lanework/Kanban/LiveStore/BoardStore.swift

import Foundation
import Observation
import os
// The one UI import in the store layer, and it earns its place: 03-board-ui.md § Motion puts the
// animate-or-snap split on the *reload*, and a reload lands here. `Motion` owns the decision and
// every curve; this file owns nothing but the `withAnimation` around the assignment (see `land`).
import SwiftUI

// MARK: - Vocabulary

/// Why a board is refusing writes — the read-only lock's cause, and now the whole of the
/// vocabulary 02-architecture.md names.
///
/// The three cases share one *scope* (§ "The lock's scope") — every mutating command disabled
/// across every window sharing the store, drops refused, ⌘-drag moves degraded to copies, editor
/// buffers kept but their debounced saves suspended — and differ only in cause and in **what
/// clears them**, which is the one thing this enum's cases are actually asked about (see
/// `BoardStore.land(_:generation:origin:)`). The banner turns a case of this into user-facing
/// phrasing (§ The banner surface); the store only owns the truth of it.
public enum ReadOnlyLockReason: Sendable, Equatable {
    /// A reload that followed a bracketed wholesale operation failed, so the last-good snapshot on
    /// screen may describe a tree that no longer exists — after a branch switch, a different branch
    /// entirely. Writes derived from it would land nonsense, so every write is refused until a
    /// reload succeeds (02-architecture.md § Live-reload resilience, "A failed reload after a
    /// bracketed operation locks the board read-only").
    ///
    /// **Clears on the next successful reload, whatever its origin** — typically once the offending
    /// file is fixed.
    case bracketedReloadFailed

    /// The board's root is gone and its bookmark re-resolution found nothing: the volume unmounted,
    /// or the folder was deleted in Finder while the board was open (02-architecture.md §
    /// Write-failure surfacing, "A vanished board root locks the board read-only"). Every write
    /// would land nowhere, so the last-good snapshot stays on screen, read-only.
    ///
    /// **Clears on the next successful reload, whatever its origin**: a reload can only succeed if
    /// the root is back, so success *is* the return signal. Pending dirty buffers then save
    /// normally.
    case vanishedRoot

    /// The board opened somewhere it cannot be written: a read-only volume (a DMG, a snapshot, a
    /// read-only share) or a permission-denied folder (02-architecture.md § Write-failure
    /// surfacing, "An unwritable board location enters the read-only lock at open"). Failing
    /// loudly, specifically, *once* beats letting every gesture fail one at a time.
    ///
    /// **Clears only on a successful *reconciling* reload whose writability re-probe passes** —
    /// unlike its two siblings, whose cause a successful reload disproves by itself. A board on a
    /// read-only DMG reloads perfectly all day long; only the probe (§ "Writability re-probes on
    /// every reconciling reload" — wake, activation) can tell that the permission or the mount
    /// actually changed.
    case unwritableLocation
}

/// The refusal `BoardStore.performWrite` throws when the board is locked read-only.
///
/// **Deliberately temporary, and deliberately not a `BoardWriteError`.** The lock is a *store*
/// condition, not a filesystem outcome: nothing was attempted, no path failed, and folding it into
/// `BoardWriteError.io(message:)` would misrepresent a policy refusal as an I/O error in the one
/// place — the banner — where the distinction is the whole point. Widening `BoardWriteError` to
/// carry a refusal case is the banner card's job (02-architecture.md § Write-failure surfacing:
/// "The operation is a closed enum, not a string"), and it will unify this vocabulary with the
/// Writer's. Until then this thin error keeps `performWrite`'s honesty at the cost of an untyped
/// `throws` on its signature.
public enum BoardStoreWriteRefusal: Error, Sendable, Equatable, CustomStringConvertible {
    case readOnlyLocked(ReadOnlyLockReason)

    public var description: String {
        switch self {
        case let .readOnlyLocked(reason):
            "the board is read-only (\(reason))"
        }
    }
}

// MARK: - BoardStore

/// The per-board hub: one live snapshot, one reload pipeline, and the read-side conditions the
/// board window renders (02-architecture.md § Layering ▸ Components).
///
/// **It enforces the one-way flow — files → watcher → loader → store → views — by never mutating
/// its snapshot from the write path.** A user action runs the Writer, the Writer touches disk, the
/// watcher notices, and the change arrives here as a reload like any external edit. The app trusts
/// its own writes no more than anyone else's; that is what makes external editors and agents
/// first-class, and it is why there is no `snapshot` setter anywhere below `apply(_:generation:)`.
///
/// ### What this type actually owns
///
/// 1. **The reload pipeline.** At most one tree walk in flight, off the main actor; signals arriving
///    during one coalesce into a single follow-up; only the newest result applies.
/// 2. **The failure rules.** A failed reload never replaces a good snapshot; an ordinary failure
///    raises the banner condition and leaves editing alone; a failure after a bracketed wholesale
///    operation locks the board read-only; the next success clears both.
/// 3. **Transient state across reloads.** `transient.resolve(against:)` runs on every applied
///    snapshot, re-grounding the selection, the drag, the pending cut, and the new-card placeholder.
///
/// ### What it deliberately does not own
///
/// The `FolderWatcher` itself — the registry owns one watcher and one store per board and wires
/// them together (`watcherBrackets`, `handleWatcherEvent(_:)`), so this type can be built and tested
/// without a filesystem stream. And the transient state itself, which lives in its own container
/// (`TransientBoardState`) rather than accreting here as fields: this type knows only *when* to
/// re-resolve it, never what the rules are. That includes the **new-card placeholder** — a
/// pseudo-card with no disk presence and no UUID, overlaid on the snapshot rather than merged into
/// it (02-architecture.md § Layering, the one named exception to the one-way flow). Nothing here
/// makes that awkward: `snapshot` is a pure value swap with no identity assumptions, so an overlay
/// is simply rendered on top of whatever the latest reload produced.
@MainActor
@Observable
public final class BoardStore {

    // MARK: Read-side state

    /// The last good tree walk. Replaced wholesale by a successful reload and **never** by the write
    /// path — see the type's doc comment for why. A failed reload leaves it exactly as it was.
    public private(set) var snapshot: BoardModel

    /// Tolerated anomalies from the load that produced `snapshot` (stray folders, an indexless
    /// UUID-shaped folder, a board-level `deleted:`). Replaced with the snapshot, so they always
    /// describe the tree currently on screen.
    public private(set) var loadWarnings: [LoadWarning]

    /// The standing read-side condition: the error from the last reload that failed, `nil` when the
    /// board is healthy. `BoardLoadError` already carries fail-fast's specifics — the offending path
    /// and what is wrong with it — which is the whole of what the banner needs to render
    /// (02-architecture.md § Live-reload resilience). This is a *condition*, not a one-shot: it
    /// stands until a reload succeeds, and it heals without ceremony when one does.
    public private(set) var reloadFailure: BoardLoadError?

    /// Non-`nil` while the board refuses writes. Cleared by the next successful reload, per "the
    /// next successful reload clears both the banner and the lock".
    public private(set) var readOnlyLock: ReadOnlyLockReason?

    public var isReadOnly: Bool { readOnlyLock != nil }

    /// Everything shared across this board's windows that is **not on disk** — selection, drag
    /// membership, the pending cut, the search query, the new-card placeholder, trash visibility
    /// (02-architecture.md § Changes from Kanban).
    ///
    /// **Created with the store and dying with it**, which is what makes its per-open values per-open
    /// without any reset logic: closing the board is the reset. `let`, because it is one container
    /// for the store's whole life — the windows observe *it*, not a slot on this class.
    ///
    /// The store's only involvement is `resolve(against:)` on every successful reload; the rules that
    /// call answers live over there.
    public let transient: TransientBoardState

    /// Where the board is **now**. Follows the folder: a rename or a move absorbed through
    /// `relocate(to:)` updates it, so every URL derived from it — the Writer's paths, card-window
    /// keys, Reveal in Finder — re-derives at the new location (02-architecture.md §
    /// Write-failure surfacing, "A renamed or moved board root follows its file identity").
    ///
    /// Observed, deliberately: the window title's folder-name fallback reads this, and a rename in
    /// Finder should be visible in the title bar without anything else being told.
    ///
    /// `snapshot.rootURL` is the root the *last successful reload* walked, and therefore lags this
    /// by exactly one reload during an absorption. That is not a second source of truth: the
    /// relocation is always followed by the watcher reattach whose reconciling reload rebuilds the
    /// snapshot at the new root, after which the two agree again.
    public private(set) var rootURL: URL

