-
Notifications
You must be signed in to change notification settings - Fork 0
Home
The repository is authoritative. Start with the canonical README; hosted wiki pages mirror the files under
docs/wiki.
Darwin is a controlled software evolution engine. It observes semantic product telemetry, creates deterministic evidence, asks GPT-5.6 for a structured mutation portfolio, and hands a human-selected manifest to a bounded Codex workflow in the target repository.
The current proof target is ProjectFlow, a separately deployed task-management application.
| Component | Location |
|---|---|
| Darwin control room | https://darwin-control-room.pages.dev |
| Darwin API | https://darwin-api.stevie-johnston.workers.dev |
| ProjectFlow study | https://darwin-projectflow.pages.dev/?study=true |
| Darwin repository | https://github.com/sjohnston1972/darwin |
| ProjectFlow repository | https://github.com/sjohnston1972/projectflow |
measured interaction
-> schema-valid semantic event
-> ordered D1 record
-> deterministic journey and evidence pack
-> evidence-citing GPT mutation portfolio
-> human-selected immutable manifest
-> bounded Codex repository workflow
-> real patch, checks, pull request, and preview
-> explicit release or rejection
-> retained Genome record and reviewable rollback
- Getting Started: local installation and first run.
- Architecture: components, data flow, persistence, and trust boundaries.
- Telemetry and Evidence: captured signals, privacy boundary, parsing, and evidence quality.
- Darwin Lab: bounded real-target automation, provenance, evidence, and safety boundaries.
- AI and Mutation Workflow: GPT context, structured validation, Codex manifest, release, and rollback.
- Demo Runbook: reliable three-minute presentation flow.
- Operations and Deployment: Cloudflare, D1, secrets, deployment, and smoke tests.
- Security and Privacy: current controls, limitations, and hardening backlog.
- API Reference: current Worker routes and contracts.
- Development and Testing: workspace commands, tests, and contribution workflow.
- Troubleshooting: common failures and recovery steps.
- Current product specification
- Technical architecture
- Real telemetry and evidence plan
- Three-minute demo source
- Generated Worker route contract
- Documentation ownership and freshness
Darwin uses these terms consistently:
| Product concept | Darwin term |
|---|---|
| target source and configuration | genome |
| measured friction | selection pressure |
| bounded code change | mutation |
| measured product outcome | fitness |
| accepted change | survived selection |
| rejected change | failed selection |
| retained version and evidence history | Genome |
Predicted impact is always a hypothesis. A passing repository build proves implementation validity, not user fitness. User fitness requires a compatible post-release evidence cohort.
Darwin intentionally supports one controlled target rather than arbitrary repository ingestion. ProjectFlow defines the mutable paths, protected paths, source context, validation commands, and change budgets that a Codex workflow must honor.
The 10,000-event simulator demonstrates deterministic scale. It is synthetic and is not mixed with measured ProjectFlow evidence.