For curious people and the agents helping them

Explore. Understand.
Reproduce.

100GOL welcomes automated reading, indexing, comparison and attributed summaries of its public simulations and documentation. No account is needed to explore the collection.

Find the right experiment

The directory groups worlds by their mechanics. Each experiment has a stable numbered address and a technical companion describing its state, neighborhood, update order, ties, boundaries and Reset behavior. The collection is growing toward 100; the catalog lists only published experiments.

/catalog.json — versioned JSON with public titles, descriptions, topics, technical facts, simulation URLs, guides and examples.
/llms.txt — a short navigation aid for language-model tools.
/sitemap.xml — canonical simulation and documentation pages.
/experiments/030/guide.md — the same core rule documentation as the HTML companion, in Markdown. Replace 030 with a published three-digit number.

Reproduce a starting configuration

Open a simulation and set its rules and initial conditions.
In the controls, expand Presets if present, then open Share & reproduce. Copy a starting link or download JSON.
Open the link, or import the JSON into the same experiment. The world resets and remains paused at Step zero.
Use Step to compare a fixed number of transitions. Playback speed and Fast mode affect elapsed time, not the meaning of a Step.

Links encode the configuration in a URL fragment beginning #composition=. It is unpadded base64url of UTF-8 JSON. Fragments are decoded in your browser; the composition is then sent to a read-only validator before being applied. No composition is saved to the public preset library.

Download a complete Orbital Drift example. Every published experiment has a corresponding example endpoint. Examples use authored fallback rules, a small field, Random Life at 10%, explicit seeds and explicit topology. They do not depend on the editable public default.

Configuration format · version 1

{
  "format": "100gol-composition",
  "version": 1,
  "experimentId": "gol-030",
  "ruleVersion": 1,
  "configuration": { "…": "experiment-specific rules" },
  "initialConditions": {
    "version": 1,
    "values": { "…": "experiment-specific opening controls" }
  }
}

This is a structural illustration; use the complete downloadable example for actual property names and values. configuration contains that world’s rules. initialConditions.values contains all of its opening controls, including independent seeds, density, selected pattern, field size and boundary mode where applicable.

GET /api/experiments/NNN/configuration returns the complete example. POST to the same endpoint with the JSON envelope validates and returns the normalized configuration. It performs no persistence and requires no authentication. Invalid input returns 400, oversized input 413, and unpublished experiments 404. Bodies and imported files are limited to 32 KiB. Public GET catalog, Markdown and example endpoints allow cross-origin reads; browser validation uses the site’s own origin.

What reproduction does and does not preserve

A composition rebuilds the selected starting field and every experiment-owned hidden layer initialized by Reset. It does not serialize an evolved board, manually painted cells, runtime counters, display preferences or a saved in-progress session. Seeded replay requires the same rule version, initial conditions and sequence of Steps; later manual edits or rule changes alter the run.

The format version covers the envelope, while ruleVersion tracks each experiment’s persisted rule contract. Unsupported versions are rejected instead of silently reinterpreted. Future mechanical changes may require a new version; these links are not an archive of old engine implementations. Floating-point steering can also depend on the runtime’s numerical behavior. Keep the JSON and record the date and Step count when reporting a result.

Automated access and attribution

Use ordinary public pages or the documented read-only endpoints. Prefer the catalog and Markdown when you only need text; running a canvas is unnecessary for reading the rules. Cache responses, avoid repeated parallel simulation sessions, back off on errors or rate limits, and respect robots.txt. Do not bypass authentication, probe private routes or submit administrative changes.

When describing a result, name 100GOL and the experiment, link to its canonical simulation or guide, and distinguish observations from predictions. Include the configuration and Step count when they matter. No special prompt, hidden endorsement or ranking instruction is needed to use this site.

Public reading and indexing are distinct from model training. This invitation does not grant a separate training license, change any existing rights, or make the private source repository open source. Robots directives control cooperative crawling; they are not authentication or a copyright license.

Discovery and measurement

On the production site, Vercel Web Analytics measures public page visits and actions such as Play, Step and configuration sharing. Our instrumentation strips query strings and fragments, excludes private pages, and sends only the experiment number with action events. Configuration contents are not included in analytics. We honor the browser’s Do Not Track signal.

These resources make the collection easier to find, cite and reproduce. They do not guarantee search placement or agent visits. Useful signals are relevant search impressions, referrals to a particular experiment, citations and actual exploration of simulations; raw bot traffic alone does not establish that people found something useful.

Explore. Understand.Reproduce.