    // MARK: Banners

    /// The board window's banner strip, as a model (02-architecture.md § The banner surface).
    ///
    /// **Owned, not injected**, and the reason is the hosting rule: the strip is "hosted by the
    /// window of origin", and a board window's strip has exactly one lifetime — this store's. A
    /// card window (m6) gets its *own* center for its own save, attachment, and raw-source Apply
    /// failures, and re-homes those rows here when it closes; injecting a shared center would
    /// erase precisely that distinction.
    ///
    /// It holds only what nothing else does — one-shot write failures, the history suspension,
    /// in-progress operations, passive signposts. The lock and the reload breakage stay this
    /// store's own state and are composed in at render time by `bannerRows`.
    public let banners = BannerCenter()

    /// The rows the board window's strip renders, in precedence order.
    ///
    /// Composed rather than stored: `readOnlyLock` and `reloadFailure` are the store's truths and
    /// `banners` holds the rest, so a stored array would be a third copy waiting to go stale. The
    /// ordering rule itself lives in `BannerCenter.rows(...)`, which is pure and tested on its own.
    public var bannerRows: [BannerRow] {
        BannerCenter.rows(
            lock: readOnlyLock,
            breakage: reloadFailure,
            oneShots: banners.oneShots,
            suspension: banners.historySuspension,
            operations: banners.operations,
            signposts: banners.signposts
        )
    }

    // MARK: Wiring

    /// The watcher's bracket calls, injected rather than owned: the registry holds the watcher and
    /// the store together, and a store that reached into a watcher it did not own could not be
    /// tested without one. `nil` means "no watcher attached" — every operation below still behaves,
    /// it simply has nothing to suspend, which is exactly the shape unit tests want.
    @ObservationIgnored
    public var watcherBrackets: (begin: @MainActor () -> Void, end: @MainActor () -> Void)?

    /// What to do when the watched root changes identity — injected for the same reason the
    /// brackets are: the response needs the board's **security-scoped bookmark**, and this store
    /// does not own one (the registry does, along with the watcher that must be re-attached and the
    /// last-known path that arms the return detection). A store that reached for a bookmark it did
    /// not hold could not be built or tested without a registry.
    ///
    /// `nil` keeps the documented no-op: the last-good snapshot stays on screen, which is what
    /// every other failure path here does, and which is exactly the shape unit tests and any
    /// storeless use want. `BoardStoreRegistry` wires it to its own recovery loop.
    @ObservationIgnored
    public var rootChangeDelegate: (@MainActor () -> Void)?

    // MARK: Reload machinery

    /// Monotonic id of the most recently *started* reload — and therefore also the number of tree
    /// walks this store has run since it opened, which is what makes the coalescing rule assertable
    /// from a test rather than merely plausible.
    ///
    /// Two rules are expressed as comparisons against it: **only the newest result applies** (a
    /// result whose generation is no longer the current one is dropped), and **the wholesale
    /// expectation binds to a load started after it was armed** (`wholesaleReloadFloor`).
    @ObservationIgnored
    private(set) var reloadGeneration = 0

    /// Whether a tree walk is running. At most one ever is: a second concurrent walk would buy
    /// nothing (both would produce the same snapshot) and would make "the newest result wins" a race
    /// rather than a rule.
    @ObservationIgnored
    private var reloadInFlight = false

    /// A reload owed but not started, because one was already running when the signal arrived — a
    /// **flag, not a queue**: any number of signals during one walk coalesce into exactly one
    /// follow-up, because the follow-up is a full tree walk that covers all of them. The origin is
    /// merged by the watcher's own precedence rule (`reconciling > appMediated > foreign`) so a
    /// foreign event folding into a pending reconciling one never downgrades it.
    @ObservationIgnored
    private var pendingReload: WatchOrigin?

    /// The generation from which a bracketed wholesale operation's expectation applies, or `nil`
    /// when no operation is outstanding.
    ///
    /// **Why a floor and not a boolean.** The rule is "the reload that ends this wholesale operation
    /// must succeed, or the board locks" — and a boolean cannot tell that reload apart from one that
    /// was *already in flight* when the operation ended, which observed a tree from before the
    /// operation touched it and therefore proves nothing about the result. Arming stores
    /// `reloadGeneration + 1`: the next walk to be *started*. An in-flight walk's generation is
    /// below the floor and passes through without consuming it; the first walk started at or after
    /// it consumes it — succeeding clears everything as usual, failing engages the lock.
    ///
    /// Ordinary (non-wholesale) reload failures never set the lock, because ordinary breakage is
    /// per-file: the snapshot still describes the tree and editing around the broken file is safe.
    @ObservationIgnored
    private var wholesaleReloadFloor: Int?

    /// Consumers suspended in `awaitQuiescence()`, resumed together the moment nothing is running
    /// and nothing is owed.
    @ObservationIgnored
    private var quiescenceWaiters: [CheckedContinuation<Void, Never>] = []

    /// Awaited off the main actor **after** a tree walk finishes and **before** its result is
    /// applied — the one seam this type keeps, `nil` in production.
    ///
    /// It exists because two of the contracts above are ordering claims about work that runs
    /// concurrently with the main actor ("only the newest result applies"; "the wholesale
    /// expectation never binds to a walk already in flight"), and a test that cannot pin a finished
    /// walk open can only approximate them with sleeps — which would make the suite slow, flaky, and
    /// silent about the very race it exists to rule out.
    @ObservationIgnored
    var loadBarrier: (@Sendable () async -> Void)?

    private static let logger = Logger(subsystem: "dev.rzen.indie.Kanban", category: "store")

    // MARK: - Init

    /// Opens a board: one synchronous tree walk, and **no fallback if it fails**.
    ///
    /// Fail-fast is the *initial-load* contract (01-storage-format.md § Malformed input): there is
    /// no last-good snapshot to keep on screen yet, so a broken board throws its `BoardLoadError`
    /// instead of constructing a store that would have nothing to show. Every rule below — the
    /// banner, the lock, "a failed reload never replaces a good snapshot" — exists only *because*
    /// this one succeeded.
    ///
    /// The walk is synchronous because the caller has nothing to render until it lands; the
    /// asynchronous, off-main pipeline starts with the first reload.
    public init(rootURL: URL) throws(BoardLoadError) {
        let result = try BoardLoader.load(boardRoot: rootURL)
        self.rootURL = rootURL
        self.snapshot = result.model
        self.loadWarnings = result.warnings
        self.reloadFailure = nil
        self.readOnlyLock = nil
        self.transient = TransientBoardState()
    }

    // MARK: - Inbound signals

    /// The single inbound signal — everything the outside world tells this store arrives here.
    ///
    /// Deliberately one door: the watcher, a test, and (later) the registry's wake/activation
    /// reconciliation all speak the same two-case vocabulary, so there is exactly one place where a
    /// filesystem change becomes a reload.
    public func handleWatcherEvent(_ event: WatcherEvent) {
        switch event {
        case let .treeChanged(origin):
            requestReload(origin)

        case .rootChanged:
            // Delegated, never guessed at. The settled response is to re-resolve the board's
            // security-scoped bookmark and either `reattach(to:)` the watcher at the new location —
            // a rename absorbed with no banner and no lock — or enter the vanished-root read-only
            // lock (02-architecture.md § Write-failure surfacing). Both halves need a bookmark this
            // store does not own, and a guess here would be a *wrong* guess: treating a rename as a
            // vanish would lock a board that is merely somewhere else.
            //
            // No reload is started either way. The delegate's two outcomes both end in one —
            // `reattach(to:)`'s reconciling reload at the re-resolved root, or the vanished-root
            // lock's eventual clearance when the root returns — and a reload fired from here would
            // walk a path that just stopped being the board.
            guard let rootChangeDelegate else {
                Self.logger.debug("rootChanged ignored — no delegate is wired (storeless use)")
                return
            }
            rootChangeDelegate()
        }
    }

    /// Starts a reload, or banks one if a walk is already running.
    private func requestReload(_ origin: WatchOrigin) {
        guard !reloadInFlight else {
            pendingReload = WatchOrigin.merged(pendingReload, origin)
            return
        }
        startReload(origin)
    }

    // MARK: - The reload pipeline

    /// Runs one tree walk **off the main actor** and applies its result on it.
    ///
    /// Off-main because a board of any size is a directory walk plus a YAML parse per item, and the
    /// whole point of the value-type snapshot is that this work can happen anywhere: `BoardLoader`
    /// is stateless statics and `LoadResult` is `Sendable`, so the only thing that has to be on the
    /// main actor is the assignment at the end. `Task.detached` rather than `Task { }`: a task
    /// created inside a `@MainActor` method inherits that isolation and would run the walk on the
    /// main actor — the exact thing this is avoiding.
    ///
    /// `origin` is not branched on: every reload is a full tree walk, so no origin is less safe than
    /// another. It is carried because the *policy* around a reload differs later — a `.reconciling`
    /// sweep re-probes writability (02-architecture.md § Write-failure surfacing) — and because it is
    /// the one thing that makes a reload's provenance legible in the log.
    private func startReload(_ origin: WatchOrigin) {
        reloadGeneration += 1
        let generation = reloadGeneration
        let root = rootURL
        let barrier = loadBarrier
        reloadInFlight = true
        Self.logger.debug("reload \(generation, privacy: .public) started (\(origin.rawValue, privacy: .public))")

