Right-clicking a card now offers Labels: up to twelve rows ranked by how many cards on the
board carry each one, tie-broken by what this user reached for last, each a checkmark that
says whether *this* card already has it. More… opens the whole inventory beside the card, in
lookup order, with a field that mints a name the board has never used.
The rows deliberately do not widen to the selection the way Copy, Cut and Send to Trash do. A
checkmark is a claim about one card, and three cards where two carry `bug` have no honest
checked state — the rows that widen are the ones that say what they will do rather than what
is already true. It is also what keeps the menu cheap: a context menu's contents are rebuilt
on every ordinary pass of every card face, so reading the selection there would resubscribe
the whole board. The board-sized half of the ranking is cached on the store and gated on
equality; what runs per face is bounded by how many distinct labels exist, not by how many
cards do.
Copy Link leaves the first group for a new Copy Special submenu, mirroring Paste Special —
the placement the owner's latest layout asks for, and the one the row's own note has been
waiting on since it shipped as a same-day deviation. It keeps its VoiceOver action even so: it
is an action that changed doors, not a door.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
CardMoveTarget (BoardCommands.swift), LaneMoveTarget's cousin: validates
the clicked card's own current lane against the live lane order rather
than the board's live selection, refuses a target that reaches outside
that lane (no coherent left for a spread), and answers an index —
the clicked card's own position in its lane, clamped — for the
adjacent live lane in either direction. Trash is never a candidate
(not a Lane); a collapsed lane is a fine landing (a fold hides cards,
it doesn't close the lane).
CardFaceView's Navigation rows now call moveCardAcrossLane(by:), which
hands CardMoveTarget's answer straight to BoardStore.moveCards(_:toLane:at:)
— the exact call a released drag makes, so rank-minting, the undo step,
the watcher echo and the banner all come free. Targeting is Copy/Cut's
own widening (targetIDs): the clicked card, or the live selection when
the clicked card is a member of it.
Enablement stays render-safe the way isSelected/selectedCount already
are: three new CardFaceView parameters (hasLeftNeighbor, hasRightNeighbor,
selectionSpansLanes) are hoisted once per lane in LaneView.scrollableCards
and handed down as compared parameters, never read from inside a card
face's own .disabled. Caught and fixed a real regression here during
development: an early cut of the multi-lane-spread check answered true
for any lane that simply didn't contain the selected card, which
flipped a compared parameter for most of the board on an ordinary
single-card select and defeated CardFaceView's equality gate wholesale
(BoardRenderPerformanceTests.selectionStillRepaints caught it at 151 of
180 card faces).
Tests: CardMoveTargetTests (KeyboardGrammarTests.swift) pins the pure
predicate — leftmost/rightmost lane, single lane, index clamping, a
widened group anchoring on the clicked member rather than its own
extent, the multi-lane-spread refusal, and an integration test feeding
the answer straight through moveCards. CardFaceViewEquatableTests
gains a case pinning the three new compared parameters.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Raw image data on a card window's ⌘V was a keyboard shortcut with no visible
trigger; the sidebar's Attachments header now grows a quiet control — beside
the existing add affordance, present only while the pasteboard holds a
picture this card could take — that pastes it through the exact seam the
retired branch used (ClipboardStore.pasteImage(intoCard:in:), the board's
"Paste Image into Card" row's own call). The file-URL branch stays on ⌘V; a
Finder copy is still unambiguous. CardBodyTextView's paste-yield mechanism
needed no change at all — it forwards by capability, not by picture-specific
logic, so a screenshot ⌘V with the body editor focused is now a genuine
no-op there, served by the new control instead.
The pasteboard's re-read gains a fourth checkpoint — a window becoming key —
alongside menu-tracking, ⌘-down and app activation: a persistent visible
control has to read true continuously while its window is frontmost, not
only at the instant a menu or chord probes it.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
CardFaceView.boardMenu/boardActions restructured to the owner's shape (card
fe66c461): Open/Copy Link/Rename/Style▸(Symbol,Color), a divider, then
Copy/Cut/Paste/Paste Special▸(Paste Image into Card), a divider, then
Navigation▸(Move Left,Move Right), a divider, then Send to Trash. Every row
routes through existing machinery — no new commands.
ClipboardStore gains copy(from:targeting:)/cut(from:targeting:) and their
canCopy/canCut twins, so Copy and Cut can widen to the clicked card exactly
as Delete and Style already do ("right-clicking something outside the
selection acts on what was clicked"), without disturbing the Edit-menu path.
LaneMoveTarget.destination is extracted out of MoveLaneCommands so the
card menu's Navigation rows validate against the identical sole-live-lane
predicate as Board ▸ Move Left/Right. Since a card id can never itself
satisfy that predicate, the two rows are wired to the real store call but
unconditionally disabled — reading the live selection per card face would
reproduce the O(board) render regression isSelected/selectedCount exist to
prevent (contextMenu's builder is not lazy).
Style ▸ Symbol and ▸ Color both open the one existing style popover — no
per-section pre-focus (StyleEditorSession has no such concept, and
StyleEditorView internals are out of scope while another pass redesigns
the pickers). Paste and Paste Image into Card reduce their .disabled
checks to selection/snapshot-free forms, proven safe by construction (a
rendered card face already guarantees a live lane / a live board card).
Journaled on the card: Copy Link kept (shipped same day, not in the
owner's list), "Delete" relabeled "Send to Trash" (board-side move, not
the permanent trash delete), quick-style recents row dropped from this
menu, Navigation's always-disabled rows, and Paste not retargeting to the
clicked card — all flagged needs owner review. DESIGN/11-command-nexus.md's
Card row is owed a rewrite, left for the main session.
