Feature completion: wired audio and Classic Bluetooth

Dated implementation receipt. The campaign identities, test counts and pending actions below belong to the recorded source cut. The newest sealed integrated candidate is Baseline02. UART01 hardware now proves bounded C1/C2/C3. Baseline02 new-image/automatic cold boot and unrelated release gates remain open.

Implementation pass, 2026-10-02; Linux starting 9ac2c0b7018ce707ec15f8845daf58ce38ec18a0, Reborn starting eb64b8b. This report covers priorities 5–7. No device configuration, flash, acoustic test or electrical measurement was performed. Qualification remains independent of source and software validation.

5–6: precision and native sample rates

Final: PARTIAL / BLOCKED_BY_HARDWARE_PROOF. F075/F076 remain NOT_IMPLEMENTED in the native kernel path. Existing direct output remains S16_LE, stereo, 44.1/48 kHz; 88.2→44.1 and 96→48 family conversion remains truthful. No exact evidence found establishes a safe wider DL1 fetch mode or native 88.2/96 rate code. This is an unresolved implementation prerequisite, not proof that MT6582 silicon cannot do it.

The exact MT6582 BSP AFE header describes DL1 rate at DAC_CON1[3:0], mono selection at bit21, serial width at I2S bit1 and rates through 48 kHz. It does not establish the later-SoC AFE_MEMIF_HD_MODE/AFE_CONN_24BIT fields. The independent pinned Y2 donor (artificery-dev/linux, 53fb57bb99c916c24b65db5b7f9723be37f67826) explicitly omits 24-bit paths/APLLs, but that donor comment alone is not a hardware limit determination. No MT8173/MT67xx constants were imported.

The retained Y2 HAL analysis, second-I2S report (Y2Player/docs/Y2_AUDIO_PATH_PHASE4_SECOND_I2S.md, private research repository), locates Set2ndI2SOut and SetMemIfSampleRate: stock CONN0=0x00400020, CON3=0x90b, with no wider DL1 fetch operation identified. Current clock and board evidence retains 26 MHz audio/audintbus muxes, 22.5792 MHz DAC crystal and CS43131 PLL-family handling. Linux6.18 CS43130-family driver accepts 88.2/96 and wide sample formats; the upstream DAC acceptance does not establish the SoC DMA/interconnect/clock path.

Current distinctions:

Concrete fixes: Reborn's C membrane accepted enum S32 but its pcm_format() switch omitted S32 entirely. Valid observed S32 Bluetooth PCM therefore failed ALSA probing/opening. Added the real ALSA mapping, an exact-format null-sink open/write test and format candidates S32→S24→S16. ALSA negotiation now intersects access, format, stereo channels and rate in one constraint set; independent test_* calls previously could accept individually valid but mutually incompatible combinations. Strict observed Bluetooth contracts still never substitute formats; wired enablement profile remains the gate.

Low-bit proof now exercises packed24 WAV→FFmpeg decode→float DSP/limiter→S24/S32 packing at 44.1/48/88.2/96; samples ±1,±127,±255,±257 remain exact. Existing application volume tests preserve low bits at unity. This proves software precision, not native DAC precision.

Changes: Reborn crates/reborn-audio/native/audio.c, crates/reborn-audio/src/native.rs, crates/reborn-media/src/native.rs; Linux audio_precision.py, capture tests.

Physical proof to add to the integrated harness

1. Record candidate hashes, boot ID, y2-audio-contract, ALSA hw_params and existing AFE state while the supported direct path is open. Preserve S16 baseline first. No plughw/implicit resampler may stand in for native acceptance.
2. Source-backed wider fetch and higher-rate programming must be resolved before enabling a new kernel mode. Needed evidence is an exact MT6582 register description/stock implementation establishing DMA fetch width and alignment, interconnect width, rate codes, generator/dividers and restoration ordering. Do not write candidate unknown register bits.
3. Generate fixtures once with python3 tools/host/audio_precision.py <new-directory>. Capture BCLK, LRCLK and SDATA at the established Y2 pads (GPIO43,44,46), with owner equipment and verified probe voltage. Retain original analyzer data, board/candidate/boot identity, measured MCLK, and capture configuration. I2S decoder must account for the one-bit delay, left/right order, slot width and signed payload alignment.
4. Normalize at least64 consecutive frames as CSV time_s,left_s24,right_s24; compare using --verify-capture capture.csv --rate 44100 --first-frame N. Checker rejects S16 truncation, channel/value mismatch and >1000ppm rate error; MATCH is only an offline capture match. It cannot authenticate hardware provenance or certify analog fidelity.
5. Once native modes are source-backed, repeat44.1→48→44.1,44.1→88.2,48→96,88.2→96,96→88.2 and high→low with close/reopen, muted safe level, ALSA params, AFE/DAC state and measured clocks. Nominal BCLK at64fs:2.8224/3.072/5.6448/6.144MHz. Verify duration/frame count, pitch, low bits, no stale PLL state, pops, channel swap or underruns. Unsupported modes must refuse and retain family fallback. This is one integrated harness lane, not a separate flash workflow.

