Runtime Atlas / Executable evidence
Don’t take the architecture on faith. Run it.
Operate six cases through the real public runtime, then inspect who accepted the intent, where the transaction settled, and what verifiable state changed.
This is a resettable browser/Core composition, not a Headless Core or server-runtime claim. Those boundaries remain on the Roadmap.
Intent → one history entry
One continuous gesture. One Undo unit.
Watch three pointer updates travel through one Feature session and settle as one reversible action.
Expected The value reaches 5, Undo returns it to 0, Redo restores 5, and history contains one entry.
Status: starting · 0/8
- 01Intent
Person, automation, or AI asks
- 02Feature
App policy accepts and routes
- 03Transaction
Factory groups the action
- 04Canonical owner
One owner changes verifiable state
- 05Projection
App surfaces render returned evidence
Worker evidence ledger
What actually ran
The worker is ready. Step once or run the remaining actions.
App-owned projection
Returned runtime result
Run or step the case. This surface only renders evidence returned by the isolated runtime.
The rule behind every case
Intent may come from anywhere. Canonical writes do not.
People, automation, collaboration, and AI all enter through bounded app-owned policy. Framework owners validate and settle state; canvas, hierarchy, properties, serialization, search, and presence remain projections or integrations - not competing sources of truth.