        Task.detached(priority: .userInitiated) { [weak self] in
            // `do throws(BoardLoadError)`: without the annotation the `catch` widens to `any Error`
            // and the loader's typed error is lost on the way into `Result`.
            let outcome: Result<LoadResult, BoardLoadError>
            do throws(BoardLoadError) {
                outcome = .success(try BoardLoader.load(boardRoot: root))
            } catch {
                outcome = .failure(error)
            }
            await barrier?()
            await self?.apply(outcome, generation: generation, origin: origin)
        }
    }

    /// Lands one walk's result and starts whatever it uncovered.
    private func apply(_ outcome: Result<LoadResult, BoardLoadError>, generation: Int, origin: WatchOrigin) {
        reloadInFlight = false

        // The stale-apply guard. Serialization means this should not trigger today, but "only the
        // newest result applies" is the rule the wholesale floor and every future overlapping-load
        // change are written against, so it is enforced rather than assumed.
        if generation == reloadGeneration {
            land(outcome, generation: generation, origin: origin)
        }

        startPendingReload()
        resumeQuiescenceWaitersIfQuiet()
    }

    private func land(_ outcome: Result<LoadResult, BoardLoadError>, generation: Int, origin: WatchOrigin) {
        // Consumed here, before the branch, because *both* outcomes end the expectation: a wholesale
        // operation gets exactly one reload to prove itself, and a second failure after it is
        // ordinary per-file breakage again.
        let endsWholesaleOperation: Bool
        if let floor = wholesaleReloadFloor, generation >= floor {
            wholesaleReloadFloor = nil
            endsWholesaleOperation = true
        } else {
            endsWholesaleOperation = false
        }

        switch outcome {
        case let .success(result):
            // **The motion language's one decision point** (03-board-ui.md § Motion, via `Motion`).
            // "User-initiated structural changes animate; foreign changes snap" cannot live at the
            // call sites here the way it did in the pathfinder — the one-way flow means the user's
            // own delete arrives back through the watcher exactly like an agent's edit — so it lives
            // at the *reload*, whose origin is already classified. `Motion` owns which origins
            // perform and in which voice; this store owns only the assignment they wrap.
            //
            // A `nil` animation is not a special case: `withAnimation(nil)` is the bare assignment,
            // which is what the snapping origins and the Reduce Motion variant both want.
            withAnimation(Motion.reloadAnimation(
                origin: origin,
                endsBracketedOperation: endsWholesaleOperation,
                reduced: Motion.prefersReducedMotion
            )) {
                snapshot = result.model
                loadWarnings = result.warnings
                // The one place transient state is re-grounded. It goes last, after `snapshot` is
                // the new one, because a view woken by the snapshot's change must never observe a
                // selection still pointing at the old tree.
                //
                // Inside the transaction deliberately: 03 § Motion has the selection highlight
                // "ride whatever transaction is active rather than easing on its own", and a
                // re-grounding that landed outside this one would be exactly the independent ease
                // that rules out.
                transient.resolve(against: result.model)
            }
            // Outside it, equally deliberately — 03 keys transactions narrowly, and the banner
            // conditions are not board structure. A lock clearing is not a thing the strip should
            // slide out on the back of a card being deleted.
            //
            // Breakage always heals on a success — it *is* the claim "the last reload failed", and
            // this one did not.
            reloadFailure = nil
            clearLockIfDisproved(by: origin)

        case let .failure(error):
            // `snapshot`, `loadWarnings` and the transient state are untouched: a failed reload
            // never replaces a good snapshot, and state over a snapshot that did not change has
            // nothing to re-resolve against. Nothing is performed here either, for the same reason —
            // and neither is anything on the lock paths below (`enterVanishedRootLock`,
            // `enterUnwritableLock`, `relocate`), none of which touch the snapshot at all. A board
            // that did not change has no motion to show.
            reloadFailure = error
            // `readOnlyLock == nil` rather than an unconditional assignment: a root that vanished
            // mid-bracket already raised its own, truer lock, and 02-architecture.md is explicit
            // that the bracket's final reload becomes a no-op there rather than a redundant
            // failure. Overwriting `.vanishedRoot` with `.bracketedReloadFailed` would also break
            // the clearing rules — the vanished root's lock must not clear on a reload that never
            // proves the root came back.
            if endsWholesaleOperation, readOnlyLock == nil {
                readOnlyLock = .bracketedReloadFailed
            }
            Self.logger.error("reload \(generation, privacy: .public) failed: \(error.description, privacy: .public)")
        }
    }

    private func startPendingReload() {
        guard !reloadInFlight, let origin = pendingReload else { return }
        pendingReload = nil
        startReload(origin)
    }

    // MARK: - The lock's clearing rules

    /// Clears the read-only lock if this successful reload actually disproved its cause.
    ///
    /// **Reason-specific, because the causes are not alike** (02-architecture.md § Write-failure
    /// surfacing):
    ///
    /// - `.bracketedReloadFailed` and `.vanishedRoot` are *disproved by the success itself*. The
    ///   first says "the tree could not be re-read after a wholesale change" and the second says
    ///   "the root is gone" — a completed tree walk at the root contradicts both, whatever origin
    ///   asked for it, so any success clears them.
    /// - `.unwritableLocation` is not. A board on a read-only DMG reloads flawlessly forever;
    ///   loading proves nothing about writing. It clears only when a **reconciling** reload — wake,
    ///   activation, a stream re-creation — re-probes writability and finds it changed ("Writability
    ///   re-probes on every reconciling reload, so a fixed permission or rewritable remount clears
    ///   the lock without ceremony").
    ///
    /// `FileManager.isWritableFile(atPath:)` is `access(2)` on the root directory: a real-uid
    /// permission question asked of the filesystem, which is what makes it answer correctly for
    /// both halves of the case — a read-only *mount* and a permission-denied *folder*.
    ///
    /// Deliberately **one-way**: a reconciling reload that finds the root unwritable does not
    /// *raise* the lock. Arming it is the open flow's job (`enterUnwritableLock()`), and inferring
    /// a lock from a probe here would be a policy decision this milestone was not asked to make.
    private func clearLockIfDisproved(by origin: WatchOrigin) {
        switch readOnlyLock {
        case nil:
            break
        case .bracketedReloadFailed, .vanishedRoot:
            readOnlyLock = nil
        case .unwritableLocation:
            guard origin == .reconciling, FileManager.default.isWritableFile(atPath: rootURL.path) else { return }
            Self.logger.debug("writability re-probe passed — the unwritable-location lock clears")
            readOnlyLock = nil
        }
    }

    // MARK: - Root identity and explicit locks

    /// Points this store at the board's new location, absorbing a rename or a move.
    ///
    /// Called by the registry when a `.rootChanged` re-resolved the board's bookmark somewhere else
    /// (02-architecture.md § Write-failure surfacing, "A renamed or moved board root follows its
    /// file identity"): the board the app has open is the *file*, not the path string, so this is
    /// bookkeeping rather than an event — **no banner, no lock, nothing was ever wrong**.
    ///
    /// It deliberately starts **no reload**. The caller follows this with the watcher's
    /// `reattach(to:)`, whose reconciling reload is the one that rebuilds the snapshot at the new
    /// root; a reload fired from here would be a second walk racing that one for no gain. Until it
    /// lands, `rootURL` is the new location and `snapshot.rootURL` is still the old — see
    /// `rootURL`'s note.
    public func relocate(to newRoot: URL) {
        guard newRoot != rootURL else { return }
        Self.logger.debug("board root relocated; Writer URLs now derive from the new location")
        rootURL = newRoot
    }

    /// Raises the vanished-root read-only lock — the registry's call, after bookmark re-resolution
    /// found nothing and the last-known path is not there either.
    ///
    /// Overwrites whatever lock was standing: a root that is gone is the most current and most
    /// specific truth about why writes are refused, and its clearing rule (a successful reload,
    /// which can only happen if the root came back) is strictly the safer one to be holding.
    public func enterVanishedRootLock() {
        Self.logger.error("board root vanished — entering the read-only lock")
        readOnlyLock = .vanishedRoot
    }

