Non-goals (V1)
Retornatus V1 is intentionally bounded. These are not bugs — they are product choices.
Out of scope
| Non-goal | Rationale |
|---|---|
| Multi-provider live LLM integration tests as default CI | Environment owns execution; harness tests structure and gates |
| Enforced OS sandbox / worktree orchestration | Prefer native host isolation; Boundaries stay advisory |
| Drop-in 1:1 Spec Guardrails command replacement | Behavioral parity ≠ clone; avoids coupling to another product’s surface |
| Remote multi-tenant control plane | Local repository-native harness |
| Permanent agent personas / identity model | Specialization via Assignment + Context + Skills |
| Automatic authoritative Rules from recurrence | Human Decision required |
PRD §65 lists explicit V1 non-goals in full.
What “done” means for V1
A developer can wake a repo, open a Change, activate a Contract, execute under the host, record Claim-bound Evidence, reach Assurance SATISFIED, preserve Learning, and recover via wake after interruption — without chat as canonical state.
See PRD §§69–70.
Reopening a non-goal
Only when a concrete failure mode appears that existing mechanisms and the Environment cannot solve (PRD §71). Prefer gates that prove host behavior over rebuilding host capabilities inside Retornatus.