worldseq documentation
Install, use, author, and operate the SaaS business simulation environment.
worldseq lets humans or agents run fictional SaaS companies. Choose what to fund and when to act, then inspect the work, customer commitments, cash, and reports that follow. Start with a local installation or browse the environments to see the kinds of decisions they contain.
Guides
- Getting started: install the standalone package and start the service.
- Play as a human: choose a challenge, make a plan, and inspect the result.
- SDKs: use the TypeScript and Python player clients.
- Evaluations and submissions: run, retain, verify and exchange results.
- Company authoring: prepare and check a declarative world package.
- Environments: browse companies and their task contracts.
Examples
- Harbor SDK agents: install and run the TypeScript or Python example in an isolated player container.
Verification
- How results are checked: distinguish feasibility, information fairness, and outcome verification.
- Protecting evaluations: understand player access, protected instances, and the trusted boundary.
- A few useful invariants: inspect the arithmetic and bounded arguments behind the checks.
- Release 0.1.0: read the scope, results, and limits of the verified release.
Reference
- Scoring and interpreting results
- API and credential boundaries
- Business mechanics
- Enterprise and SLA agreements
- Finance and workforce
- Growth experiments
- Usage billing
- Calendar and storage limits
- Assessment requirements
- Counterfactual assessments
- Evidence and licensing
- CEO decision coverage and limitations
Operators
- Harbor integration
- Private instances
- Private suites
- Journal reuse
- Remote execution
- Release qualification
Development
- Architecture
- Contributing
- Quality controls
- Controlled mechanism cases
- Mechanism authoring
- Validation campaigns
- Capability studies
Release receipts belong to the release they identify. Documentation describes supported contracts; it does not award certification to new instances or convert development demonstrations into protected evaluation evidence.
Agent experiments
Run and publish agent experiments with Codex, OpenCode, Claude Code, or Pi. Configure paired cases, retain trajectories, and inspect verified company outcomes.
Challenge tag definitions explain the scope of the company labels.