    /// Raises the unwritable-location read-only lock — the open flow's call, after probing the
    /// root's writability (02 § "An unwritable board location enters the read-only lock at open").
    /// Public now so the vocabulary and its clearing rule ship together; m4's open flow is the
    /// producer.
    ///
    /// Does **not** overwrite a standing lock: a board that is already locked for a vanished root
    /// or a failed bracketed reload has a cause that outranks "and it is also read-only", and both
    /// of those clear on a success that would then re-probe anyway.
    public func enterUnwritableLock() {
        guard readOnlyLock == nil else { return }
        Self.logger.error("board location is not writable — entering the read-only lock")
        readOnlyLock = .unwritableLocation
    }

    // MARK: - Write gate

    /// Runs a synchronous Writer operation inside the watcher bracket, so the churn it produces
    /// rounds back as one app-mediated reload rather than a scatter of foreign ones.
    ///
    /// The store does **not** touch its snapshot here, before or after. `operation` puts bytes on
    /// disk; the watcher notices; the reload applies. That indirection is the one-way flow, and it is
    /// why this method's only jobs are the gate and the bracket.
    ///
    /// **A failure posts to the banner before it is rethrown** (02-architecture.md § Write-failure
    /// surfacing): the strip is how the one-way flow keeps its honesty — the action visibly did not
    /// happen, and the banner is the only thing that says why — so no call site is trusted to
    /// remember, and a `try?` at some future call site cannot make a failure silent. The refusal
    /// below is deliberately *not* posted: the lock row is already standing, and a second row per
    /// refused gesture would bury it under echoes of itself.
    ///
    /// - Throws: `BoardStoreWriteRefusal.readOnlyLocked` if the board is locked read-only — checked
    ///   *before* the bracket opens, so a refusal costs no suspended watcher and no owed reload.
    ///   Otherwise rethrows whatever `operation` threw, which is a `BoardWriteError`. The untyped
    ///   `throws` is the price of those being two different error types today; see
    ///   `BoardStoreWriteRefusal` for why they are, and why they will not stay that way.
    ///
    /// One ergonomic wart, recorded so it is not rediscovered: when `operation` returns a value,
    /// Swift cannot infer `T` and the closure's thrown type at the same time — the thrown type
    /// widens to `any Error` and the call fails to compile. Such a call site spells the closure out
    /// (`{ () throws(BoardWriteError) -> ItemID in … }`). `Void`-returning operations, which are
    /// most of them, infer cleanly.
    @discardableResult
    public func performWrite<T>(_ operation: () throws(BoardWriteError) -> T) throws -> T {
        if let readOnlyLock {
            throw BoardStoreWriteRefusal.readOnlyLocked(readOnlyLock)
        }
        watcherBrackets?.begin()
        // `defer`, not a trailing call: a Writer operation that fails partway has still touched disk,
        // and an unbalanced bracket would leave the watcher suspended for the rest of the session.
        defer { watcherBrackets?.end() }
        // `do throws(BoardWriteError)`: without the annotation the `catch` widens to `any Error` and
        // the Writer's typed error is lost on the way to the banner.
        do throws(BoardWriteError) {
            return try operation()
        } catch {
            banners.post(error)
            throw error
        }
    }

    /// Runs an operation that rewrites the tree **wholesale** — pull-rebase, branch switch, undo
    /// restore (06-history-undo.md, 07-sync-collab.md) — under the bracket, and arms the rule that
    /// its closing reload must succeed.
    ///
    /// Two things distinguish this from `performWrite`:
    ///
    /// - The bracket is load-bearing rather than tidy: a reload landing mid-operation would render a
    ///   half-checked-out tree.
    /// - Failure of the closing reload **locks the board** (`ReadOnlyLockReason.bracketedReloadFailed`).
    ///   After a wholesale change the last-good snapshot may describe a different branch entirely, so
    ///   writes derived from it would land nonsense — unlike ordinary per-file breakage, where the
    ///   snapshot still describes the tree.
    ///
    /// The expectation is armed on **every** exit path, a thrown error included: an operation that
    /// died partway is precisely the case where the tree's state is unknown and the next reload had
    /// better be the authority on it.
    ///
    /// - Throws: `BoardStoreWriteRefusal.readOnlyLocked` if the board is already locked — a locked
    ///   board refuses to *start* wholesale work, not just ordinary writes. Otherwise rethrows
    ///   `operation`'s error. (Spelled `throws` rather than `rethrows` because of that refusal: a
    ///   `rethrows` function may only throw errors its closure threw.)
    public func performWholesale(_ operation: () throws -> Void) throws {
        if let readOnlyLock {
            throw BoardStoreWriteRefusal.readOnlyLocked(readOnlyLock)
        }
        watcherBrackets?.begin()
        defer {
            // Ordered: arm first, then close the bracket. `endBracket()` is what schedules the
            // post-bracket reload, and with a `nil` watcher a consumer may signal by hand the instant
            // this returns — either way the floor has to be in place before any walk can start.
            wholesaleReloadFloor = reloadGeneration + 1
            watcherBrackets?.end()
        }
        do {
            try operation()
        } catch let error as BoardWriteError {
            // Same honesty rule as `performWrite`, applied to the one error type the banner has
            // phrasing for. A wholesale operation is usually git's (m7), whose own failure
            // vocabulary is not `BoardWriteError` and whose surfacing — the suspended-history
            // condition, the in-progress row swapping for an error — is the committer's to drive;
            // but a `BoardWriteError` escaping here is an ordinary failed write and may no more
            // bypass the strip than one from `performWrite`.
            banners.post(error)
            throw error
        }
    }

    // MARK: - Lane width

    /// Writes a lane's width — the one commit point both width mechanisms share (03-board-ui.md §
    /// Lane): the right-edge drag's release and the stepper's ⌥⌘→/⌥⌘← both land here, and they differ
    /// only in what they did to the *window* on the way (the drag grew it, the stepper did not).
    ///
    /// **The value written is an integer, replacing whatever was there.** `width` is a lenient field
    /// on the read side — missing, malformed, zero and negative all render as one unit
    /// (`LaneLayoutMath.displayUnits`) with the author's bytes left alone — but an explicit width
    /// change is the user overwriting that value, so the Writer puts a plain integer in its place
    /// (01-storage-format.md § Frontmatter).
    ///
    /// Three ways this does nothing, all deliberate: a count below 1 clamps to 1 (a lane spans at
    /// least one unit), an id that is not in the snapshot is ignored (the lane vanished under the
    /// gesture — the reload that removed it is the authority), and a count already equal to what the
    /// lane displays writes nothing (a drag that ends where it started must not stamp `modified` or
    /// mint a git commit).
    ///
    /// Failures are already the banner's: `performWrite` posts every `BoardWriteError` before it
    /// rethrows, so the rethrow is swallowed here rather than propagated to a gesture that has no
    /// second thing to do about it. The lane stays at its old width, which is the truth — nothing was
    /// written.
    public func setLaneWidth(_ id: ItemID, units: Int) {
        let clamped = max(1, units)
        guard let lane = snapshot.lanes.first(where: { $0.id == id }),
              LaneLayoutMath.displayUnits(of: lane) != clamped
        else { return }

        let folder = rootURL.appendingPathComponent(id.rawValue)
        // The closure's signature is spelled out because of `try?`: with the error discarded at the
        // call site Swift stops inferring the typed `throws(BoardWriteError)` and widens it to `any
        // Error`, which `performWrite` will not take. Same wart as the value-returning call sites
        // `performWrite`'s doc comment records, arriving from the other direction.
        try? performWrite { () throws(BoardWriteError) -> Void in
            try BoardWriter.updateIndex(inItemFolder: folder, operation: .resize(title: nil)) { document in
                document.set(FrontmatterKeys.width, to: .int(clamped))
            }
        }
    }

    // MARK: - Styling

    /// One item a style gesture is about to act on: where its `index.md` is, and what the two styled
    /// keys currently say there.
    ///
    /// The editor reads these for its per-dimension current-value display (`StyleFieldState.resolve`)
    /// and `applyStyle` reads the *same* values to decide what is a no-op, so the display and the
    /// write can never disagree about what is already on disk. `id` is `nil` for the board root,
    /// which has no `ItemID` by design (see `ItemID`'s doc comment).
    public struct StyleSubject: Sendable, Equatable {
        public let id: ItemID?
        public let folder: URL
        public let background: FieldValue<String>
        public let icon: FieldValue<String>
    }