Tests: LaneMoveTarget.destination (new), targeted copy/cut (new), plus
existing ClipboardStore/PasteTarget/MoveLane/CopyLink/Trash-menu/PasteImage/
PasteFile/Style/CaretChord/render-performance/equatable-gate suites —
156 tests, all passing.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Follow-up to d0c5461's .backgrounds/ folder: generating, pasting, or
choosing solid now trims the app's own prior file in .backgrounds/ as
part of the same write when it repoints or unsets background.image
away from it — silent, best-effort, never blocking the gesture that
triggered it. A scheduled heal at every board open sweeps whatever
that trim declined or missed: any .backgrounds/ file the board's
current background.image no longer names, announced with a loss-row
notice in the loose-file relocation's own voice. Legacy root-level
references are untouched by both paths — the tidy scopes to
.backgrounds/ only, since that is the one folder the app can prove it
wrote into.
Removal is via FileManager.trashItem, matching the attachment-removal
precedent (recoverable, never a hard delete).
Supersedes applySolidBackground's earlier "facets.png survives on
disk" contract for the settled case: the ruling reads "unsets" as one
more shape of "repoints away from a .backgrounds/ file", so a settled
solid choice now trims the generated picture it displaces, same as
switching producers does. The superseded test and doc comment are
updated to the new behavior; the echo-window (no-reload-yet) case is
unaffected and still leaves the file in place.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Added .onHover modifier to the collapseButton to explicitly set NSCursor.arrow when hovering over the collapse chevron, preventing the lane resize handle's resize cursor from appearing over the button.
The image branch's precedence ladder always had a second clause: a file URL on the
pasteboard suppresses it, "a different gesture with a different answer" that the
code deliberately declined rather than guessed at. This fills it in: one or more
file URLs paste through the same `importAttachments` a Finder drop takes — one
collision ladder, one folder refusal (`FinderDrop.partition`), one set of banners
— outranking raw image data riding beside it (a Finder-copied image file carries
both; the actual file lands, not a re-encoded copy of its bytes) while still
deferring to the app's own clipboard type. Both ⌘V surfaces read the same
`ClipboardStore.fileURLPayload`, so the board's fallback and the card window's own
branch stay in step by construction rather than by two hand-kept-in-sync checks.
Fixed a real leak in the body editor's paste yield along the way: `public.file-url`
conforms to `public.url`, which `NSTextView` legitimately reads for a pasted
hyperlink, and `NSPasteboard.availableType(from:)` matches by conformance rather
than exact type — so a Finder copy carrying a generic URL representation beside
its file URL would have been silently swallowed as text and never reached the
window's attachment branch at all. The yield now declines outright on any
file-URL pasteboard before the generic capability check runs.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
New app-written background images (Theme tab ▸ Pattern, and Edit ▸ Paste
as Board Background) now land in a hidden `.backgrounds/` folder at
board root instead of beside index.md, matching the `.trash/` app-managed
pattern. `background.image` stores the qualified relative reference
(`.backgrounds/facets.png`); the resolver needed no change at all, since
it already accepted any relative path inside the board root — the same
mechanism that already resolved `art/backdrops/sunset.png` resolves the
new location for free. `BoardWriter.writeBoardImage` now creates its
destination folder if missing, since `.backgrounds/` won't exist until a
board's first generated or pasted background.
The Finder collision-ladder (`BoardStore.boardImageName`) is rescoped to
`.backgrounds/`'s own contents, and its overwrite-in-place check now
recognizes only the qualified form as "ours" — a legacy bare
`background.image: facets.png` from before this change is read as a
foreign reference rather than migrated, so a regeneration writes a fresh
`.backgrounds/` file and orphans the old one in place, per the no-migration
ruling. The Theme tab's Pattern/Solid mode-detection was updated to
recognize both the legacy and current spellings as the generator's own
output.
The board loader needed no change: `.backgrounds/` is a hidden,
non-UUID-shaped name, and `.skipsHiddenFiles` already keeps every hidden
entry off the lane walk before any name-based exclusion is consulted —
pinned with a new loader test. Deliberately did not add `.backgrounds` to
IntegrityRules' claimed-name/squatter-displacement table: that table
mirrors a specific existing DESIGN ruling this card doesn't amend.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
⌘V grows an image-data branch, below the app's own clipboard format and refused
outright while a file URL is on the pasteboard: a screenshot or a browser's Copy
Image lands as "Pasted Image.png" in the anchor card's attachments/, through the
very import path Finder file drops and ⇧⌘A take — one bracket, one Finder-style
collision ladder, one set of banners, and the same silence a drop's arrival has.
A card window's ⌘V pastes onto its own card; a focused text field still wins the
selector natively. A file-shaped flavor travels byte for byte, PNG preferred when
several are offered; TIFF and BMP are re-encoded to PNG, being interchange
encodings rather than files anyone wants in a folder.
The hero key gets the setter it was born owing: "Set as Hero" on any image row of
the attachment list, "Remove Hero" on the row that holds it, with menu-bar twins
so the context entry is nobody's only home. It writes as a restyle — one key, one
bracket, one invertible step on the window's own stack — and replaces rather than
refusing, because a card has one hero and the row that has it says Remove instead.
Edit ▸ Paste as Board Background is the same payload's other destination, taking
the existing background.image convention at its word: the picture into the board
folder as "Pasted Background.png", the colour subkey untouched, the generator's
overwrite-our-own-name rule inherited and its echo memo taught to tell the two
producers apart.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
`CuratedSymbols` was one flat ~65-glyph list serving every style-editor target alike. It is
now three level-specific sets — `boards` (project/container/identity), `lanes`
(workflow/stage/status), `cards` (work-item/content) — each ~30-40 entries, seeded from the
original list and `SymbolPickerCatalog.defaultSet`, reorganized by which level a glyph actually
reads as being about. Overlap is kept where a glyph genuinely fits everywhere (`flag`, `star`).
Wiring:
- `BoardInfoPopover`'s board-glyph `SymbolPicker` now passes `CuratedSymbols.availableBoards`
instead of the picker's domain-agnostic default.
- The card sidebar's `SymbolPicker` (`CardSidebarSections`) now passes
`CuratedSymbols.availableCards` instead of the old flat `available`.
