| Commit message (Collapse) | Author | Age | Files | Lines |
| ... | |
| |
|
|
|
|
|
| |
state) at sync time, newest-first per txid, instead of adopting the newest core's list wholesale. The wholesale apply lost history: a device whose history fetch keeps failing (mobile 429 storms — the balance scan is cheap per-address calls, history is the heavy /txs pulls) publishes a THIN core with txs:[] and a fine balance, and every fresh device then adopted the newest=thin core and showed a balance with no transactions — adam's phone today, with three ~2KB thin cores on the relay shadowing the fat 7KB core from Aug 18 that still holds the history. Union is safe because history is append-mostly and every SUCCESSFUL local scan still prunes rows its addresses no longer report; a wiped device also self-heals the fleet, since its republished core carries the union'd txs back to the relay
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MGzU12QrAA4G6BYGzGVdXz
|
| |
|
|
|
|
|
| |
the pager's Next; right = newer) instead of bouncing straight into the Receive tab — the same swipe falls through to tab navigation only once the pages run out in that direction. historyTab publishes what's on screen ({page, pages} in _histPager, cleared for detail views/spinner/empty) and the wallet-screen swipe gesture consults it when the touch began inside the list card (.history-card)
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MGzU12QrAA4G6BYGzGVdXz
|
| |
|
|
|
|
|
| |
not after the slide lands. The old model held ALL renders until the glide settled (a mid-glide morph used to strip the drag's inline snap-none and yank the card — so nothing could paint until rest, and the history switch waited out the whole animation). New model: the strip alone is frozen against the morph (_skipMorph short-circuits that subtree) while renders run freely, so the release can commit the account switch immediately — desktop knows the chosen card exactly (it computes the target itself), touch predicts it from the release position + scroll velocity (nearest card, pushed one over on a real flick — the same physics mandatory snap applies). settle stays the authority: it unfreezes the strip, runs a syncing render, and corrects a rare wrong touch guess when the snap actually lands. Renders stay held only while the finger is DOWN (pointermove writes scrollLeft on the main thread — a render there eats frames)
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MGzU12QrAA4G6BYGzGVdXz
|
| |
|
|
|
|
|
| |
key is now kept in the wallet's nostrlogin state and resumeLogin reattaches from it instantly, so opening chat no longer greets passkey users with "Your nostr signer isn't connected". Kept deliberately and only for the passkey door: a pasted nsec keeps its used-once-never-stored promise, but a passkey key exists only through this app's PRF derivation and re-deriving it costs a WebAuthn ceremony (system dialog + biometric) on every open — and the same storage already holds the wallet seed, so this is the same exposure class. Existing passkey sessions heal on their next reconnect: that ceremony saves the key, and none is asked for again
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MGzU12QrAA4G6BYGzGVdXz
|
| |
|
|
|
|
|
| |
Savings button (and the unfold panel) used to render only on the SELECTED card, and the strip's flex row stretches both cards to the tallest — so the whole strip resized right as a swipe settled onto Spending. The extras now belong to the spending card itself, selected or not, so card heights never move during a drag; tapping the button on the peeking card selects Spending and opens the move panel in one go
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MGzU12QrAA4G6BYGzGVdXz
|
| |
|
|
|
|
|
| |
after verification moved off (chunk-60dfb4nt in profiles). The verify worker grows two ops: 'openWraps' opens community (Concord) wraps given the symmetric stream conversation key — nip44 decrypt + the seal's schnorr verify + hash checks per wrap, no secret crossing the thread boundary since stream keys are group material; 'unwrapDMs' runs the full NIP-17 unwrap (two ECDH ops + seal verify) when the wallet holds the raw identity key. Remote signers can't export their key, so their DM path stays on the main thread, exactly as before. The time-sliced main-thread queues survive as fallbacks — a dead or stalled worker only changes WHERE the crypto runs (workerCrypto resolves null and callers re-enqueue), never whether chat catches up. Bonus: subChannel derives its stream key once per channel instead of once per wrap (groupKey does hkdf + ECDH per call — real curve work on a 200-wrap backfill)
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MGzU12QrAA4G6BYGzGVdXz
|
| |
|
|
|
|
|
| |
the Ark mark: a faucet paying the LN address settles on the registrar's CLN and arrives as an anonymous arkoor from the float (the mailbox protocol carries no memo), so the wallet genuinely couldn't tell the rail. The registrar now records a receipt {sat, hash, ts} per LN settlement it forwards via the float (7-day ring per name) and serves them at GET /lnreceipts (NIP-98, owner/manager only — it's a feed of incoming amounts); the wallet fetches them through the names feature (which holds the signer) and fuzzy-matches receive movements by exact amount + delivery within ±3 min, stamping viaLn + the payment hash so a receipt can never claim a second row. History row and detail view show ⚡ for viaLn receives; the tag rides the sync snapshot to other devices via merge enrichment. viaClink deliveries need none of this — the wallet's own LN-receive flow already tags itself. Best-effort by design: an amount+time collision only mislabels an icon
