ADR-0014: Delivery scope and sequencing

Status

Accepted

Date

2026-06-24

Context

The execution plan presents a 90-day schedule with parallel team pods, an Architecture Review Board, and an extension certification pipeline. It never states a team size. The actual team is 2–3 engineers. A schedule that assumes parallel pods but is executed by 2–3 people serially is fiction, and every “how much do we build” decision (ADR-0005, 0007, 0008, 0010, 0013) only resolves correctly once the headcount is honest.

The plan also front-loads roughly 18+ weeks of serial work (kernel skeleton + operational-loop POC + slices of offline/writeback/agent) into “90 days.”

Decision

Read the 90-day plan as a forcing-function RFC, not a staffed commitment. For a 2–3 person reality:

Sequence strictly serially. The plan’s parallel pods collapse into one ordered thread; realistic horizon is ~6 months, not 90 days.
Prove ONE vertical slice end-to-end before any breadth: a single Action (projects.CreateWorkPackage) → Event → synchronous projection → read-your-write → policy check → generated-SDK call. This is the walking skeleton; it exercises the spine (ADR-0003) on one thread before widening.
Apply “shape for the vision, size the runtime for the proof” everywhere — the discipline that produced ADR-0005, 0007, 0008, 0010, 0013.
Defer the org machinery (ARB, certification pipeline, parallel pods) until headcount justifies it — consistent with deferring multi-author realm governance (ADR-0008).

Alternatives Considered

Treat 90 days as a real target with parallel pods

Rejected: requires a funded team that does not exist; would force shallow, broad work that fails the spine.

Solo/exploratory framing (design artifact only)

Rejected: this is a real build by 2–3 people, not a pure pitch artifact — but the deliverables (RFCs, ADRs, a walking skeleton) resemble that framing early on.

Breadth-first (stand up many realms/modules in parallel)

Rejected: violates “prove the spine on one thread first”; breadth before the truth model (ADR-0003) is validated multiplies rework.

Consequences

The first concrete deliverable is the walking-skeleton slice, plus the two freeze-gating spikes (ADR-0003 truth model, ADR-0006 language build-off).
Milestones are sequenced, not parallel; estimates assume 2–3 people.
Org/process scale-ups (ARB, certification) are explicitly out of scope until the team grows — their absence is a deliberate decision, not an oversight.
Sensitive to: a funding/headcount change would supersede the sequencing (parallelism becomes available) but not the “prove the spine first” principle.