roadmap: add Phase 1.5 — undo & redo (snapshot-based)

- effort S (~0.5-1 day): rides on Phase 1 machinery (serde snapshots,
  replace_state, the store.track() autosave trigger) — no component
  changes needed, page-agnostic
- coalesced snapshots (one per ~1 s window), capped stacks, Ctrl+Z /
  Ctrl+Shift+Z with focus guard for text fields
- sequencing note: lands before the edit-heavy Phases 2-7 so their edits
  are undoable from day one; nothing depends on it
- Phase 8 e2e item: drop the persistence round trip (already covered by
  persistence.spec.ts), add undo shortcuts
- ARCHITECTURE gaps list: note the missing undo/redo

ARCHITECTURE.md

@@ -157,6 +157,8 @@ currently mounted on any page.
 ## Current gaps
 
 - Quests, notes, contacts: models and mock data only
+- No undo/redo (planned as Phase 1.5; snapshot-based, rides on the
+  persist machinery)
 - `Character` and `Wiki` pages render placeholders
 - Prepared spells list is empty; `Spell` is a name-only stub; attunements
   have no UI; `Settings` has only `version`

ROADMAP.md

@@ -98,6 +98,28 @@ export/import round trip). Notes:
 - Version check is the migration hook: `parse_dashboard` returns
   `ImportError::{Parse, NewerVersion, OldVersion}` (RU messages).
 
+## Phase 1.5 — Undo & redo (S, ~0.5-1 day)
+
+Snapshot-based undo rides on the Phase 1 machinery: serialize → push →
+`replace_state`. It is page-agnostic: the `store.track()` trigger already
+fires on any mutation, so no component changes are needed and every later
+phase's edits are undoable from day one.
+
+- [ ] `state::undo.rs` (hydrate-gated): undo/redo stacks of serialized
+  `Dashboard`, capped (e.g. 100); push coalesced on the store trigger
+  (one snapshot per ~1 s window → one undo step = one edit burst)
+- [ ] a new edit after undo clears the redo stack; Clear and import push a
+  snapshot first, so they are undoable too
+- [ ] `undo()` / `redo()` call `replace_state`; expose stack depths as
+  `RwSignal<usize>` for button disabled states
+- [ ] header buttons: ↩ Отменить / Вернуть (disabled when empty)
+- [ ] shortcuts: Ctrl+Z, Ctrl+Shift+Z (Ctrl+Y); ignore when focus is in an
+  input/textarea/select, so typing never triggers undo
+- [ ] unit tests for stack semantics; e2e: edit → undo → redo; reload keeps
+  the undone state (undo writes through autosave)
+- **Acceptance:** every edit since the last window is reversible with a
+  keystroke; undo survives a reload.
+
 ## Phase 2 — Character page (M, ~1-2 days)
 
 `/` currently renders the literal string "character".
@@ -167,8 +189,8 @@ New route `/contacts`. Model and mock data exist; zero UI.
 ## Phase 8 — Polish, tests, release (L, ~2-4 days)
 
 - [ ] responsive SCSS pass (sidebar and header on mobile widths)
-- [ ] Playwright e2e: persistence round-trip, add item from search, quest CRUD,
+- Playwright e2e: add item from search, quest CRUD, notes filter, undo
-  notes filter
+  shortcuts (persistence round trip is covered by `persistence.spec.ts`)
 - [ ] rewrite README: usage, dev workflow, deploy
 - [ ] release build via cargo-leptos; verify SSR + hydration; pick a deploy target
 - **Acceptance:** fresh clone → `cargo leptos build` → app runs; e2e green.
@@ -179,7 +201,9 @@ New route `/contacts`. Model and mock data exist; zero UI.
 
 - Phase 1 before Phases 2-7: every new page must persist. Building pages first
   against mock data and persisting later costs a data-migration pass instead.
-- Phases 2 and 3 can run in parallel after Phase 1 (different components).
+- Phases 2 and 3 can run in parallel after Phase 1.5 (different components).
+- Phase 1.5 (undo) is page-agnostic: it must land before the edit-heavy
+  phases so their edits are undoable; nothing in Phases 2-7 depends on it.
 - Phases 4-6 are independent of each other and share one CRUD-list pattern —
   build the pattern once (e.g. a generic list editor) and reuse it.
 - Phase 7 reuses the Phase 3 add-to-inventory path; keep that path shared.