- The style editor's own curated grid (`StyleEditorView`, the Style… popover's only remaining
anchor) reads `CuratedSymbols.availableForStyleEditor(level:spansLevels:)`: a homogeneous
target reads its own level's set, and a target that somehow spans more than one level (today
unreachable — 04-interactions.md's cards-XOR-lanes rule keeps a live selection homogeneous)
reads the three combined, via a new `BoardStore.styleTargetSpansLevels` seam that asks the
question `styleLevel(of:)` deliberately collapses.
- `CuratedSymbols.combined` (the three sets' stable-order union) also replaces the old `.all`
in `SymbolPickerCatalog`'s full-catalog fallback.
- `SymbolPickerCatalog.defaultSet` is kept as the fallback for a caller naming no level (a
future saved-search picker, say) rather than retired.
DESIGN/03-board-ui.md and DESIGN/05-card-window.md's Styling/sidebar prose amended minimally
where they named "the curated set" as a single list.
Tests: three new/rewritten suites in KanbanTests/StyleModelTests.swift (set shape, availability,
overlap, `combined`, the style-editor level/span decision, the board-anchor width tripwire), one
new test in KanbanTests/StyleWriteTests.swift (`styleTargetSpansLevels`), and the old
single-list-pinning tests in KanbanTests/SymbolPickerTests.swift and KanbanTests/CardSidebarTests.swift
updated to the new set names. 2834 tests, 487 suites green (KanbanTests, arm64); one unrelated
flaky failure (RootRecoveryTests.vanishAndReturn under full-suite load) passed clean in isolation.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
A card whose thread holds one comment or more now draws a second trailing chip beside the
paperclip: a secondary-tinted bubble glyph plus its count, shown only when the count is above
zero (design ruling 2026-08-09, card e729e30a). Same styling family as the attachments chip —
caption size, secondary tint, decorative and hidden outright from the accessibility tree — but
this one carries a visible count rather than staying icon-only, per the ruling's own "bubble-style
SF Symbol + count." It sits after the attachments chip at the row's trailing edge, in both the
live title row and the drag replica.
The count is a new `Card.commentCount` field the loader fills with a readdir over `comments/`'s
identity-shaped children that carry their own `index.md` — `BoardLoader.commentCount(in:)`, built
on the same `identityShapedChildren` predicate a trash entry's held-card count already uses. Never
a parse: `.draft` and `.trash/` are excluded for free, the same dot-prefixed hidden-entry skip
`CommentThread.load` documents for both, so the walk stays exactly the O(cards) shape
01-storage-format.md § Enhanced schema already commits to. Because the count rides inside the
`card: Card` parameter `CardFaceView` already takes — not a new parameter of its own — drawing the
chip costs nothing beyond a field read on an already-compared value: no new Observable read joins
the body, and the equatable gate already covers it via `Card`'s synthesized `Equatable`.
The one divergence from the comments pane's parsed count is documented rather than hidden: a
comment folder whose `index.md` exists but fails to parse is a `Stray` the thread read excludes by
opening and rejecting it, a cost this readdir does not pay. The face may then read one comment
high until that folder is fixed or removed — the trade the ruling's "cheap directory-entry count…
not a parse" asks for, over paying full parse cost on every card of every load. Every well-formed
comment, and every card with no malformed one, agrees with the pane exactly.
VoiceOver: `AccessibilityPhrases.cardValue` gains a `comments: Int` parameter, appended after
attachments and before the cut-pending phrase — the same left-to-right order the two chips draw
in, so a sighted read and a VoiceOver read never disagree about which comes first. The trashed
lane row's own call site (an opaque unit with no comments to speak of) passes `comments: 0`.
Docs: DESIGN/03-board-ui.md's card-face section describes both chips and retires the stale "closed
with no growth" sentence, honestly recording the 2026-08-09 growth (the hero banner landed hours
earlier, this chip after it) as exposure of facts the card already carries rather than a body
excerpt. DESIGN/10-accessibility.md's flattened-element sentence gains the comment count.
DESIGN/01-storage-format.md's Enhanced schema paragraph records the chip as shipped. WISHLIST #9
is marked shipped in place — not renumbered, since #10 and #11 are cross-referenced elsewhere.
Tests: CardCommentCountListingTests (BoardLoaderTests.swift) pins the readdir against a synthetic
tree — no comments/ folder, an empty one, non-identity-shaped and index-less strays excluded,
.draft/.trash/ excluded for free, agreement with CommentThread.load's parsed count in the
well-formed case, and the one documented divergence on a malformed index.md.
AccessibilityPhrasesTests covers cardValue's new parameter alone, alongside attachments, and
all three fragments together. ViewEquatableTests pins that a comment landing on a card is a gate
difference. BoardRenderPerformanceTests adds a render-cost guard: one comment added to one card
on a hosted 180-card board re-renders a handful of bodies, not the board.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
A card whose `hero:` names one of its own attachments draws that picture as a
banner across the full width of its plate, above the icon-and-title row,
aspect-fill cropped into a fixed 2.75 em band — 36pt at the standard body, and
em-scaled like every other figure the board draws, so it grows with the system
text size and with the board's zoom rather than shrinking against a title twice
its usual size. The figure sits deliberately under the 44pt a plain one-line
card is tall: a hero card should read as a card with a picture on it rather than
a picture with a caption, which is 03's standing rule that the title dominates.
The key's grammar is a **bare filename**, and that is what separates it from the
board background's `image` subkey rather than a nervousness about paths. A board
names a file anywhere under its root, so a path is that key's reading and where
it leads is the renderer's question. A card names one of the files it already
owns — the flat `attachments/` folder the app lists, relocates into, and carries
through every move, copy, trash and restore — so `hero: art/sketch.png` is not an
awkward spelling of a hero image, it is a value the key cannot mean. It therefore
has no reading at all: a value carrying a separator, or spelling `.`/`..`, or
empty, is malformed at the document layer, which renders it as absent and leaves
the coerce tier's trace, exactly as `width: 1.5` does. The bytes stay as written,
the resolver re-checks containment anyway, and the whole degrade family below
that — a name pointing at a missing file, an unreadable one, or one that is not
an image — ends the same way: no banner, no defect, nothing written.