7: Classic codec implementation

Final: PARTIAL / BLOCKED_BY_HARDWARE_PROOF and BLOCKED_BY_DISTRIBUTION for optional codecs. Private integration profile intentionally enables compiled AAC, aptX, aptX-HD and LDAC endpoints while platform_qualified=false and distribution_approved=false. SBC remains fallback. Existing explicit owner policy can disable experiments. Public package review remains separate; this does not grant public rights.

BlueALSA5 quality is configured at daemon startup. y2-platform codec-settings exposes requested/effective settings and pending restart; setters persist validated choices durably and do not interrupt an active transport. Effective requires current-boot receipt plus a live matching daemon argv. Product UI consumes the typed platform contract. Actual negotiated codec remains a D-Bus observation, separate from preference/request/accepted selection. Raw remote advertisement stays unknown where GetCodecs supplies only mutual intersection.

The BlueALSA patch exposes read-only EncoderStats on PCM1: active, bitrate, SBC bitpool, LDAC quality index, ABR enabled, adaptation count and errors. Atomic fields avoid locks, filesystem writes and D-Bus signals on the encoder thread. Active clears on cleanup. Platform reports XQ only for the actual negotiated mode plus observed bitpool; ABR-linked/enabled with zero transitions is never adaptation proof. Bitrate means encoder payload, not measured radio throughput. Other codecs retain unknown bitrate rather than fabricated measurements.

Offline encoder contract opens no hardware and runs real libraries. ARM/QEMU results: SBC452/551; aptX512bytes and HD768bytes from512 input frames; AAC44.1/48 produces LATM; LDAC starts909/606/303 or990/660/330 and synthetic congestion produces respectively3/2/0 EQMID changes to303/330. This is deterministic encoder functionality, not RF, audible quality or target CPU-load evidence.

Changes: Linux codec manifest/source ledger, radio policy, capabilities, Bluetooth normalization, codec controls/observation, connectivity startup, Buildroot profile/package/patch installation, codec helper and radio-policy tests. Reborn audio files above plus reborn-platform/{codecs,bluetooth}.rs; Product UI/client work is recorded in the main pass report.

Remaining physical proof: per codec record daemon endpoints, mutual peer support, actual codec/PCM, encoder stats, audible output, CPU/load/temperature, reconnect and SBC fallback; run Wi-Fi idle/scan/transfer matrix. LDAC ABR must show real adaptation under observed link load, not just synthetic qualification. Pause/release PCM before selection; preserve sole reconnect ownership and uncertain-outcome marker. No LE Audio, aptX Adaptive or Lossless added.

Source/distribution review

Reviewed primary BlueALSA5 LDAC encoder, pinned local SBC encoder/quality selector and corresponding library headers. FDK NOTICE, libfreeaptx COPYING and AOSP LDAC NOTICE remain explicit source/notice/certification review inputs. No legal clearance was inferred. Source ledger records source URLs, exact hashes, license hashes, reviewed source and unresolved approval independently of technical enablement. Fresh legal-info must include the local BlueALSA modification and helper source.

Targeted validation

• Linux audio5 tests:4 pass,1 built-DT test deferred to integrated kernel build; capture test exercises12 valid/truncated/wrong-clock combinations.
• Platform evidence3 pass; radio/codec policy7 pass; Bluetooth platform7 pass before final manifest extension.
• Reborn audio8 and platform69 tests passed; new real low-bit decoder test passes8 format/rate cases. Final workspace fmt/clippy/test and ARM rebuild are integration gates.
• Patched BlueALSA compiles ARM; QEMU --version returnsv5.0.0. Real optional-library contract passes12 printed codec/rate/quality cases, including actual software ABR movement.
• No previous binary certifies the integration candidate; its clean build/ELF/package results belong to the final receipt.

Post-integration repair: first connection previously had no codec_policy until an explicit codec request, hiding the normal UI selector and LDAC settings. The Bluetooth worker now discovers eligibility read-only from current manager endpoints, peer mutual codecs and immutable build inventory. Discovery is cached for the observed transport epoch; disappearance/change invalidates it. A private D-Bus regression verifies that a playing stream exposes eligible choices without SelectCodec/Connect, and that stale peer choices disappear. Strict clippy passes. Saved-preference connection behavior and the SBC-only quality-row integration are covered by the Product lane.

Reliability/privacy closure: persistent codec history is platform-owned, event-only, sanitized to64 valid peer addresses and five known codec counters (maximum255), with a65536-byte publication limit. Unknown fields/events cannot produce unbounded writes; observed negotiation clears that codec's rejection count. The release scanner rejects the entire private Bluetooth state directory and diagnostics exclude peer identities/history. Invalid saved quality preferences now report a settings failure without hiding an independently observed live transport; mutation/startup still refuse invalid policy. Targeted history regression and all9 Bluetooth platform tests pass.