Checked against BTX 0.34.15, the current release (tags v0.34.12..v0.34.15 read in full, and every chain number below measured on a 0.34.15 node). Mainnet passed block 244,000 on 2026-10-10 (07:22 UTC), so the three rules 0.34.14 scheduled there are live. None of them touch an ordinary send: the size, fee, dust, sigop and mempool limits are byte-for-byte the same as in 0.34.12.
Big sends from a wallet of many small coins go through. One BTX transaction can carry at most 314 coins: each everyday-key coin adds 3,818 bytes and nodes refuse anything over 1,200,000 bytes (unchanged in 0.34.15). The wallet spends its largest coins first, so a wallet of mining payouts could send less every time, each send using up the biggest coins: reported as 6,000 BTX, then 4,000, then not even 3,000, refused with "larger than the BTX network accepts (1.2 MB)". A wallet holding a few large coins never meets the limit, which is why another wallet could move 20,000 BTX at once. Now a send that needs more coins than one transaction can carry goes out as several transactions to the same address, up to 20 (6,280 coins). The preview says how many, lists what each pays and shows the total fee; together they pay exactly the amount. Every one is signed and read back from its own bytes before the first is sent: each under the limit, no coin in two of them, the recipient paid exactly its share, no output but the last one's change, and the fee as reported and under the cap. They spend separate coins and do not depend on each other, so a big send may confirm over a few blocks. If one is not accepted, the receipt (amber, "X of Y BTX") says what is on its way and offers to send the rest using the same signed transactions, so nothing can be paid twice; it stays until "New send", and the Send screen is locked while parts go out. A send that fits one transaction is built exactly as before.
Measured on a BTX 0.34.15 test chain with 800 mined coins: 314 coins (1,198,909 bytes) accepted, 315 refused with tx-size; the old single transaction for 5,789 BTX (747 coins, 2,852,146 bytes) refused with tx-size; the same amount as 3 transactions, all accepted and mined in one block, the recipient credited to the satoshi. In the real wallet screens, against a second test chain built the same way: a 25 BTX send unchanged; 5,000 BTX as 2 transactions with the second refused on purpose, then sent with one click; 799.31375026 BTX, the most the wallet said it could send, as 2 transactions, leaving the wallet empty. Recipient total 5,824.31375026 BTX, exact. Signing measured about 5 ms per coin (Node on an M5), so a 314-coin transaction takes 1.5 to 2.5 seconds. A second round reproduced and fixed a way a retry could have sent a newer preview instead of the missing part (about 9,140 BTX delivered for a 6,000 BTX send) before anyone used it; the fuzzer behind that review found nothing else in 4,500 really signed cases.
Node wallet files from current BTX nodes import again. Since 0.34.8 a node writes its descriptors as pqhd([fingerprint/87h/0h/0h/0]/*), and the wallet.dat import (now "From a node wallet") only knew the older pqhd(fingerprint/0h/0h/0/*), so every current wallet.dat was refused with "No BTX post-quantum (pqhd) descriptors found". Both shapes name the same keys. It also scanned only the first 16 addresses of each chain whatever the node had used, so a node wallet past that would have been swept in part without a word; it now reads how many the node handed out from the file. Checked against a wallet made by btxd 0.34.15 (throwaway, never funded, kept in tools/fixtures): all 23 addresses the node gave out derive the same. The wallet.txt route was already fine and still is. Encrypted wallet.dat files are still refused, as before. And a node address holding more coins than one transaction carries (a miner's payout address) is now swept in parts, with the same planner and checks as Send; before, it was refused. Proven on a test chain with the 0.34.15 file: 400 coins on one node address, swept as 2 transactions of 200, the destination credited exactly.
The screens fit the window. Measured at the app's own 560 x 720 window. Settings was a centred box with a narrow side menu that left 352 px for the content, so its nine sections ran to nine screens of scrolling and the menu labels wrapped; it now fills the window with the menu as chips across the top, and keeps the side menu in a wide window. The window can now be made wider and go full screen (it was capped at 580 px); a wide window gets a 720 px column. Adding a wallet no longer runs 345 px past the window (112 px now, only the read-only option below): the four restore routes sit in a 2 x 2 grid. The tab bar stays in reach while a long history scrolls. The unencrypted-key notice is one row instead of three. A send preview brings its Confirm button into view, and a partly sent send is amber, not green. The restore buttons are shorter ("From a master key", "From a recovery file", "From a .btxwallet file", "From a node wallet"), and a footnote that repeated the hint above it (and named the chain server) is gone. Smaller fixes: "Protect with a passkey · your passkey" said the same thing twice on Linux; the no-passphrase hint said "Touch ID" on Windows too, and now names a passkey only where there is one; the server picker's dot and name sat apart; the version shown before the app reports its own said v0.27.0. Copy throughout the app lost its em-dash asides.
History drops a payment the server says is gone. A payment reorged out of the chain, or a pending send that was evicted or replaced, used to keep its row forever, still counting confirmations. Now a row that should be on the newest page but is not is looked up by its txid (at most 5 per refresh): "not found" retires it, and for a row cached as confirmed only once the server's own tip has reached that block; an answer with a new status updates it; anything else keeps it. A row missing from a page is never dropped on that alone.
Smaller hardening. A coin amount from the server must be a safe integer (any integer, even 1e300, was accepted); lifetime sums keep their own, unbounded parser. A broadcast gets more time on a slow uplink (25 s plus 1 s per 50 kB of transaction, so 73 s for a full part). Guards that had no test now have one, including the empty Developer ID entitlements that keep the app launchable.
Labels no longer turn to gibberish when you switch tabs quickly. Opening a tab replays a short scramble on its labels. Switching away and back within that animation's 650 ms stopped it half-way and took the scrambled text as the label, for good ("Send to (btxZ!4 Fc/Z$JT2"), until the app restarted. A label the app updated during the animation (the fee hint) could also be put back to its old text. Found while checking the merged 1.6.0 work in Apple's WebKit, reproduced in Chromium with full-speed timers, and older than this release.
What "settled" means is said more carefully. From 0.34.14 a BTX node with default settings will move to a heavier competing chain of any depth once it has either verified every block of it itself or watched it grow for 30 minutes; under 0.34.12 and 0.34.13's default no automatic reorg went past 72 blocks (btx src/node/chainstatemanager_args.cpp: "BTX has no finality gadget, so that ceiling is opt-in"). The 20-confirmation count and the fork check are unchanged, both rest on measured fork races, but the help no longer calls a payment "final" after 20 and says a very large amount deserves a longer wait.