That last promise is about *height* as much as about ink, so the band is given no
height at all until a picture has actually decoded. A card whose hero cannot be
drawn lays out identically to a card with no key, structurally rather than by a
branch somebody has to remember; the price is one settle per hero as a board
opens, and none after that. Everything else the face draws is attached outside
the new stack and is untouched by it — the accent stripe still runs the plate's
full leading edge across the band's corner, the selection and file-hover strokes
still ring the whole plate, the cut and drag dims still cover it, and the drop
model still registers the plate's real height, so a hero card is simply a taller
card the masonry already understands. The trash draws it too, by the one-face
rule.
Decoding is ImageIO's downsampling path off the main actor at a quarter of the
backdrop's pixel budget (`BoardBackdrop.decode` gained the limit as a parameter
rather than being copied), and the results live in one app-wide, deliberately
non-observable cache keyed on path plus the file's date and size. Non-observable
because a tracked write there would invalidate every hero face on the board,
which is the O(board) invalidation this view was rebuilt once already to shed;
each face holds its own picture in view state and seeds it from the cache, which
is also what lets the drag replica — whose preview builder is non-escaping and
cannot await anything — carry the band at the face's real height. Taking a stamp
twice from one URL value turned out to answer with the first read's date and size
however many times the bytes had changed, so `stamp(of:)` now drops its cached
resource values first; noticing a replacement is the only thing a stamp is for.
The face takes the resolved URL as a compared input rather than resolving it, for
selected-ness's reason one axis over: resolving needs the card's folder, which a
face does not know, and finding it from the snapshot would be a board walk per
face. The lane and the trash column each know their own container and compute it
once for the whole strip.
There is no in-app setter this version — the key is written by hand or by an
agent, which is why the guide bumps to v13 with a clause spelling the grammar out
beside the other card keys, and why `attachments/` gets the one-line pointer an
agent that has just written `` will need. "Set as Hero"
from the attachment row is future work, as is the card window and print, which
draw the same model and show no banner today.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Lateral card navigation was pure geometry: the nearest drawn frame in the
direction. That loses the walk in the card's own title — stepping from a
10-card lane's 8th card into a 3-card lane clamps to its 3rd, and coming
back out, "the nearest frame at that height" is the 3rd card's height. The
information the user was walking at stopped being on screen, so no rule
over rectangles could have recovered it.
So it is remembered instead. `TransientBoardState.lateralOrdinal` holds the
1-based position a run of ←/→ started from, counted over the cards the board
is showing, and `NavigationMath.lateralHop` lands each hop on
`min(ordinal, target lane's count)` of the next lane that is showing cards —
collapsed and query-emptied lanes hopped over on `firstCard`'s rule rather
than by the accident of registering no frames. 8th → 3rd → 8th.
Every reset comes from one funnel and needs no enumeration anywhere: the
ordinal is a defaulted `nil` parameter on `select`, so a click, a marquee, a
↑/↓ step, an ⌥-jump, a ⌫ successor, a lane-domain arrow and the reload's
focus recovery all end the run by saying nothing. `resolve` adds the one
rule a value referencing no item can need — the ordinal never outlives the
head it was counted from — while a reload that leaves the cursor standing
leaves the run standing too.
A wide lane's interior masonry columns keep their spatial step and carry the
ordinal through untouched: a column hop is not a lane hop, and stickiness is
lane-granular over the logical order. With no lane in the direction the
geometry has the last word, which is how → still reaches the shown trash.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
The lane header no longer draws a new-card button. Every other
creation path is untouched: File > New Card (Cmd-N), Return on a
selected lane, and double-click on lane empty space.
- LaneView: drop the button and its overlay slot, leaving the
collapse chevron as the header's one piece of trailing chrome.
- NewCardTarget: drop the button's click-names-target carve-out over
the Cmd-N target rule (Return on a selected lane is now the only
direct-target path left).
- AccessibilityPhrases: drop the button's spoken label.
- BoardMetrics: drop its width reserve; laneHeaderTrailingReserve now
reduces to the collapse chevron's own reserve.
- DESIGN/03, 04, 10, 11: update the lane title-bar inventory, the
Cmd-N override clause and search-clearing mechanism list, the
accessibility tree-shape sentence, and the pointer-grammar row that
described the button.
- Tests: drop the two pinned accessibility-label tests and the
newCardButtonReserve assertions; retarget the truncation-headroom
test at the now-solo laneHeaderTrailingReserve; fix stale doc-quote
comments and the manual accessibility-verification checklist.
Full KanbanTests suite: 2807 tests, 482 suites, all passing.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Card 737a949f: "Add an option to card context menu to copy a link to
the card folder." Implements the design ruling verbatim.
- New context-menu row "Copy Link" (CardFaceView.boardMenu, board
side only — trash cards are excluded, matching "sole selected live
card"). Writes the clicked card's folder in one pasteboard item
carrying two representations: the file:// URL under .fileURL and
the plain absolute path under .string (FolderLinkPasteboard.swift).
Enabled on a sole selected live card; disabled on a multi-selection
and wherever edit-shaped actions already disable, per the ruling.
Also exposed as a VoiceOver custom action alongside its siblings.
- Menu-bar twin: Board ▸ Copy Link (BoardCommands.swift,
CopyLinkCommand), no default chord — the every-function-a-menu-item
contract in DESIGN/11-command-nexus.md is still current, so this is
the twin that contract calls for, homed the way Open Card/Rename/
Style… already are.
- DESIGN/11-command-nexus.md: new Board-menu row and an updated Card
context-menu row.
- Tests (KanbanTests/CopyLinkTests.swift): the target predicate's
enablement (sole card / multi-selection / lane / trash / empty /
inline-editing), the pasteboard write's exact bytes via a fake
pasteboard (both representations, exact folder URL), and a
disabled-target no-op.
