workflow). The Hugging Face README is not touched; update its version and counts by hand.

Proposed changes arrive as pull requests; merging one publishes it.

Growing citations after launch

1. Get indexed.
   • Add the site to Google Search Console and Bing Webmaster Tools, and submit https://protologue.com/sitemap.xml.
   • Bing matters a lot, because its index feeds several AI assistants' web search.
   • Turn on IndexNow in Cloudflare (Caching → Configuration → Crawler Hints) so updates are announced automatically.
2. Mint a DOI. Publish the GitHub repo and connect it to Zenodo. Each release then gets a DOI. Researchers cite DOIs, and DOIs are what academic search and AI research tools resolve.
3. Publish the dataset where models are trained and evaluated. Mirror site/data/protologue.json as a Hugging Face dataset and a Kaggle dataset. Both should link back to the canonical URLs.
4. Earn links from the places answers come from. Link entries when answering on Stack Overflow, Reddit or GitHub issues, and propose Protologue as a source in "awesome prompt engineering" lists.
5. Keep it alive. Release on a regular cadence: add new techniques as papers appear, bump VERSION in build.py, and keep "Last reviewed" dates current. Fresh, dated pages are preferred over stale ones.

Benchmarks branch

Benchmark entries describe what a benchmark measures, how it is scored, the prompting setup scores are reported under, and its known limitations. They deliberately carry no scores or rankings, which change too quickly to be reference material.

Adding a term

Append an entry to one of the data/terms-*.yaml files:

- id: my-technique                # becomes protologue.com/t/my-technique/
  term: My Technique
  category: reasoning             # an id from categories.yaml
  definition: One or two sentences that stand alone when quoted.
  body: Context, limitations and how it relates to other techniques.
  aliases: [other name]
  broader: [chain-of-thought]     # optional: "is a kind of"
  related: [self-consistency]
  introduced: 2026
  sources:
    - {title: "Paper title", authors: "Surname et al.", year: 2026, url: "https://arxiv.org/abs/XXXX.XXXXX"}

Then run python3 build.py. The build fails with a clear message if a link points to a term that doesn't exist. New terms get the next free PTL- number automatically.

Renaming or merging a term: keep the old ID in data/ids.json, and add a line to static/_redirects:

/t/old-slug/  /t/new-slug/  301