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MGzU12QrAA4G6BYGzGVdXz
|
| |
|
|
|
|
|
| |
buffers, frame-drop + scroll-position tracing) now that the stutter is fixed — the render-gate-during-drag fix stands on its own
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
showed scrollLeft climbing (19→40→64) then JUMPING to 0 and stalling ~80ms before gliding — a position jump at a steady 17ms/frame, not a dropped frame. Cause: a background render (ark update, balance tick) firing mid-drag; the morph rebuilds the strip and strips the drag's inline scroll-snap-type:none, so mandatory snap yanks the card to its origin. Fix: renders are held for the sub-second of a carousel drag and one flushes when motion stops, so the DOM never churns under the finger. Verified: a mid-drag render now leaves scrollLeft untouched
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
survives in the buffer regardless of read timing
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
jump during a flick (a stutter that isn't a dropped frame)
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
rebuilding it costs ~60ms (sort + dedupe + gift resolve + secp256k1), and its 1.5s TTL meant an unhurried flick always hit a cold rebuild, stuttering the slide's tail — 'warms up if I keep going fast, cold if I pause, refresh reproduces' is the classic TTL-expiry signature. Now keyed on a state revision (_save bumps state._rev) so it's reused until the data actually changes and never re-cools between flicks. Diagnostics still on to confirm
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
frame gap into __dl so a real drag's jank appears next to the release/settle it overlaps
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
release decision (vel/travelled/dir/from/best) and settle outcome, to capture what a real failing flick does. To be removed once diagnosed
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
unfolded on the balance card with no way to dismiss it short of completing or switching tabs. A centered Cancel link closes it back to the plain card
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
release. My new travel-fraction variable was also named 'moved', colliding with the drag-started flag of the same name in the enclosing scope — a temporal-dead-zone ReferenceError that threw up() before it could commit or even snap back, so a release did nothing. Renamed to 'travelled'. (This is what made 4f0e444's drags dead.)
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
latest video shows the card sliding a third of the way toward the neighbor then springing back to where it started — a short flick that never crossed the halfway settle point. Release now commits to the neighbor on a deliberate small move (past ~15% of a card) or any quick flick, instead of only past the midpoint; otherwise it snaps back. Direction comes from flick velocity, or from how far the grabbed card actually travelled
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
synchronous verifyEvent (a schnorr check per event, ~560ms on a relay backlog — the profiled carousel-stutter cause) is switched off, and every event is now gated through a Web Worker before any caller sees it. Nothing unverified is ever trusted — the worker runs the same nostr-tools verifyEvent, just on another thread, with a synchronous fallback if the worker is unavailable (standalone file://) or stalls. Bundled to dist/verify-worker.js, precached in the SW
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
every community room up front (500 control + 500 guestbook + channels each — the bulk of the ~560ms of boot secp256k1 behind the carousel stutter). Now boot builds rooms from the decrypted local cache ONLY (subscribe:false), so the unread dot still reads last-known state, and the live backfill's verify/decrypt waits until the user actually opens Chat (homeView/room view subscribe then). ensureRoom split into cache-build + subscribeRoom
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
every spendable coin and built the exit txs (heavy secp256k1) synchronously on each render, and the cold first compute after a refresh landed right on the carousel drag. Now the render only reads the cached answer; the recompute runs in requestIdleCallback (keyed by coin set + feerate, so it fires only when something changed) and re-renders when done — the EC never competes with input
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
sign + vtxo decode, real secp256k1) for every spendable coin in one synchronous map — a ~100ms+ EC burst on boot, the local work blocking a carousel drag right after a refresh (community/DM wraps arrive over the network later; this fires immediately on init). The requests now kick off in ≤6ms slices that yield; network calls still overlap
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
the exact cold first-drag window showed ~700ms of secp256k1, and it was the COMMUNITY room's gift-wrap decryption — a relay backlog of 500+ wraps decrypted inline in the subscription callbacks, one synchronous EC burst on boot (the DM inbox already yielded; community rooms didn't). Now their decrypt work is time-sliced through the same yield-every-8ms queue, so it spreads over a few frames and the wallet stays responsive. Measured: first cold flick dropped 6 frames before, and this was the dominant cost
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