Caught and fixed during self-review: an early version read the
context menu's widened-selection helper (targetIDs) inside the
Copy Link row's .disabled(...), which reads store.selection. Since
.contextMenu's content closure is evaluated on every ordinary body
pass (not only when the menu opens), that resubscribed every card
face on the board to every selection change — the exact O(board)
regression RENDER-INSTRUMENTATION.md's isSelected/selectedCount
split exists to prevent, caught by BoardRenderPerformanceTests and
MarqueeRenderCostTests. Fixed by reading the already-hoisted,
non-Observable `selectedCount` parameter instead, which answers the
same "how many ride along" question at zero extra subscription cost.
xcodegen generate, the Kanban scheme build, and the full KanbanTests
suite (2816 tests) are clean.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
A lane folds to a fixed slim vertical strip carrying its glyph, its card-count
badge and its title turned on its side, and the strip is deliberately not part
of the window's division: the expanded lanes' units divide what is left once
each folded strip's fixed width has come off the top, so folding a lane is a
re-divide trigger of the Show/Hide Trash family — the window never moves and
the siblings grow into what the lane gave up.
The state is a first-class lane frontmatter key, `collapsed: true`, and
document state exactly as `width` is: the files are the board, so an agent
folds a lane by writing one key. Absent means expanded, expanding removes the
key rather than writing `false` (the remove-at-default family beside a
one-unit `width`, the empty rename's `title` and the None well's
`background`), and the lane's `width` rides along untouched so expanding
restores the lane the user had. The read is `width`'s leniency one type over —
a boolean scalar or a quoted boolean word reads as itself, everything else has
no reading at all and renders as expanded, bytes preserved either way.
Toggling is the header's always-visible collapse chevron, the lane context
menu's single Collapse Lane / Expand Lane row, and a plain click anywhere on
the strip; a modified click on the strip stays the ordinary selection grammar,
so a folded lane is still selectable by pointer. The title reads bottom-up and
is justified to the top of the room below the strip's chrome (owner ruling
2026-08-08), truncating against the strip's own height.
While folded the lane draws no cards at all, which is what makes every
exclusion true by construction rather than by a guard per gesture: no card
face means no marquee target and no navigation frame, and no registered grid
means the masonry's drop zones have nothing to resolve against. What did need
code is the half that names absolute destinations — the option-arrow jumps and
the arrow seed scan past a folded lane, the lane domain's down-arrow is inert
on one, and New Card skips it (a selection inside one falls through to the
last-active lane, the stale selection's rule). A drop on the strip appends at
the lane's end, cards and Finder files alike, with an accent edge standing in
for the shadow the strip has no masonry to open; there is no hover-to-auto-
expand yet. Lane reorder works on the strip, and a dragged folded lane carries
its fold, so its shadow and its replica are the strip rather than its units.
The write is `writeLaneWidths` clause for clause — one `updateIndex` bracket,
the same stamp behaviour, the same three do-nothing paths — with two new
`WriteOperation` cases and two new undo verbs rather than one of each, because
a banner or an Edit-menu row that said "resize" after Collapse Lane would name
a control the user never touched.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
The right-edge drag's growth was capped at the screen's visible frame,
because each snap tick grows the window; on a window near the screen edge
that left a lane stuck at a tick or two of headroom. Settled 2026-08-08
(03-board-ui.md § Lane, superseding the pathfinder's hard stop): at the
screen the window pins and each further tick re-divides the fixed strip
width across one more unit — siblings compress, the stepper's mechanism
arriving under the drag's fingers. The regimes meet with no pixel jump
(the re-divided standard at the fit IS the frozen standard, by the
exact-fill identity), shrinking mirrors the way back, the rubber band
moves to the strip's own capacity, and a window with no headroom at all —
full screen included — re-divides from the very first snap.
New pure arithmetic in LaneLayoutMath (pinnedStripWidth, resizeStandard,
resizeMaxUnits, resizeWindowDelta, snappedUnits over per-count slots);
LaneResizeSession splits the tick across the regimes and derives its
standard from the live count; the handle and BoardView hand the session
the strip's whole divide. 2709 unit tests green (+11).
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Edit ▸ Undo/Redo become the app's own replaced rows and the board toolbar
pair takes explicit targets, both reading the focused session's
BoardUndoManager through FocusedValues.undoStack (board windows publish the
session's manager, card windows their own) — the nil-target route died with
the SwiftUI window latch, 13-native-undo.md ▸ Rules ▸ command surface,
re-ruled 2026-08-08. The rows enact the routing predicate themselves: text
focus routes ⌘Z to the first responder's own manager, title and enablement
included, re-derived at fire time with a beep for the stale window.
NativeHistoryProvider turns @Observable so both surfaces re-derive on stack
changes; a checkpoint-notification ticker covers plain text managers.
.responderAction leaves ToolbarItemSpec with its only user;
windowWillReturnUndoManager stays wired for AppKit's own asks.
