Design Review — ADR

gov-af5bbdc0b162

Verdict: APPROVED (0 violations) — but downstream-advocate attached conditions. The verdict card below the conversation shows the full per-role breakdown.

Domain model

Board owns columns (ordered)
Column {id, title, tasks[]}
Task {id, title, priority: Priority}
Priority = "Low" | "Medium" | "High" (closed enum, default Medium — Define condition)

type Board = { columns: Column[] }
function move(b: Board, taskId: string, toColumnId: string): Board   // pure, returns a new Board

ADR-001 — storage

localStorage behind a versioned boundary (schemaVersion); seed-empty on corrupt / old-version / unavailable. Rationale: client-only MVP, no backend in scope; the boundary keeps a future server swap a one-file change.

ADR-002 — drag mechanism

Hand-rolled HTML5 DnD with the test-binding pinned to fireEvent (user-event v14 has no HTML5 DnD). Rejected: a DnD library — 30kB + opaque to the governed test.

ADR-003 — move() is a pure reducer

State transitions are a pure move(board, taskId, col) returning a new board — testable without the DOM, and the same function backs keyboard-move when a11y lands post-MVP.

The crux (counterfactual)

If move() mutated in place, the same-column no-op and the undo path both break — so the reducer purity is the load-bearing decision the cycles must prove, not the drag wiring.

downstream-advocate returned CONDITIONAL underneath the green headline: keyboard task-move deferred to post-MVP (acknowledged a11y tradeoff); degraded-storage seeds empty. Both carried into the build contract.