scrollend doesn't fire reliably for programmatic scrolls, so relying on it left the card visually switched but the account state stale. Back to the onScroll debounce — but the real mid-glide stall was never the debounce (it trails the glide's own scroll events and fires after motion stops); it was an EXPLICIT 150ms settle timer in the drag release firing mid-glide. That's gone, so the switch render lands only once the card is at rest. scrollend kept as a bonus early trigger where it does fire
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
stall showed up ~150ms into a slide — exactly when the settle timer fired the account switch and rebuilt the history list while the card was still moving. The switch now fires on scrollend (motion truly at rest, card already in place), so its render is invisible; the debounce timers survive only as long fallbacks for browsers without scrollend
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
(NIP-44 ECDH, ~500ms of secp256k1 in a boot profile — the last of the carousel jank) ran as one uninterrupted loop. Each synchronous run is now capped to ~8ms, one frame's budget, then yields so input and paint get a turn; the backfill just spreads over a few extra frames instead of freezing them
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
first 9s after refresh showed ~800ms of secp256k1 (q0/mul/TJ) dominating the main thread; the source was _key(index) recomputing a BIP32 derive + a getPublicKey point-mul on every call, and reconcile() calls it for every spendable coin (56 here), with decode/send/exit/fee-math hitting it again. A key for an index is constant per wallet, so it's cached by index now (mailbox key too) — dozens of EC muls per boot collapse to one each
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
Every _save() — fired dozens of times during a boot reconcile — did a JSON.parse of stored state, a merge over every vtxo and movement, a JSON.stringify, a synchronous localStorage write, and saveCache(). A CPU profile of the first 9s pinned it: ~230ms in saveFeatureState/loadFeatureState/setItem bursts blocking the main thread, which delayed the mouse drag's scrollLeft writes (mouse drag is main-thread; touch scrolls on the compositor). The in-memory state is always current and every live read uses it, so the disk write is now coalesced — immediate when idle, one write ~200ms after a burst settles (800ms cap), flushed on page-hide and wallet switch so nothing is lost
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
back. On open, when Savings is empty and Ark has a balance, the wallet flips you to Spending — a convenience. But it fires on every Ark update and only latches once the balance SYNCS, which after a refresh arrives a second late — right after you've dragged to Savings, so it overrides your choice and slides back. Now any deliberate pick (carousel scroll/tap/flick, swipe, accounts screen) sets accountUserChosen, and the auto-select consumes its one-shot without flipping when you've already chosen
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
vertical scrollbar: a short page has none, the history list filling in makes the page tall enough to get one, the viewport narrows ~15px, and the snapped card re-centers into the narrower width — a horizontal nudge right as history lands. scrollbar-gutter:stable reserves the gutter so growing content never changes the width. And the 'false start' — a boot render's alignment firing a scrollTo mid-drag — is guarded: alignment now bails while a finger/mouse is down
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
snap-type was restored on a fixed 400ms timer, which during rapid back-and-forth fired while the next glide was only partway — re-applying mandatory snap then makes the browser snap to the nearest point, the card you started from, yanking it back. Snap now restores on the scrollend of that specific glide, guarded by a per-drag token so a stale restore from an earlier throw can't fire during a later one (timer kept only as a fallback for the no-scroll case)
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
no longer reads layout (offsetLeft/scrollLeft) on every one of the many boot renders — an _alignedTo guard short-circuits before any layout touch once the card is in place, and the scroll's own settle marks itself aligned so it never re-aligns behind the user. The pass moved to rAF, and each slide gets layout containment so the history list building in below can't force it to reflow mid-scroll
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
transaction history instead of replacing it, so returning to the list is a scroll, not an unintuitive re-tap of the toggle. A tapped history detail (or the receive celebration) still claims the whole area
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
a small button in the grouped row with copy tx id / view explorer / copy address — one consistent size for every secondary copy action
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
secondary copy/view action (copy gift link, copy tx id, view on explorer, copy address) now rides one wrapping row at its natural width, rather than each stretching full-width and getting split apart by the block buttons between them. Revoke (now danger-styled) and Back keep the full width they deserve as primary actions
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
and exit each settle as a real bitcoin tx — txid, bc1q address, block explorer — so they wear the Bitcoin mark in the history row and detail header instead of the Ark logo; pure off-chain hops (receive/send/renew) keep the Ark mark, Lightning keeps its bolt
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
coin id — the folded row rides ui state (and nav history) since it isn't a stored movement. Ordinary send details also stop hiding their inputs: 'Coins spent' lists what a payment gathered (up to 24), beside the destination vtxo it minted