Live-probed on the fixture board (21/21): the row retitles to "Undo Add
Lane" and crosses via real ⌘Z key events, ⇧⌘Z redoes via a window-server
chord, the toolbar pair validates and fires, search-field and body-editor
⌘Z stay text undo with board stacks untouched, and a card window crosses
its own stack with no fall-through. 2698 unit tests green.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
The excision's follow-up sweep: present-tense prose that still implied an edition axis now reads
correctly under one version ("in every tier" clauses dropped or turned substrate-shaped, the
announcer's "every free-tier bracket today" is "every bracket today" — nothing passes a phrase),
and forward-looking promises pinned to the mooted pro-m1/pro-m2 milestones now name the thing
itself (the change narrator in Kanban/Changes/, the foreign-change journal successor) or fall to
past tense. Kanban.entitlements' network-client comment sheds its "dormant until Pro ships"
framing for the pivot's own reasoning: the key stays because the sync capability to come needs
it regardless. Untouched on purpose: the storage layer's coerce/tolerate/refuse tiers, the
chooser's bundled/user/keyless tiers, verbatim design-doc quotations, and genuine past-tense
record. One dangling reference repaired en route: EndToEndVerification.md cited the long-renamed
InertGitTests. Comment-only throughout; 2,686 unit tests green, unchanged.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Step 7 of strategy/01-git-excision.md, the companions. The agent guide bumps to v11: the Git section teaches repo-resident etiquette alone (stage only your own paths, commit your own changes, leave app-maintained files to the app) — existing boards heal to the new text on next open. README re-anchors: the four git feature bullets out, tiers say the complete Mac experience is free, and one bullet states the format's git-friendliness promise. The changelog drops the never-shipped git entries. DESIGN re-rules: 06 retired with Undo routing migrated to 13 (now the sole substrate's doc, seam kept open), 07 retired as written pending the ops-service workstream, 14 retired as superseded record, 12 carries the second pivot note, the index reflects all of it; the charter gets a pointer note (the anchors' full re-ruling stays with the user). InertGitTests renames to GitAgnosticStorageTests — the excision restores its original claim app-wide. And the sweep: ~70 comment sites across 36 files adjudicated against the keeper list, every present-tense description of the excised machinery made past tense or repointed, keepers untouched. 2,707 tests green.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Step 3 of strategy/01-git-excision.md: the popover strip is Info/Theme/Sync, the titlebar widget says the name alone, the card window's History section and its slot go, Board ▸ Pull/Push comes out with the RemoteCommands scaffold, and View ▸ History re-tags from the commit trail to the deferred foreign-change journal. The Sync placeholder re-annotates to the future ops-based sync service. Four git test suites leave with the surfaces they pinned (BoardGitSetup, BoardInfoPopover, BranchSwitch, GitUndo). The git engine still compiles underneath, unreferenced by UI. 2,890 tests green.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
BoardInfoTab gains .sync last in the strip, rendering BoardSyncTabView's one
honest caption. The tab claims the position where 07's remote half will
live without ruling where its setup surfaces land — that stays the open
Redesign card's question.
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
The titlebar widget becomes a two-line identity block: the board glyph at
22pt spanning both lines, the title over the branch (git-mode only, smaller
and secondary), the em-dash retired. New Branch… returns to the switch menu
behind a divider, revealing an inline name field — the pre-split shape. The
board settings sheet retires whole: add-git and commit identity render
inline in the Git tab's postures (BoardGitSetup.swift), the availability
rule collapses into BoardGitSetupSection.resolve, and Board ▸ Board
Settings… leaves the menu bar. Where 07's remote/credential setup surfaces
land is deliberately left open — filed on the Redesign board.
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
HistoryStore.compose(boardRoot📒) returns non-optional and runs for
every session — the nil the gate produced was the only nil it ever had.
makeHistoryProvider is a one-axis decision: git-mode boards bind the git
provider, everything else native, in every tier; Session.tier stays
recorded, dormant. BoardGitSection shrinks to the four mode postures
(.absent and .proPointer die, BoardGitNote and the .git probe with them);
every board carries all three popover tabs (BoardInfoTab.available
retired); the titlebar branch shows on any git-mode board; the settings
sheet and card History section stop reading tier. InertGitTests is
repurposed as UntouchedGitTests — the file layer still never opens .git,
now load-bearing for mode-none boards. The accessibility audit reaches the
settings sheet at last: the fixture board hosts it in every tier, so the
free-fixture disabled-row test becomes an open-and-audit test.
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
With branch creation relocated to the settings sheet, a single-branch
board's switch menu opened onto nothing and read as broken (ruled
2026-08-06, built with the Git tab): a bare Text — AppKit's standard
disabled item, read by VoiceOver as disabled text — now says "No other
branches" where the switch entries would be. The filter behind it becomes
the pure switchTargets(branches:current:) seam, pinned by
BranchSwitchTargetTests: current excluded, nil current passes all through,
repository order preserved.
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
The popover's third tab settles (03 ▸ Git tab, 2026-08-07): BoardGitTabView
takes the parked section — the six-posture matrix, the Pro pointer, the
nested/unverifiable prose, the branch controls, the Board Settings… door —
with the tab label doing the old "Git" header's naming work, and one new
caption for the no-repository posture. BoardInfoTab.available drops the Git
tab exactly where BoardGitSection resolves absent, carrying 12's
absent-no-placeholder rule up to the tab strip; selection resets to Info per
open, now a ruling. Daily face only — the remote block still joins with
pro-m2.
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
iconColor stops being hand-written-only (user-ruled, superseding 03's
"schema yes, control no"): it rides StyleCommand.apply → applyStyle as
the third styled dimension — per-dimension no-op skip, one bracket, one
history step, ExpectedField.iconColor for staleness. The SymbolPicker
grows an opt-in colour row (leading None plus seven Palette.foregrounds
hues, None removes the key); the board popover is its one caller. The
window-title widget now draws the board's resolved glyph in that tint
beside the name. Doc realignment filed on the Redesign board (Minor).
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
A selection change re-ran every CardFaceView on the board (180 bodies ≈ 85 ms on
the 6×30 fixture, 515 ≈ 233 ms on a real 515-card board, debug): the face's body
read store.selection in three places — isSelected, the drag replica's count, and
the context menu's styleTarget — and Observation invalidates every reader of the
property, past the equatable gate entirely. The band overlay stayed cheap, which
is why the marquee tracked the cursor while the highlight lagged ~0.4 s behind.
Now LaneView and TrashLaneView hoist one selection read per body and hand each
face isSelected/selectedCount as compared parameters; StyleMenuItems takes its
target as a deferred closure; TrashLaneRowView gains the same treatment plus the
Equatable gate it never needed before. Select-one-card: 180 bodies → 1. A
growing band costs the selection's own running size; the real board's crossing
fell 233 → 112 ms — the remainder is lane bodies re-measuring their masonry, a
separate lane-level finding recorded in RENDER-INSTRUMENTATION.md.
Also: select() gains defaultsSoleMember — the marquee's explicit nils never
avoided the sole-member default, so a one-card band acquired a selectionHead and
could scroll the lane out from under its own drag.
