LLM is fuel, not the brain
The LLM parses, proposes, and generates candidates; decisions, guard, verification, and adoption are carried out by GHW's existence degree, rules, and feedback loops.
GHW
Goal-Harness-World
A digital life framework that closes the loop across math, rules, memory, feedback, and product KPIs
Auditable self-evolution
The L1-L5 self-evolution chain passes SelfEvolutionGuard, CompileVerifier, SelfEncoder, and delta-E feedback, so proposals never bypass the rules.
Product digital life runtime
ProductIntent, KPI Bridge, ChangeProposal, Consent, Audit, and the SDK/API let GHW embed into real products.
Human symbiosis loop
TaskQueue, HumanSignal, HumanGSP, ValueTrendTracker, and chat memory bring human tasks and feedback into the long-term trajectory of E.
What is GHW
GHW is a digital life runtime. It is neither an agent that hands every judgment to an LLM, nor a plain task orchestrator. It splits the system into four layers:
- World: observes the world and maintains the M/R/N three-source state and the existence degree E.
- Goal: selects the next goal based on discrepancy and attention.
- Harness: executes actions, calls the LLM, runs self-evolution, and guards boundaries.
- Agent: wires perception, action, reflection, self-evolution, migration, and persistence into a loop.
The core principle is: the LLM may propose, but cannot decide; the digital life itself decides whether to adopt, via E, Guard, Verifier, Consent, and ΔE feedback.
Current Completeness
The project already has the core loops of a research framework and a product runtime:
- L1-L5 self-evolution, rule proposals, code verification, self-migration, and state persistence are working.
- Memory, reflection, task queue, human feedback, value trends, and UI provenance are closed-loop.
- ProductRuntime supports ProductIntent → KPI → Proposal → Consent → Apply/Rollback → Audit.
- The C SDK and the
ui_serverHTTP API expose the product runtime capabilities.
Parts still being engineered include real product Adapters, signed audits, role permissions, more deployment templates, and a long-term stress-test matrix. See the Roadmap.
Who This Is For
- To understand the GHW architecture: start with Core Concepts and Architecture Overview.
- To embed GHW into a product: start with Product Digital Life Runtime.
- To call the APIs: see HTTP API and C SDK.
- To deploy and operate: see Deployment & Operations.
Build the Docs Site
The docs are plain Markdown built with zdoc — no config file, no Node.js:
zdoc docs site/docs --output site/public --title "GHW"
The output in site/public/ is a static site ready for any web server or GitHub Pages.