    /// The live items `target` names, in display order — lanes left to right, each lane's cards top
    /// to bottom.
    ///
    /// **Vanished targets are simply absent**, ancestor walk included: a tombstoned card, a card
    /// under a tombstoned lane, and an id that names nothing all contribute no subject, which is the
    /// same silent skip `commitRename` gives a vanished rename target — "nothing is ever written into
    /// a vanished folder". A style editor whose set has emptied dismisses (`StyleEditorSession`), so
    /// an empty result is a frame's worth of nothing to show rather than a state to handle.
    public func styleSubjects(of target: StyleTarget) -> [StyleSubject] {
        switch target {
        case .board:
            return [StyleSubject(
                id: nil,
                folder: rootURL,
                background: snapshot.background,
                icon: snapshot.icon
            )]

        case let .items(ids):
            var subjects: [StyleSubject] = []
            for lane in snapshot.lanes where !lane.isDeleted {
                let laneFolder = rootURL.appendingPathComponent(lane.id.rawValue)
                if ids.contains(lane.id) {
                    subjects.append(StyleSubject(
                        id: lane.id,
                        folder: laneFolder,
                        background: lane.background,
                        icon: lane.icon
                    ))
                }
                for card in lane.cards where !card.isDeleted && ids.contains(card.id) {
                    subjects.append(StyleSubject(
                        id: card.id,
                        folder: laneFolder.appendingPathComponent(card.id.rawValue),
                        background: card.background,
                        icon: card.icon
                    ))
                }
            }
            return subjects
        }
    }

    /// Which level `target` sits at — the editor's symbol grid needs it for its leading well, "the
    /// level's default symbol" (03-board-ui.md § Styling ▸ Controls).
    ///
    /// A set naming any card is a card set: 04-interactions.md's cards-XOR-lanes rule means a live
    /// selection never mixes the two, so the branch below is a total answer rather than a policy —
    /// and if a mixed set ever reached here, `doc.text` is the level whose default would actually be
    /// removed by the leading well.
    public func styleLevel(of target: StyleTarget) -> StyleLevel {
        switch target {
        case .board:
            return .board
        case let .items(ids):
            let namesACard = snapshot.lanes.contains { lane in
                !lane.isDeleted && lane.cards.contains { !$0.isDeleted && ids.contains($0.id) }
            }
            return namesACard ? .card : .lane
        }
    }

    /// Writes a style gesture — the **one** commit point every anchor shares (03-board-ui.md §
    /// Styling ▸ Controls: "One component, one behavior, three anchors"), and the quick-style recents
    /// row with them.
    ///
    /// **One bracket, whatever the target set's size.** "Choosing a well applies to the whole
    /// selection — one gesture, one commit on git boards" (§ Controls), so every target's `index.md`
    /// is rewritten inside a single `performWrite`: the churn rounds back as one app-mediated reload,
    /// and the auto-committer (m7) sees one operation rather than N.
    ///
    /// **No-ops are skipped per dimension and per target** — `setLaneWidth`'s rule, for its reason: a
    /// well clicked twice, or a batch where half the cards are already that colour, must not stamp
    /// `modified` or mint a commit on the items that were already right. A dimension whose value is
    /// already what the gesture asks contributes nothing; a target both of whose dimensions are
    /// no-ops is dropped entirely; and a gesture that changes nothing anywhere never opens the
    /// bracket at all.
    ///
    /// **`iconColor` is not a parameter, and that is the design**: it is "resolved — schema yes,
    /// control no" (§ Capabilities). The field renders when hand-written and the app offers no
    /// control for it, so there is nothing here to pass.
    ///
    /// Failure is `performWrite`'s: the banner is posted before the rethrow, which is swallowed here
    /// like every other gesture with no second thing to do. A batch that fails partway leaves the
    /// targets written before it written — the Writer is "atomic per filesystem operation, not per
    /// gesture" — and the reload shows the true state, which is the honest one.
    public func applyStyle(to target: StyleTarget, background: StyleChange = .keep, icon: StyleChange = .keep) {
        let edits: [(folder: URL, background: StyleChange, icon: StyleChange)] = styleSubjects(of: target)
            .compactMap { subject in
                let background = Self.effective(background, against: subject.background)
                let icon = Self.effective(icon, against: subject.icon)
                guard background != .keep || icon != .keep else { return nil }
                return (folder: subject.folder, background: background, icon: icon)
            }
        guard !edits.isEmpty else { return }

        try? performWrite { () throws(BoardWriteError) -> Void in
            for edit in edits {
                // `.style(title: nil)`: `updateIndex` enriches it off the document it reads, so a
                // failure names the item by the title it still has (see `WriteOperation.style`).
                try BoardWriter.updateIndex(inItemFolder: edit.folder, operation: .style(title: nil)) { document in
                    Self.apply(edit.background, to: FrontmatterKeys.background, in: &document)
                    Self.apply(edit.icon, to: FrontmatterKeys.icon, in: &document)
                }
            }
        }
    }

    /// `change` narrowed against what is already on disk: `.keep` when it would write what is
    /// already there.
    ///
    /// The comparison is against the **valid** reading, not the written text: a malformed value —
    /// `background: [a, b]`, a sequence where a scalar belongs — is never equal to a palette name, so
    /// choosing a well always replaces it, which is what "choosing any well replaces it" (§ Controls)
    /// promises about a value the app could not read.
    nonisolated static func effective(_ change: StyleChange, against field: FieldValue<String>) -> StyleChange {
        switch change {
        case .keep: .keep
        case let .set(value): field.value == value ? .keep : .set(value)
        case .remove: field.isMissing ? .keep : .remove
        }
    }

    private static func apply(_ change: StyleChange, to key: String, in document: inout FrontmatterDocument) {
        switch change {
        case .keep: break
        case let .set(value): document.set(key, to: .string(value))
        case .remove: document.remove(key)
        }
    }

    // MARK: - Creation

    /// Creates a lane at the board's right end — File ▸ New Lane ⇧⌘N (11-command-nexus.md).
    ///
    /// **Untitled, and deliberately with no inline editor.** 03-board-ui.md gives lane titles one
    /// editing surface — "Inline rename on the header" — and 04-interactions.md gives that surface
    /// one entry point, Board ▸ Rename, "since Return on a lane creates a card". Nothing in either
    /// doc opens an editor *at creation*, so a new lane appears with the untitled placeholder and
    /// the user renames it if they want a name. Titles are optional at every level; a lane with no
    /// `title` key is a legitimate resting state, not a half-finished one.
    ///
    /// Like `setLaneWidth`, the rethrow is swallowed: `performWrite` has already posted the banner,
    /// and a menu item has no second thing to do about a failure.
    public func createLane() {
        let root = rootURL
        try? performWrite { () throws(BoardWriteError) -> Void in
            _ = try BoardWriter.createLane(inBoard: root, title: nil)
        }
    }

    // MARK: - The new-card placeholder's commit

    /// Turns the open placeholder into a real card — the write half of 02-architecture.md §
    /// Layering's one named exception to the one-way flow.
    ///
    /// The five outcomes, all settled:
    ///
    /// - **No placeholder, or one already committed** — nothing to do. (Idempotence matters: Return
    ///   commits, and the field's focus-loss handler fires immediately afterwards.)
    /// - **An empty title discards it** — "creating-then-abandoning never leaves an empty card
    ///   behind" (04-interactions.md ▸ Grammar). Whitespace counts as empty: a title of three
    ///   spaces is a slip, not a deliberate untitled card.
    /// - **A vanished lane discards it** — the anchor is gone, so there is nowhere to file the
    ///   card; the reload that removed the lane is the authority.
    /// - **A failed create discards it too** (settled, 02 § Layering): "the overlay never waits for
    ///   a card that cannot arrive". The failure is already the banner's.
    /// - **A successful create hands off**: the overlay flips to `.awaitingArrival` and stands until
    ///   the watcher round-trips the real card, so the user never sees a hole where they just typed.
    ///
    /// - Returns: the created card's id, or `nil` on any of the discard paths — which is what the
    ///   ⌘↩ call site needs to know whether it has a card window to open.
    @discardableResult
    public func commitPlaceholder() -> ItemID? {
        guard let placeholder = transient.newCardPlaceholder, placeholder.phase == .editing else { return nil }

        let title = placeholder.draftTitle.trimmingCharacters(in: .whitespacesAndNewlines)
        guard !title.isEmpty,
              let lane = snapshot.lanes.first(where: { $0.id == placeholder.laneID && !$0.isDeleted })
        else {
            transient.discardPlaceholder()
            return nil
        }

        let laneFolder = rootURL.appendingPathComponent(placeholder.laneID.rawValue)
        let visible = lane.cards.filter { !$0.isDeleted }
        // `nil` means "append", which is `createCard`'s own default — so the anchored case is the
        // only one that needs a rank at all.
        let position = Self.insertionIndex(after: placeholder.anchorCardID, among: visible)

        let created = try? performWrite { () throws(BoardWriteError) -> ItemID in
            let id = try BoardWriter.createCard(inLane: laneFolder, title: title)
            guard let position else { return id }

