This is a test build. Stable stays on beta_65. Flash it from wadamesh.com with the Beta toggle on, or turn on "Get test builds (beta)" on the device.
If your T-Deck has been typing the wrong letters, this is the fix. One change matters here; the rest is a footnote.
Older T-Decks could not be typed on
Pressing a produced q and m produced qa , which makes the keyboard useless for writing, navigating, or locking the screen. Reported by @Vybo in #341, and it has been broken since beta_70.
It affects the older T-Deck keyboards, the ones without the controllable backlight. At startup the firmware asks the keyboard whether it speaks the newer column protocol that modifier latching needs. An older keyboard answers a five-byte read with the pressed character in the first byte and padding after it, and every lowercase letter has its high bit clear, so that answer passed the test. Read as a key-position report it is nonsense:
'a' = 0x61 -> bits 0,5,6 of column 0 -> "q "
'm' = 0x6d -> bits 0,2,3,5,6 -> "qa "
Nothing can tell the two keyboards apart while nobody is typing: idle, both answer with zeros. Requiring proof before committing sounds right and is useless: proof only arrives when someone types, nobody types during boot, so every device falls back and silently loses modifier latching. That was measured, not assumed.
So the initial guess stays optimistic and becomes reversible. Raw mode now inspects each frame, and three or more bits in the first byte with a printable value and every other column empty is not something a key matrix can report, but is exactly an ASCII character. On that, the firmware switches to the older protocol, delivers the character, and carries on.
Checked against the real key tables: all 26 lowercase letters trigger it, so any letter corrects the device on the first press. Going the other way needs three or more of q, w, a, space held simultaneously and nothing else, which the printable-range test narrows further.
Newer keyboards are unaffected and keep latching, verified on hardware.
One more thing worth saying: the beta_72 change meant to make this detection more reliable made it worse, because retrying gave an older keyboard more chances to look like a newer one. That is why upgrading to 72 did not help anyone affected.
Translations
Two Hungarian strings from @pisti87 (#340): Mark read, which had never been translated, and a shorter wording for joining the public channel so it fits its button. Already on devices, since language files publish separately from firmware.
For testers
If your keyboard was typing nonsense, you should be able to install this normally rather than needing a cable.
A reminder while it is fresh, because it caught someone this week: two separate settings control whether an arriving message lights a dark or locked screen. Message flash wakes the screen; At a glance draws the preview text on it. Both live in Options > Display, and turning off At a glance alone will not stop the screen lighting up.
Still open and not fixed here: a map that renders nothing after beta_71 (#339), a screen that will not stay dark after a manual lock (#333), and the web flasher hanging at "wrapping up" on the T-Display P4 (#342).