Problem
Nine phases of runtime, each with its own plan, is nine chances to define a runtime rather than build one. Every subsystem after the first should plug into an existing runtime instead of inventing its own lifecycle, its own storage assumption, and its own idea of where state lives.
That is not hypothetical — DriftMeetingRepository, features/coins/, features/money/, and relational_store each did exactly that, because there was no runtime to plug into.
Expected outcome
The first executable milestone. One thing proven end to end:
Operation → Persist → Replay → Projection → UI
Scope — deliberately small
| In |
Out |
| Supervisor |
AI |
| Vault |
Sync |
| Operation Log |
Workflows |
| Runtime API (the six functions) |
Capabilities beyond a minimal Notes reference |
| Empty Projection Engine |
Everything else |
The projection engine is empty on purpose. It proves the seam exists and that a projection can be driven, rebuilt, and dropped. What it projects comes later.
Why this ordering
Every subsystem after the skeleton inherits a runtime that already has a lifecycle owner, a storage contract, and a public surface. Nothing after this has to decide where state lives, because that question is answered and enforced.
The Supervisor and the replay engine are the two architectural centres of Airo Mind (design spec §11c-3). A skeleton containing both, and nothing else, is the smallest thing that can prove them.
Definition of done
What this replaces
Not the phase plans — it sequences them. Phases 1–6 still deliver what they describe; the skeleton is the vertical slice through them that must work before any of them is finished horizontally.
Blocked by
Phase 1 (#1193) and the Supervisor (#1302).
Problem
Nine phases of runtime, each with its own plan, is nine chances to define a runtime rather than build one. Every subsystem after the first should plug into an existing runtime instead of inventing its own lifecycle, its own storage assumption, and its own idea of where state lives.
That is not hypothetical —
DriftMeetingRepository,features/coins/,features/money/, andrelational_storeeach did exactly that, because there was no runtime to plug into.Expected outcome
The first executable milestone. One thing proven end to end:
Scope — deliberately small
The projection engine is empty on purpose. It proves the seam exists and that a projection can be driven, rebuilt, and dropped. What it projects comes later.
Why this ordering
Every subsystem after the skeleton inherits a runtime that already has a lifecycle owner, a storage contract, and a public surface. Nothing after this has to decide where state lives, because that question is answered and enforced.
The Supervisor and the replay engine are the two architectural centres of Airo Mind (design spec §11c-3). A skeleton containing both, and nothing else, is the smallest thing that can prove them.
Definition of done
emit_operationand nothing elsereplay_passes == 1is asserted, not documented — C2What this replaces
Not the phase plans — it sequences them. Phases 1–6 still deliver what they describe; the skeleton is the vertical slice through them that must work before any of them is finished horizontally.
Blocked by
Phase 1 (#1193) and the Supervisor (#1302).