            // The rank is computed here rather than passed to `createCard` because the create's
            // contract is "append after the visible siblings" and widening it would give every
            // caller a position to think about. The reposition rides the Writer's own same-parent
            // degenerate reorder — "a move whose destination is the item's current parent degrades
            // to a plain reorder" — inside the *same* `performWrite`, so the pair rounds back as
            // one app-mediated reload rather than showing the card at the bottom for a frame.
            var rank = Ranks.insertionRank(amongVisible: visible.map(\.order), at: position)
            if rank == nil {
                // Midpoint precision exhausted between the anchor and its neighbour
                // (01-storage-format.md § Ordering). Compact, then place against the fresh ranks:
                // the new card is not among the renumbered siblings — it was appended past them —
                // so the compacted ladder lines up one-for-one with `visible`.
                try BoardWriter.renumberVisibleChildren(of: laneFolder)
                rank = Ranks.insertionRank(amongVisible: Ranks.renumbered(count: visible.count), at: position)
            }
            guard let rank else { return id }

            _ = try BoardWriter.moveItem(
                at: laneFolder.appendingPathComponent(id.rawValue),
                toParent: laneFolder,
                sourceBoardRoot: rootURL,
                destinationBoardRoot: rootURL,
                order: rank
            )
            return id
        }

        guard let created else {
            transient.discardPlaceholder()
            return nil
        }
        transient.commitPlaceholder(expecting: created)
        return created
    }

    /// The display position a new card takes, or `nil` for "append at the bottom".
    ///
    /// An anchor that is not among `visible` degrades to `nil` rather than failing: the card the
    /// ⌘N target rule named was deleted or moved away mid-typing, and the lane — the anchor that
    /// actually matters — is still there. Appending is the honest fallback; refusing to create
    /// would punish the user for someone else's edit.
    nonisolated static func insertionIndex(after anchor: ItemID?, among visible: [Card]) -> Int? {
        guard let anchor, let index = visible.firstIndex(where: { $0.id == anchor }) else { return nil }
        // Already last: "immediately after it" and "at the bottom" are the same position, and
        // append needs no rank of its own.
        return index + 1 < visible.count ? index + 1 : nil
    }

    // MARK: - Inline rename

    /// Writes the open rename editor's draft — the third inline editor's commit
    /// (04-interactions.md ▸ Grammar), reached by Return **and** by focus loss ("a rename commits
    /// … the deliberate exception being the placeholder, because nothing exists on disk yet").
    ///
    /// Four rules, all from 04 and 03:
    ///
    /// - **The editor closes first, unconditionally.** Every path below ends with it gone, and
    ///   retiring it up front is what makes this idempotent — Return commits and the field's
    ///   focus-loss handler fires an instant later against no editor at all.
    /// - **A vanished target writes nothing, silently.** "A target that is tombstoned, deleted, or
    ///   gone at commit time discards the editor and its keystrokes silently … nothing is ever
    ///   written into a vanished folder, and no partial `index.md` can resurrect deleted data."
    ///   Liveness is effective — a card under a tombstoned lane is vanished too.
    /// - **An empty commit removes the `title` key** (03-board-ui.md § Card face; 04 ▸ Selection:
    ///   "Committing an empty rename on an existing item removes its `title` key"), rather than
    ///   writing `title: ""` — titles are optional, and the face shows the untitled placeholder.
    /// - **An unchanged title writes nothing.** `setLaneWidth`'s rule, for the same reason: an
    ///   editor opened and dismissed with Return must not stamp `modified` or mint a commit.
    ///
    /// The folder is re-derived from the *current* snapshot, which is what makes a foreign move
    /// mid-rename invisible: the editor follows the UUID, and the write lands wherever the item is
    /// now.
    public func commitRename() {
        guard let editor = transient.renameEditor else { return }
        transient.discardRename()

        guard let target = Self.liveItem(editor.targetID, in: snapshot) else { return }

        let typed = editor.draftTitle.trimmingCharacters(in: .whitespacesAndNewlines)
        let newTitle: String? = typed.isEmpty ? nil : typed
        guard newTitle != target.title else { return }

        var folder = rootURL.appendingPathComponent(target.laneID.rawValue)
        if let cardID = target.cardID {
            folder.append(component: cardID.rawValue)
        }

        try? performWrite { () throws(BoardWriteError) -> Void in
            // `.rename(title: nil)`: `updateIndex` enriches it off the document it reads, so the
            // banner names the item by the title it still has rather than the one that failed to
            // land (see `WriteOperation.rename`).
            try BoardWriter.updateIndex(inItemFolder: folder, operation: .rename(title: nil)) { document in
                if let newTitle {
                    document.set(FrontmatterKeys.title, to: .string(newTitle))
                } else {
                    document.remove(FrontmatterKeys.title)
                }
            }
        }
    }

    /// Where a live item lives and what it is currently called, or `nil` when the id names nothing
    /// the board renders.
    ///
    /// **Effective liveness, ancestor-walked** — the same rule `CardWindowHost.cardWindowFate`
    /// applies to a card window and `ItemReferenceSet` applies to the selection: a card under a
    /// tombstoned lane renders nowhere, so it is as gone as a deleted one. The path is returned as
    /// its two identity components rather than as a URL so the caller builds it off the store's
    /// *current* `rootURL`, which a mid-session folder rename may have moved.
    nonisolated static func liveItem(
        _ id: ItemID,
        in snapshot: BoardModel
    ) -> (laneID: ItemID, cardID: ItemID?, title: String?)? {
        for lane in snapshot.lanes where !lane.isDeleted {
            if lane.id == id {
                return (laneID: lane.id, cardID: nil, title: lane.title.value)
            }
            if let card = lane.cards.first(where: { $0.id == id && !$0.isDeleted }) {
                return (laneID: lane.id, cardID: card.id, title: card.title.value)
            }
        }
        return nil
    }

    // MARK: - Board rename

    /// Writes the board's own `title` — the board popover's rename field (03-board-ui.md § Board
    /// popover), and the one rename in the app with no item to aim at.
    ///
    /// **It edits frontmatter, never the folder**: "Rename edits the board's frontmatter `title`
    /// only — the folder is never renamed by the app; the Finder document name is Finder's to
    /// change" (§ Board popover, 01-storage-format.md § Board naming). The app's display name and
    /// the Finder document name may therefore diverge, which is accepted rather than reconciled.
    ///
    /// The three commit rules are `commitRename`'s, deliberately identical — one rename vocabulary
    /// whatever level it is aimed at:
    ///
    /// - **Trimmed**, so a title of three spaces is a slip rather than a name.
    /// - **An empty commit removes the key.** A board with no `title` falls back to its *folder
    ///   name* (§ Board naming) — never the "Untitled" placeholder cards and lanes show, and never
    ///   `title: ""`, which would be a real if blank title with nothing to fall back to.
    /// - **An unchanged title writes nothing**, so a popover opened and dismissed with Return
    ///   neither stamps `modified` nor mints a commit.
    ///
    /// There is no vanished-target guard, because a board cannot tombstone itself out of its own
    /// window (01-storage-format.md § Deletion): the only way this target goes away is the root
    /// itself vanishing, which is the read-only lock's story, and `performWrite` refuses under it
    /// before anything touches disk.
    public func renameBoard(_ title: String?) {
        let typed = (title ?? "").trimmingCharacters(in: .whitespacesAndNewlines)
        let newTitle: String? = typed.isEmpty ? nil : typed
        guard newTitle != snapshot.title.value else { return }

        let folder = rootURL
        try? performWrite { () throws(BoardWriteError) -> Void in
            // `.rename(title: nil)`: `updateIndex` enriches it off the document it reads, so a
            // refusal names the board by the title it still has (see `WriteOperation.rename`).
            try BoardWriter.updateIndex(inItemFolder: folder, operation: .rename(title: nil)) { document in
                if let newTitle {
                    document.set(FrontmatterKeys.title, to: .string(newTitle))
                } else {
                    document.remove(FrontmatterKeys.title)
                }
            }
        }
    }

