Y2Linux 1.0 Beta 8.24 (Baseline 123)

Status: PUBLISHED as GitHub release baseline-123 (2026-10-23). Release 1.0.0-beta.8.24, Reborn 1.0.0-beta.8.24, kernel 6.18.56-y2linux-1.0.0-beta.8.24, update sequence 8800, root 2026.08-platform-v1.163, Linux 2a21690, Reborn e199d9c. Built with tools/production/release.py (136 s, resumed workspace, update based on baseline-122): container suite 867 OK, host suite 894 OK, Reborn workspace 626 tests, clippy and fmt clean; all seal checks PASS, no MediaTek firmware in any image. Installed on the owner's player before publishing. Milestone audit BETA-8.24 in roadmap-gap-audit.

Image	SHA-256
BOOTIMG	`3d59f0ac40f78efaaa3c6c5f32ede081bc1d801001418a18253ce68737ee1607`
Y2ROOT	`2b803ed645b8c4317c01c862cc779612dc6c1386f9792318d98e08b19e53f5ce`
RECOVERY	`861437fa5911566a22bdfdd713bf8222ea4b25f437bc5c200686c01b19b70118`
Y2DATA (no SSH key)	`7b87a7eb3d68f5021693910b5def632dbb357cbbe2ecc3028184973b356bed85`

Why 8.23 was not enough (#52)

Evidence	What it shows
#52 report y2linux-report-20261022-220205 (install 2813306c, 8.23, first boot after the OTA)	the crystal came up in 30 ms and the player took about 4.5 min of skipping (95 kept-sink reuses), then failed at a 48 kHz start (`PLL_READY -110`, ALSA -110) followed by XTAL_ERR (`status1=49`, `errors=5`). In this player's 8.21, 8.22 and 8.23 sessions every failure is a 48 kHz start; 44.1 kHz starts never failed. Reporter: three to four times more skips than before, and a restart did not bring the sound back
Stock `system.img` `/etc/audio_policy.conf`	the primary output (speaker, wired headset/headphone) has `sampling_rates 44100` only: Android resamples 48 kHz before the HAL. With the stock kernel's `set_freq` (rate bits only, no PLL), stock never runs the CS43131's PLL
Y2Linux 8.17–8.23	48 kHz sources played at 48 kHz through the codec PLL (fed by the crystal); the one thing stock never does is where #52's crystal falls over

What changed since Beta 8.23

Area	Change	Evidence
No codec PLL (#52)	The AFE front end and I2S back end offer 44.1 kHz only (stock's only output rate), and the codec refuses any rate its 22.5792 MHz crystal does not clock (`-EINVAL` before any clock work, logged). Reborn's wired plan always asks for 44.1 kHz; the decoder converts 48, 96 kHz and other rates (as stock's mixer did). Every track now plays on in the one open headphone sink, whatever its rate	Linux `f66d71d` (`kernel/audio/mt6582-afe.c`, overlay 0030, `tools/platform/capabilities.json`, `tests/test_codec_clock.py`); Reborn `83f0082`
Self-healing (#52)	A crystal error flag seen at a stream start power-cycles the codec (rails off 500 ms, then the stock power-up) before that stream, which waits for the restarted crystal. `y2_hp/xtal` gains `cycles=`	Linux `f66d71d` (register model)
Cold start at probe	A probe that finds the codec's rails on switches them off for 500 ms before powering up, and logs `cold start`. Correction: the commit and code comment assume a restart leaves VGP2 on; on the owner's player a software restart and an OTA restart both reset VGP2 (it comes up at its 2.5 V reset value and off), so a restart already cold-starts the DAC and this path does not trigger (`cold_start=0`). It stays as a safety net	owner's player, two restarts

Device results (owner's player, build 123)

Check	Result
OTA 8705 → 8800	restarted by itself 73 s after Update, Acknowledged on health boot 1
After boot	`ready_ms=30`, `cycles=0`, `cold_start=0`, `SYS_CLK_CTL_1=04`, `PWDN_CTL=f0`
Rates	48 kHz FLAC, 96 kHz FLAC and 44.1 kHz MP3 each play at 44100 Hz (`hw_params`), crystal selected (`SYS_CLK_CTL_1=04`), no PLL line
Skips across rates	4 × 20 Next/Previous 0.3 s apart ending on 48 and 44.1 kHz alternately, 10 alternating 48/44.1 kHz starts: all play; `sink_reused` +50, `sink_reopened` 0; no crystal/ASoC kernel line (13/13 PASS)
Software restart	VGP2 enabled (4 users) before, at its 2.5 V reset value and off at the next probe: `cold_start=0`

Open

#52/#54 confirmation from the reporters. The self-healing cycle could not be triggered on the owner's player (its
crystal never flags an error); host register model only. Crystal errors during a kept sink (same-stream track changes) are seen only at the next stream start.
Not checked on a player: a real sleep and wake, a full power-off, the click check by ear (unchanged from 8.23).
The overlay comment on the cold start stated the disproven VGP2 assumption; corrected on main after the build (in the next release).
Owner decision (2026-10-23): the crystal changes of 8.21–8.23 stay. Most copy stock, and which of them fixed #52 is
unknown. After the reporters confirm 8.24, only non-stock parts that prove unnecessary are trimmed: the probe cold start first, then the power cycle if no report shows cycles>0, then the ready wait once ready_ms is no longer needed. The screen-on hold's battery cost is measured before it is touched.