MarqueeRenderCostTests pins the shape: redundant samples cost zero bodies, a
growing band pays per crossing, and selectionStillRepaints holds a ≤8 budget.
Model facts off the live snapshot (Lanes, Cards with the trash's freight as a
quiet tail, Attachments — the welcome-count live-only rule) and disk facts off
one background whole-folder walk (File, Files, Size, Created, Modified — .git
and .trash included, so the rows agree with Finder's Get Info), with a Reveal
in Finder link as the door to the folder the rows describe.
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
The symbol/name header keeps the top; below it a segmented tab bar replaces
the styling embed and the git section, both parked in-file — their seams are
settled, test-pinned design, and the tab sessions rehome surfaces, not rulings.
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
Measured on real events 2026-08-07, correcting the 2026-08-06 hosted
finding: a bare count-1 tap on LaneView's empty-space layer fires in
~1-3 ms with no drag source at all — the hold that made the empty
.onDrag look necessary was the sterile NSApp.postEvent stream
over-disambiguating. And the provider was actively harmful: even an
empty drag source claims the mouse-drag at threshold, starving the
marquee's simultaneous DragGesture after one sample — the band froze
and the mouseUp never arrived. The layer goes dragless; drags from
empty space belong wholly to MarqueeControl. PointerClick's and the
layer's comments retell the corrected story. Alongside: openCard is
typed @MainActor throughout, which makes the closure Sendable and
lets CardFaceRole carry it under CardFaceView's nonisolated ==.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
View ▸ Appearance (11-command-nexus.md): three radio-exclusive rows,
app-wide, persisted, needing no window in front — the View menu's new
last group. AppearanceStore owns the override's rules (absent key =
Auto, lenient reads degrade to Auto, remove-at-default) with an
injectable apply seam so test hosts never touch NSApp; the one real
apply hands NSApp.appearance its answer in applicationDidFinishLaunching,
the global side effect KanbanApp.init must not carry. The board toolbar
gains its first .picker item — an NSMenuToolbarItem whose rows re-fetch
their spec fresh, checkmark read at menu-open like every other menu row —
and Appearance joins the search field as the second default item,
centered beside it (03-board-ui.md ▸ Toolbar, ratified 2026-08-07).
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
The whole-lane wash-and-solid-ring scaled its emphasis with the lane's
area: what read as a ring on a card read as a slab on a column. Resettled
as the system focus ring's vocabulary at selection strength — a hairline
accent edge with a soft accent glow bleeding outward, riding the stroke
shape rather than the translucent plate (a shadow behind it would bleed
through as murk), and no wash at all: nothing tints the cards' backdrop,
and lane colour stays the accent band's alone. Under Increase Contrast the
halo yields to the solid full-alpha ring — "strengthens borders" means
crisper, and a blur is the one thing a border cannot become — so the
setting sees exactly the indicator it always has; cards and trash rows are
card-scale and keep the solid ring at every contrast. One selection
vocabulary, pinned by Accommodations.drawsSelectionHalo and its test.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
"At drag start it replaces the item's original space" is now true from the
very first frame: begin takes the run's own resting slot as a seed, set in
the same transaction that lifts the run out. Without it the vacated gap
closed un-animated and sprang back open at the first dropUpdated — a
shuffle carrying no information. The seed bypasses propose deliberately (a
pickup is not a new landing spot, so no alignment tick), and the first real
sample's re-propose of the same slot is the early-out's ordinary silence. A
⌥-pickup seeds nothing — a copy's resting layout keeps the originals in
place, so there is no vacated space to hold. Card pickups compute the seed
through the same LaneView.rendered rules the retargets count in; lane
pickups seed their strip slot the same way. DRAG-REORDER.md's own-slot
paragraph records the rule, and SeededPickupTests pins all four corners.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
The measured defect: LaneView's empty-space double-click was a second
sequential .onTapGesture(count: 2) stacked over the single tap, and that
recogniser held every click in the lane — its own empty space and every
card face alike — hostage to the system double-click interval while it
disambiguated (~475 ms click-to-selection on a hosted board). The fix is
structural: the empty-space surfaces live on a background layer behind the
masonry, so a card's click never shares a gesture path with a lane
recogniser; one .onTapGesture branches on PointerClick.count (AppKit's own
clickCount, read the way ClickModifier reads the keyboard) — first click
selects, second creates, Finder's cadence with nothing to disambiguate; and
the layer carries a load-bearing empty .onDrag, because without a drag
source macOS holds primary clicks pending multi-click disambiguation
(measured: never fires alone, ~90 ms with one present). A measured viewport
floor makes each lane's blank space actually belong to the layer — a
ScrollView proposes nothing along its scroll axis, so only an explicit
minimum stretches the content — with the trash column as its twin, less the
padding that sits inside its scroll content. The template chooser's stacked
pair collapses to the same one-recogniser branch, and the attachment rows
move their double-click to a simultaneous gesture (instant there, because
the row's real .onDrag forces immediate delivery). PointerLatencyTests pins
the recovery with synthetic pointer events on a hosted board.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
View ▸ Zoom In / Zoom Out / Actual Size (⌘+ / ⌘− / ⌘0): 75%–200% in eight
rungs, app-wide and persisted (the Show Comments precedent) — a viewing
comfort, not a property of any one board. The level travels as
BoardZoomContext in the environment, injected on BoardView alone so the
banner strip, search bar, sheets and popovers stay at the system size; the
environment is also what carries it through CardFaceView's equality gate,
which compares nothing that moves with the level. Every BoardMetrics figure
follows zoom.bodyPointSize — card and lane chrome, drag replicas and the
count badge, the resize handle, the trash column — and the drop registry
carries the ruler for event-time reads, with the autoscroller's three
reaches turning font-derived (reachSide named as the stripGap it always
equalled). Lanes still divide the window; zoom never moves the window or
its floor. The toolbar gains a catalog-only Zoom In/Out pair mirroring the
menu rows' predicate; zoom holds shut mid-drag (frozen geometry), each rung
announces itself to VoiceOver, and the render suite pins both invariants:
a rung repaints every face, a no-op Actual Size repaints nothing.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
background is {color:, image:} and only a mapping at every level; the board's image paints the full window under a transparent title bar, with a thin-material frost strip keeping the chrome legible and the standard accommodations intact.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Redesign card 6d5a97fd resolved off the 2026-08-01 live try-out: leading is
board identity, center is view controls, trailing stays the user's catalog.
centeredItemIdentifiers, not a flexible-space sandwich — centers against the
window, holds as catalog items install, reaches machines with a saved
arrangement; defaultItems and its pinned tests untouched. 03 records the
grammar and the accepted tensions (widget-vs-field budget in narrow windows,
HIG's trailing convention yielding).