    // MARK: - Lane reorder

    /// Commits a lane drag: `id` lands at display position `index` among the board's live lanes,
    /// counted **with the dragged lane itself removed** — which is the index
    /// `LaneReorderMath.proposedIndex` produces.
    ///
    /// Within-board only. A cross-board lane drag is the locality model's (04-interactions.md ▸
    /// Drag and drop) and belongs to m5's drag card; here source and destination board roots are
    /// the same URL, so `moveItem` takes its same-parent degenerate-reorder path and rewrites
    /// exactly one file — the moved lane's `order`.
    ///
    /// **A drag that ends where it started writes nothing**: `index == from` re-inserts the lane in
    /// its own slot, and a no-op must not stamp `modified` or mint a commit — the resize drag's
    /// rule, and for the same reason.
    public func moveLane(_ id: ItemID, toIndex index: Int) {
        let lanes = snapshot.lanes.filter { !$0.isDeleted }
        guard let from = lanes.firstIndex(where: { $0.id == id }) else { return }

        var remaining = lanes
        remaining.remove(at: from)
        let target = min(max(0, index), remaining.count)
        guard target != from else { return }

        let root = rootURL
        let folder = root.appendingPathComponent(id.rawValue)
        try? performWrite { () throws(BoardWriteError) -> Void in
            var rank = Ranks.insertionRank(amongVisible: remaining.map(\.order), at: target)
            if rank == nil {
                // Compact and place again. Unlike the card case the dragged lane *is* among the
                // renumbered children — it is a real folder on disk — so its fresh rank is dropped
                // from the ladder before the neighbours are consulted.
                try BoardWriter.renumberVisibleChildren(of: root)
                var compacted = Ranks.renumbered(count: lanes.count)
                compacted.remove(at: from)
                rank = Ranks.insertionRank(amongVisible: compacted, at: target)
            }
            guard let rank else { return }

            _ = try BoardWriter.moveItem(
                at: folder,
                toParent: root,
                sourceBoardRoot: root,
                destinationBoardRoot: root,
                order: rank
            )
        }
    }

    // MARK: - The trash

    /// Whether physically removing an item on this board destroys the only copy of it — and
    /// therefore whether Delete Immediately stands an alert between one keystroke and unrecoverable
    /// deletion (03-board-ui.md § Trash, "Delete Immediately confirms exactly where the loss is
    /// real").
    ///
    /// **Every board is `true` today**, because every board is history mode *none*: nothing in the
    /// app keeps a second copy, so a purge is final everywhere.
    ///
    // m7-git: git boards answer `false` here — "on git boards it acts immediately, since the content
    // remains reachable in history" (06-history-undo.md's delete-never-forgets). Repo-nested boards
    // stay `true` alongside mode none: the app manages no history for them either. The named
    // predicate exists now so the committer card changes one expression rather than hunting the
    // confirmation logic out of two menu items and an alert.
    public var purgeIsUnrecoverable: Bool { true }

    /// Tombstones the current selection — File ▸ Delete ⌘⌫ and its plain-⌫ grammar twin
    /// (04-interactions.md ▸ The map, 11-command-nexus.md).
    ///
    /// A convenience over `delete(_:)` so the two call sites cannot disagree about *what* the
    /// command acts on.
    public func deleteSelection() {
        delete(selection.ids)
    }

    /// Tombstones every live item in `ids` — cards or lanes, in one bracket.
    ///
    /// **One `performWrite` whatever the set's size**, matching the style batch's rule and for its
    /// reason: one gesture, one app-mediated reload, and (on git boards) one commit rather than N.
    /// A lane's tombstone rewrites only the lane's own `index.md` — hiding the subtree is the
    /// renderer's ancestor walk, not a stored flag (`BoardWriter.deleteItem`).
    ///
    /// **Tombstoned ids are silently skipped**, not refused: the paths are resolved on the live side
    /// only, so a selection the next reload will drop writes nothing rather than re-stamping a
    /// `deleted:` that is already there. An empty resolution never opens the bracket at all.
    ///
    /// The selection is **cleared**, not moved to a successor. 04-interactions.md ▸ The map asks for
    /// the Finder-style successor sibling ("repeated ⌫ walks down a lane"), which needs the
    /// navigation order the keyboard grammar defines — that is m5's card. Clearing is the honest
    /// interim: what was selected renders nowhere now, and the reload's resolve rule would empty the
    /// set a moment later anyway.
    public func delete(_ ids: Set<ItemID>) {
        let folders = TrashModel.paths(of: ids, on: .live, in: snapshot).map { $0.folder(under: rootURL) }
        guard !folders.isEmpty else { return }

        try? performWrite { () throws(BoardWriteError) -> Void in
            for folder in folders {
                try BoardWriter.deleteItem(at: folder)
            }
        }
        // m5-keyboard: the successor-selection grammar replaces this line.
        clearSelection()
    }

    /// Put Back: removes `deleted:` from every tombstoned item in `ids`, in one bracket
    /// (03-board-ui.md § Trash).
    ///
    /// **Restore fidelity is perfect because nothing ever moved** — the item re-enters the visible
    /// set at its recorded `order` among its current siblings, and the folder is exactly where it
    /// has been all along (`BoardWriter.restoreItem`).
    ///
    /// **Putting back a lane splits its contents by flag for free.** The write is the lane's own
    /// `index.md` and nothing else, so cards hidden *with* the lane return with it while cards
    /// carrying their own `deleted:` stay tombstoned — and their rows reappear in the trash, which
    /// is exactly the two-step recovery the design settled on.
    ///
    /// A card whose lane is itself tombstoned cannot be reached this way through the UI (it has no
    /// row — `TrashModel.entries`), but the path resolution admits it, and the outcome is the pure
    /// view's honest one: the key is removed, and the card still renders nowhere because its lane
    /// is still tombstoned. Putting the lane back then shows it.
    ///
    /// The selection is deliberately left alone: the restored items flip liveness, and the reload's
    /// resolve rule ejects them from a `.trashed` set as a vanish — the same silent shrink an
    /// external restore would produce.
    public func putBack(_ ids: Set<ItemID>) {
        let folders = TrashModel.paths(of: ids, on: .trashed, in: snapshot).map { $0.folder(under: rootURL) }
        guard !folders.isEmpty else { return }

        try? performWrite { () throws(BoardWriteError) -> Void in
            for folder in folders {
                try BoardWriter.restoreItem(at: folder)
            }
        }
    }

    /// Delete Immediately ⌥⌘⌫: physically removes every tombstoned item in `ids` (03-board-ui.md §
    /// Trash), in one bracket.
    ///
    /// **The confirmation is not here.** Whether the loss is real is `purgeIsUnrecoverable`'s
    /// question and the alert is the window's; a store method that put up its own dialog could not
    /// be driven from a test, and the same purge is reached by two surfaces (the menu item and the
    /// trash row's context menu) that must not each grow their own copy of the rule.
    ///
    /// Purging a **lane** takes its whole folder — every card inside it, tombstoned or not. That is
    /// what the lane entry subsuming its subtree means on disk.
    public func deleteImmediately(_ ids: Set<ItemID>) {
        let folders = TrashModel.paths(of: ids, on: .trashed, in: snapshot).map { $0.folder(under: rootURL) }
        guard !folders.isEmpty else { return }

        try? performWrite { () throws(BoardWriteError) -> Void in
            for folder in folders {
                try BoardWriter.purgeItem(at: folder)
            }
        }
        // Nothing the set named exists any more, on either side of the boundary — unlike Put Back,
        // where the items merely changed sides, there is no vanish for the reload to notice on the
        // trashed side that would not equally be a vanish here.
        clearSelection()
    }

    /// Empty Trash… ⇧⌘⌫: purges **every** tombstone on the board, in one bracket.
    ///
    /// **Whole-trash scope, search-independent** (03-board-ui.md § Trash, settled): the targets come
    /// from the snapshot, never from the filtered view — "a bulk command about the trash itself never
    /// silently narrows to the visible subset". The filter does not reach this method at all, which
    /// is the strongest form of that guarantee.
    ///
    /// Cards carrying their own `deleted:` under a tombstoned lane go too, without being listed:
    /// they live inside the lane folder this removes (`TrashModel.emptyTrashTargets`).
    public func emptyTrash() {
        let folders = TrashModel.emptyTrashTargets(in: snapshot).map { $0.folder(under: rootURL) }
        guard !folders.isEmpty else { return }

        try? performWrite { () throws(BoardWriteError) -> Void in
            for folder in folders {
                try BoardWriter.purgeItem(at: folder)
            }
        }
        clearSelection()
    }