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
— an ordinary send row, because it was the user's own spend and nobody needs an alarm about which device made it. (The ASP's status API returns only a spent bit, so the shown time is when this wallet caught up)
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
row is now a plain Sent wearing a quiet 'synced' tag, and tapping it unfolds the reassurance — the coins left your wallet in a session this device didn't see (another device, a restore that dropped history, or a server-side renewal), and the balance is correct; the row just keeps the history adding up
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
self-custodial edition: a name's owner registers an https URL (NIP-98, owner or manager key) and every Lightning settlement for that name POSTs {type, address, amount, hash, preimage, memo, secret} there, firing on settlement independent of the ark delivery. URLs are vetted hard (https only, no credentials, every resolved address must be public unicast — RFC1918/CGNAT/loopback/link-local refused, re-checked before every delivery), retries back off 30s/5m/30m from a persisted queue, and GET /webhook shows the config plus the last delivery's outcome. Proven live: a throwaway name, a 21-sat Lightning payment, one webhook delivered
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
profile answer (the 5s query racing still-dialing relays) is retried in minutes instead of being trusted for the 24h TTL, faceless rows are never persisted (they were evicting real faces from the 150-cap), and the logged-in identity's own profile is pinned against eviction entirely
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
ASP's service charge (base + lifetime ppm) record separately and the history detail shows them on their own lines — settling the mystery of a ~350-sat chain fee inside a ~650-sat total. The community row's member count anchors to last session's settled figure instead of ticking upward through the guestbook replay, and a profile's Pay button reached from search actually reaches Send — the lingering search screen was winning the router
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
freezes the tab (relay sockets die as the approval publishes) and its callbackUrl then RELOADS the page, discarding the client key mid-handshake — so the pending connect now persists to sessionStorage, the wait races the live subscription against a stored-event poller kicked on every return to visibility, and a fresh boot resumes the handshake instead of shrugging. Proven both ways with a scripted signer: approval after reload, and approval stored while the page was gone
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
class changes between screens is a different control wearing the same node, and its background transition was crossfading the old style into the new one — repaints aren't style animations, so the swap now snaps and the transition comes back after the paint
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
button or finger is still down (release is the choice, shared settle logic for mouse, touch and glide) — and the coins table's dotted meta line becomes little labeled pills: an hourglass countdown, depth, and exit cost each in their own chip, the expired one wearing the warning color
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
emits standalone.html (~950KB, everything inlined, boots from a bare file:// with no server), How it works grows a 'Keep a copy' card linking the download, and the tab strip's phantom double-tap dies — the pointerdown's paired click was arriving after the render had already swapped in fresh handlers, and with toggling tabs that ghost click un-did the tap it belonged to. The timestamp guard now lives on the DOM node, which survives the morph
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
~100ms throws the card in the flick's direction, so a quick short pull changes accounts instead of demanding a half-width haul; a slow release still settles nearest
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
one puts it away), every landing — login, account creation, an account snapped into the carousel — is the account's own timeline, and that timeline is now SCOPED: Savings shows the on-chain ledger and gift reservations, Spending shows the ark side. The carousel meanwhile learns mouse and mind-reading: desktop gets a real grab-and-pull drag (snap lifted mid-drag, restored on the glide), the peeking neighbor switches on a single click, and the strip wears grab cursors. And a thread's avatars and author names finally go somewhere: profiles stack on top of the conversation and back returns to it
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
riding a full-bleed strip — the neighbor card peeks in from the screen edge and snaps into place, instead of two faces crowding one card's chrome. The active card carries the move/top-up buttons and unfold panel; the peeking one stays a clean preview
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
faces with the neighbor peeking at the card's edge, dragged and snapped by native scroll-snap — settling on a face IS the account switch, programmatic switches glide the carousel to match, and the old swipe-to-flip gesture yields to the carousel's own drag (it still serves single-face wallets)
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|
| |
|
|
|
|
|
| |
padding (11px/16px, asymmetric in a fixed 42px round) was shoving the ✕ off-center — flex-centered with the padding dropped
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01MrkLsNScXe9b6BviUSdmMr
|