Code Rooms
# Proof-Grown Systems
Proof-Grown Systems do not scale by narrative. They scale by accumulated
evidence.
## Core Rule
```text
No structure without contract.
No success without proof.
No checkpoint before artifacts.
```
## Required Artifact Types
- Intent blueprint: what the human is actually asking for.
- Capability gap: what is missing and who owns it.
- Proof contract: what must be demonstrated.
- Handoff envelope: what the next agent must reload.
- Non-claims: what must not be said yet.
## Proof Quality
A proof must be causal. It should show that a specific command, runtime action,
browser interaction, local provider fixture, or validator produced an inspectable
artifact.
Good proof includes:
- exact command or operation identity;
- input fixture;
- output artifact;
- positive assertion;
- negative assertion when safety matters;
- non-claims.
Bad proof:
- "it should work";
- "tests probably cover it";
- generated UI without runtime capability;
- screenshots that do not bind to a contract;
- provider calls without dry-run boundary.
## Capability Maturity
Use this language for agents:
- `ghost`: visual/intended, not implemented.
- `gap`: missing capability with owner and contract path.
- `candidate`: local proposal with a proof plan.
- `local`: works locally, not shared truth.
- `signal`: repeated evidence, still bounded.
- `pattern`: stable shape across more than one context.
- `proven`: named class with documented proof envelope and non-claims.