    /// Drag-to-restore: a tombstoned card row dropped over a live lane comes back **into that lane**
    /// (03-board-ui.md § Trash, 04-interactions.md ▸ The trash).
    ///
    /// Two writes in **one bracket**, and the order is load-bearing: `restoreItem` first — the folder
    /// is still where the trash row said it was — then, only when the destination differs, the move.
    /// Doing it the other way round would have the second call chasing a folder the first one had
    /// already relocated.
    ///
    /// **Same lane is a plain Put Back**: the key is removed and nothing else is touched, so the card
    /// returns at its recorded `order` rather than at the bottom. "Folder moved only if the
    /// destination lane differs" is the design's own wording, and the position-perfect restore is the
    /// point of the trash being a pure view.
    ///
    // m5-drag: two things arrive with the drag card's `DropSlot` port. (1) The **positional** drop —
    // the design restores "at the drop position", and the append below is the interim; the rank comes
    // from `Ranks.insertionRank` over the destination's visible cards, exactly as `commitPlaceholder`
    // computes it. (2) **Cross-board locality** — a drop on another board is a live *copy* by
    // default with the tombstoned original staying put, and ⌘-drag forces the true restore-move.
    // Both need the drag controller's target vocabulary; this method is deliberately within-board.
    ///
    /// Silent no-ops, all of them the reload being the authority rather than this gesture: a
    /// destination lane that is gone or tombstoned, a card that is not a trash row (its own flag
    /// unset, or its lane tombstoned so it has no row to drag), and an id that names nothing.
    public func restoreByDrag(cardID: ItemID, intoLane laneID: ItemID) {
        guard snapshot.lanes.contains(where: { $0.id == laneID && !$0.isDeleted }),
              let source = snapshot.lanes.first(where: { lane in
                  !lane.isDeleted && lane.cards.contains { $0.id == cardID && $0.isDeleted }
              })
        else { return }

        let root = rootURL
        let cardFolder = TrashModel.ItemPath(laneID: source.id, cardID: cardID).folder(under: root)
        let destination = root.appendingPathComponent(laneID.rawValue, isDirectory: true)
        let crossesLanes = source.id != laneID

        try? performWrite { () throws(BoardWriteError) -> Void in
            try BoardWriter.restoreItem(at: cardFolder)
            guard crossesLanes else { return }
            // `order: nil` is the Writer's own append — computed over the destination's *visible*
            // siblings, which the arriving card is not yet among.
            _ = try BoardWriter.moveItem(
                at: cardFolder,
                toParent: destination,
                sourceBoardRoot: root,
                destinationBoardRoot: root,
                order: nil
            )
        }
    }

    // MARK: - Selection (delegated)

    // The thin pass-throughs to `transient`, and the only ones.
    //
    // **Conveniences, not a second home.** The selection is the transient state every command site
    // touches — menu validation, ⌫, paste anchoring, Select All — and `store.selection` reads better
    // at each of them than `store.transient.selection` while meaning exactly the same thing. Nothing
    // is stored here: `selection` is computed and the two mutators forward, so there is no second
    // copy to go stale. Drag membership, the pending cut, the query and the editors get no such
    // shortcuts — they have one or two call sites each, and a delegate per field would be the
    // grab-bag reassembling itself on this class.
    //
    // `isEditingInline` earns one for the selection's reason and no other: **every** board-mutating
    // menu item validates against it (04-interactions.md's focused-editor rule), and a rule read
    // that often should read as one word.

    /// The board's selection, re-resolved against every snapshot this store applies —
    /// `TransientBoardState.selection` under a shorter name.
    public var selection: ItemReferenceSet { transient.selection }

    /// Whether an inline title editor is open — `TransientBoardState.isEditingInline`, which owns
    /// what it means and why every mutating command reads it.
    public var isEditingInline: Bool { transient.isEditingInline }

    /// Replaces the selection, and **records the lane it lands in** as the last-active one.
    ///
    /// The lane bookkeeping lives here rather than in `TransientBoardState` for one reason: it
    /// takes a snapshot to answer "which lane is that". 04-interactions.md's ⌘N target rule calls
    /// for "the lane that most recently held selection or a creation", and a *card* selection is
    /// its lane holding selection just as much as the lane's own header click is — so both are
    /// noted here, and creation notes itself in `beginPlaceholder`.
    public func select(_ ids: Set<ItemID>, liveness: Liveness, anchor: ItemID? = nil) {
        transient.select(ids, liveness: liveness, anchor: anchor)
        transient.noteActiveLane(Self.lane(holding: ids, in: snapshot))
    }

    /// **Every pointer click on a selectable surface goes through here** — card face, lane header,
    /// lane empty space, trash row — so 04-interactions.md § Selection's grammar is stated once
    /// (`SelectionGrammar`) rather than four times with three of them subtly different.
    ///
    /// The store's whole contribution is supplying the three inputs the grammar cannot see (the
    /// snapshot, the selection, the anchor) and storing the outcome. An emptied outcome clears
    /// rather than storing an empty set on a side, because that is what "nothing selected" is
    /// everywhere else in the app.
    ///
    /// - Parameter togglesOnRepeat: the lane's click-again-to-unselect — see `SelectionGrammar`.
    public func click(_ target: SelectionTarget, modifier: ClickModifier, togglesOnRepeat: Bool = false) {
        let outcome = SelectionGrammar.click(
            target,
            modifier: modifier,
            selection: selection,
            anchor: transient.selectionAnchor,
            snapshot: snapshot,
            togglesOnRepeat: togglesOnRepeat
        )
        guard !outcome.selection.isEmpty else {
            clearSelection()
            return
        }
        // The anchor is passed through explicitly: `select`'s default would otherwise re-anchor a
        // ⇧-range's single-member edge case on the target, and the grammar's answer is the one that
        // knows whether this click was an origin or an extension.
        select(outcome.selection.ids, liveness: outcome.selection.liveness, anchor: outcome.anchor)
    }

    /// **Select All** — "all visible cards on the board" (04-interactions.md ▸ The map), with the
    /// trash's own reading of the same command when the trash side is the one in play.
    ///
    /// Two branches, and the trash's is the narrow one: it fires only when the column is **shown**,
    /// the selection is on the trashed side, and it still names a row — the exact conditions under
    /// which "all" could mean anything but the board. It then selects every trash row **of the
    /// selection's kind**, because 04 ▸ The trash's card-entries-XOR-lane-entries rule binds a
    /// wholesale selection as tightly as it binds a click. A trashed selection naming nothing (a
    /// foreign Put Back, a purge) falls through to the board rather than selecting the trash
    /// wholesale on a guess.
    ///
    /// The anchor **survives if it is still in the set** and is dropped otherwise: Select All is not
    /// a click, so it names no new origin, but it has no business discarding one that is still
    /// standing inside what it selected.
    ///
    // m5-search: "filter-respecting, like every surface" (04 ▸ The map). The universe here is
    // `SelectionGrammar`'s order lists, which is where the filter threads in — one change, and both
    // this command and every ⇧-range narrow together.
    public func selectAll() {
        if transient.isTrashVisible, selection.liveness == .trashed, !selection.isEmpty,
           let kind = SelectionGrammar.kind(of: selection, in: snapshot) {
            apply(Set(SelectionGrammar.trashEntries(of: kind, in: snapshot)), on: .trashed)
            return
        }
        apply(Set(SelectionGrammar.liveCards(in: snapshot)), on: .live)
    }

    /// Select All's storage half: an empty universe clears rather than storing an empty set, and the
    /// anchor is kept only while it is still inside what was selected.
    private func apply(_ ids: Set<ItemID>, on side: Liveness) {
        guard !ids.isEmpty else {
            clearSelection()
            return
        }
        let anchor = transient.selectionAnchor.flatMap { ids.contains($0) ? $0 : nil }
        select(ids, liveness: side, anchor: anchor)
    }

    /// Selects nothing — Escape's last step outward (04-interactions.md ▸ Grammar).
    ///
    /// The last-active lane deliberately survives: it is a high-water mark of where the user has
    /// been working, and ⌘N after a deselect is exactly the case it exists to answer.
    public func clearSelection() {
        transient.clearSelection()
    }

    /// The lane a selection sits in, or `nil` when it names no single one — a live lane selects
    /// itself; live cards select their lane, but only when they all share one (a cross-lane
    /// selection has no single home to remember).
    nonisolated static func lane(holding ids: Set<ItemID>, in snapshot: BoardModel) -> ItemID? {
        guard !ids.isEmpty else { return nil }
        var found: ItemID?
        for lane in snapshot.lanes where !lane.isDeleted {
            let names = ids.contains(lane.id) || lane.cards.contains { !$0.isDeleted && ids.contains($0.id) }
            guard names else { continue }
            guard found == nil else { return nil }
            found = lane.id
        }
        return found
    }

    // MARK: - Quiescence

    /// Suspends until no reload is running and none is owed.
    ///
    /// The store is not a request/response object — a signal in does not produce a result out — so
    /// this is how a consumer (and every test below) says "let the pipeline settle" without polling.
    /// Returns immediately when the store is already quiet.
    public func awaitQuiescence() async {
        guard !isQuiescent else { return }
        await withCheckedContinuation { continuation in
            quiescenceWaiters.append(continuation)
        }
    }

    private var isQuiescent: Bool { !reloadInFlight && pendingReload == nil }

    private func resumeQuiescenceWaitersIfQuiet() {
        guard isQuiescent, !quiescenceWaiters.isEmpty else { return }
        let waiters = quiescenceWaiters
        quiescenceWaiters.removeAll()
        for waiter in waiters {
            waiter.resume()
        }
    }
}