Trust model
Security claims stop where evidence stops.
Enforced by SPACLY
- Candidate release ref is read from a verified same-origin manifest.
- Every candidate route response body is SHA-256 matched to that manifest.
- Candidate paths are same-deployment relative paths.
- Ordinary assessments are immutable; decisive READY/BLOCKED outcomes cannot be resampled.
- Challenge evidence is independently fetched before consuming a bounded route challenge slot.
- Owners cannot challenge their own READY candidate.
- Authorization binds a deterministic evidence root.
Not claimed
- A release ref is not proven to be a Git commit unless the deployment pipeline itself publishes that relationship in the manifest.
- Semantic model explanations are not necessarily identical across validators; only consequential status/provenance fields are consensus-bound.
- Personalized/A-B content, provider correlation and transient web behavior remain residual risks.
- Live Studionet performance/finality/fees are not claimed before funded-wallet proof.
Accepted is not finalized. Finalized is not execution success until the receipt says so and finalized contract state is re-read.