v1.4.6

A crash reported hours after 1.4.5, and the Arctis GameBuds finally reporting their battery.

Added

• The Arctis GameBuds report battery and connection state. The profile said no status request was known for the family, so the tray never showed a percentage and every status row stayed blank. That was out of date rather than true: SteelSeries' own device specification describes the frame in full, for the GameBuds and the Xbox variant alike. Each earbud reports its own 0-100 level; the tray shows the lower of the two, since the bud that dies first is the one that ends the session, and you count as connected when either bud is — which is what the vendor's own firmware does, so wearing a single bud no longer reads as offline. The charging case keeps its percentage to itself: it answers on its own USB product rather than through the dongle. (#202)
• Two diagnostics for problems that need someone else's hardware. scripts/diagnose-sonar-eq.py answers "where does the EQ chain break?" in five read-only checks ordered so the first failure is the answer — safe mode, the filter-chain unit's LADSPA path, a plugin that fails to load, an EQ node missing or unlinked, a config carrying no filters at all. scripts/reverse-engineering/gamebuds_battery_probe.py finds out whether a headset reports a battery at all, by listening while you put a bud in its case rather than by asking you to capture USB traffic on Windows. (#181)

Fixed

• A clips folder on a disconnected drive no longer stops the window opening. With clips on removable media, pulling the drive leaves the mount point behind: it still looks like it exists and answers an I/O error on the first read. That error escaped out of the Clips page while the main window was being built, so the whole window failed — no mixer, no Sonar, no settings, because a disk was not there. The folder is now read defensively, entry by entry, since on failing media some reads succeed and others do not. (#204)
• Clip capture stops relaunching itself when it cannot start. The same missing drive made every automatic start fail, and nothing remembered that: with a game running, ASM asked the desktop portal for a new screen capture every five seconds, for as long as the game lasted. A failed automatic start is now left alone until something changes — a different game, or you pressing the button yourself — and the reason is written to the log instead of being repeated in silence. (#204)
• A microphone chain referencing a plugin by name is repaired again. The check that spots it required the plugin and the node type to sit on the same line, and the noise-reduction nodes are written across two — so a config that could not load was never regenerated, which is the failure that check exists to prevent, left open for the one channel formatted differently.
• Two tests that only passed on the author's machine. One needed a Python new enough for tomllib, the other needed LADSPA plugins to be installed; both went green locally and red everywhere else, which proves nothing about anybody's machine.