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
Four rulings close Redesign Contradiction 3452893f (2026-08-06): the in-app
escape hatch is ratified in full, reversing 2026-07-29's palette-only rule —
the combo's Other… opens the system Colors panel, a pick landing on a palette
color stores the name, anything else the hex. Free-picked colors change no
contrast story: they land on the same runtime ink computation hand-written hex
always got (10 amended to say so; no warning surface is owed). Anchor
ownership: the card sidebar's background story is the combo alone — the well
grid's background half stays with the other anchors (StyleEditorView gains
showsBackground beside showsSymbols; the popover's symbol half already went to
its inline SymbolPicker). Quick-style recents stay palette-vocabulary — a
panel pick never enters them.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
SymbolPicker: one well at rest, a 6x6 curated grid in a popover (leading
well = the level default, clearing the key), and an optional search over
the OS's full symbol inventory read from CoreGlyphs metadata. Geometry
font-derived off StyleEditorLayout's base, grid enlarged by a deliberate
1.3x. Wired inline with the rename field in the board info popover
through the StyleCommand funnel; the Title header retires. Curated set
and the search's standing vs 03's full-browser refusal await ratification.
The pathfinder surface stack landed in e1ffe71 entangled with staged-index
placement drift: the replica's wash block sat inside replicaHeight (a CGFloat
has no .background — HEAD did not build), the body plate rode outside the cut
dim, the lanePlate property and the DESIGN bullet sat away from their kin.
Pure relocation, no new lines: every block moves to the anchor it was written
for. Verified by an isolated build of this tree.
Claude-Session: https://claude.ai/code/session_01SR4XGjmBE16ZUYWpfFHXwY
Gap #12 ruled into 04 ▸ the trash-drop bullet: the refusal fall-through's
honest reading (the column declines to be a target, it does not cancel
the held drag) covers only promises the refusal leaves keepable — lane
shadows keep drawing and an ⌥-copy lands where they are. A standing
proposal naming the trash itself is the promise the refusal just broke,
so it withdraws: the drag goes proposal-less over the column, release
with no proposal cancels in agreement with the empty picture, and
lifting ⌥ re-asks at the same address. Implementation filed on the
Implementation board.
Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
NSHapticFeedbackManager .alignment at propose()'s change edge — the
one funnel every retarget commits through, whose existing guards
(value-equal refusal, hold freeze) already make firings per-landing-
spot rather than per-pixel; withdrawal stays silent (losing a target
is not an alignment), and the external-file mode gets nothing. Force
Touch hardware only, silent no-op elsewhere; performanceTime .default
syncs the tick to the reflow the change triggers. The performer rides
an injectable seam beside holdTimeout for the same reason it has one.
Drag-perf card 11b85111 — the cluster's last.
Claude-Session: https://claude.ai/code/session_01CqjXB7ASoWtbyoGod68k97
DragReplicaStyle spells the pathfinder's constant ambient shadow once
(black 16%, radius 2.5, y 1.5) and all three replica faces — card,
lane, trashed lane — wear it via one modifier; the ghost fans compose
from the same faces so every copy inherits it, dimmed by its own
opacity. 03 § Motion's "pickup lift (scale + shadow)" supplies the
scale from the system session but no shadow on the preview image — the
replica is the only place it can come from. BoardMetrics.replicaPadding
was already written to keep exactly this shadow from clipping and was
simply unused until now.
Drag-perf card 24397310.
Claude-Session: https://claude.ai/code/session_01CqjXB7ASoWtbyoGod68k97
Modifiers were sampled only inside resolveOperation, reached from
dropUpdated and the commit — and drop callbacks arrive only while the
mouse moves, so ⌥ pressed against a still pointer changed nothing
until the next twitch. DragSession now arms a local .flagsChanged
watch for exactly the session's lifetime (begin arms, end stops, the
watchdog guarantees end; ⇧/⌃/caps don't count, a settled hold
swallows) and publishes modifierGeneration; the three retargets record
which window's surface resolved the proposal (RetargetOrigin — weak
registry identity, never board root, so cross-board and two-windows-
one-board both answer correctly), and the hovered window replays that
same retarget with the operation re-resolved FIRST, since the index
space is a function of it. The monitor is injectable — the real one
needs a live event stream no test bundle has.
Drag-perf card d491e7d3.
Claude-Session: https://claude.ai/code/session_01CqjXB7ASoWtbyoGod68k97
DragAutoScroller.run() consumes an AsyncStream of CADisplayLink
targetTimestamps from NSView.displayLink(target:selector:) on the
lane's existing anchor view (which was already the right NSView — no
new plumbing). The stream is the lifetime: cancellation ends the
iteration, termination invalidates the link, the link releases its
target. DragAutoScrollClock keeps the retired loop's two non-trivial
rules — first tick scrolls nothing, no tick integrates past 50ms (a
paused link resumes with the whole gap as its delta) — and the math is
frame-rate independent by test: one second at the edge moves 800.0
points at 120Hz and at 60Hz alike. Run-loop mode .common is
load-bearing: a drag runs AppKit's event-tracking mode.
Drag-perf card 2e08fd31.
Claude-Session: https://claude.ai/code/session_01CqjXB7ASoWtbyoGod68k97
rzen