Platform v1 roadmap

Current state — 2026-10-16: Y2Linux 1.0 Beta 7.6 (Baseline 67)

The public beta line is current: 1.0 Beta 7.6 (baseline-67, update sequence 6100) is the latest GitHub release; players update themselves over Wi-Fi (Settings → About & Updates → Update Now). Per-release evidence: the 1.0 Beta 7.6 report (earlier ones in docs/README) and the milestone rows BETA-1 … BETA-7.6 in roadmap-gap-audit; what works and what does not: current platform state.

The sections below are the history of how the platform got here (deep-sleep work, POWER-FINAL candidates); their "latest" labels refer to their own dates.

Current deep-sleep investigation — 2026-10-07

POWER-FINAL-06 bring-up candidate (2026-10-09, latest, source + offline only): vendor infra_pdn=1 (INFRASYS/DDRPHY power-down) on the already-present pcm_suspend, with PERICFG save/restore, automatic retained-mode fallback, a recoverable MUSB L1 guard and park-not-unregister on a terminal USB stop (the cause of the Wi-Fi-time Oops). Physical status NOT_RUN; owner flashes. Corrects the 05 report: Y2Linux already ran pcm_suspend with infra_pdn=false. After hardware qualification the production cleanup/refactor and a clean rebuild follow. MORE_WORK_REQUIRED, no milestone promotion.

POWER-FINAL-05 installed (2026-10-08): 04 + vendor PMIC LP_SEL on five always-on LDOs, runtime-revertible. Probed on device: 1573 s suspended, same boot, golden-table gap 22→17; benefit unmeasured (no current sensor). Next: overnight drain, Power/RTC wake, early-boot on 05, optional USB-meter top-up measurement; the vendor pcm_suspend state (INFRA/DDRPHY and 26 MHz off) is the next large, high-risk lever (image 06). MORE_WORK_REQUIRED, no milestone promotion.

04 diagnostic successor installed and trialed (2026-10-08, latest): DEEPSLEEP-FINAL-04 (Linux 5e26757) is flashed; owner-driven normal-policy trials passed: 469 s first sleep and a one-hour run with six maintenance wakes and 3805 s suspended, same boot, taint 0. This is encouraging but not a fix (diagnostic-only change, nine clean entries, stall rate unknown). Next: reboot + early-unplug test, then overnight drain; Power/RTC wake, formal 30-min/2-h gates and the display/poweroff/electrical gaps stay open. MORE_WORK_REQUIRED, no milestone promotion.

Late 2026-10-08 analysis (source only): the recurring pre-SPM device-phase stop is localized to an unrecorded wait after a returned callback; a diagnostic successor source with per-device WAIT/LOCK records and a built-in stall sampler is prepared and tested but not built, installed or run. Classification remains MORE_WORK_REQUIRED; the next step needs the owner's decision to build and flash a diagnostic successor.

Evening 2026-10-08 follow-up: archive67 records an orderly user-requested poweroff of boot81dce40c at about09:55:39 CEST and a new bootad0d5416 about 16 seconds later, using host-anchored BOOTTIME rather than device wall time. This does not change the earlier unclean pre-SPM failure. The03 recovery guard still matches failed boot0132fb21. On the new boot, SPM0/0 and421 guard refusals confirm no whole-system sleep through the seven-hour unplugged interval; BOOTTIME and MONOTONIC advance together. Journaled provisional SOC fell 57%→32% before cable return, with voltage about3.862→3.736V. Live connected reading at17:00 is Charging31%, with no charger faults. The displayed level is not established as unchanged by device evidence; no measured current/capacity exists. Exact03 software, application, filesystems and USB remain healthy in the read-only check. Preserve the guard. Classification stays MORE_WORK_REQUIRED with pre-SPM diagnosis first and all longer gates open.

**Earlier, morning2026-10-08:03 automatic suspend FAILED, followed by an unexpected reboot.** Archive66 preserves the read-only morning SSH evidence. The owner reports about88%→55% over approximately six unplugged hours and confirms no manual restart. Journal88% and final pre-cable55% agree. Boot0132fb21 attempted kernel suspend at398.407s, then ends with retained DPM_PREPARED/seq8; boot81dce40c reports the preceding shutdown unclean. Exact reset cause remains unknown. This new boot spent roughly6.44h running, with SPM0/0 and380 refusals from the pending03 recovery guard. The guard remains intact.

CPU0 runtime DORMANT did work (807610 entries/resumes, zero restore failures), but it is not whole-system sleep. Pre-attempt ADC4.092187V and last Discharging sample3.821484V show depletion across the reboot; provisional SOC and differing load prevent capacity/current or battery-health conclusions. Morning taint0, application responsive/paused, PCM closed, no scan, radios off, ext4 errors and warnings0, USB/charging healthy. No device changes or new test were made.

The evidence refresh and overnight report keep readiness MORE_WORK_REQUIRED. Next is the recurring pre-SPM device-phase failure on03; controlled30-minute/two-hour/endurance and the independent display, poweroff, cold-C3 and electrical gates stay open. No milestone promotion. The pre-observation plan and installation-only status below are historical.

Previous checkpoint:03 passes quick SSH installation verification (archive65). Linuxcfb7e09 / Reborn97ff8c3 / rootv1.33 / kernel03 and installed binary/helper hashes match the sealed baseline. Boot0132fb21..., taint0, working first frame, USB, charging without reported faults, and all three ext4 error counters0. Automatic policy and synthetic USB wake fix are present; operation owners clear, old02 quarantine identity inapplicable, no collector/alarm/backstop armed. No runtime changes were made. This is INSTALLATION_SANITY_PASS only: USB remains attached and SPM entries/resumes0. Exact03 full sleep and endurance remain NOT_RUN. The owner's exploratory overnight plan and morning archive stand; actual start is not yet recorded. All reliability/electrical/qualification gaps stay open.

Latest owner request, 2026-10-08: build a fresh03 baseline. Package all completed production fixes with Reborn97ff8c3 into one BOOTIMG/Y2ROOT successor, rootv1.33, after the full software and package checks. Preserve Y2DATA, sealed02 and matching symbols. Existing live02 remains guarded: archive64 was read-only, and the proposed overnight enablement below did not happen. No agent flash. The build boundary audit keeps exact-source physical verification and the independent reliability, charging/electrical and endurance gaps open. After owner installation, verify03 identity and USB/charging before deciding the overnight observation.

Later owner steering, 2026-10-08: the owner will leave the Y2 unplugged for the remainder of the night and return in the morning. Prepare an exploratory normal-product observation on the already physically tested live source: save the starting state and clear only the between-test diagnostic quarantine under operation locks. No new code, alarm or forced suspend; all normal inhibitors, battery checks and recovery protections remain. Archive64 rechecks the same untainted boot and completed prior return. The refreshed audit records this later authorization; the previous parked plan below is superseded only for tonight's observation. Thirty-minute/two-hour acceptance and all unresolved reliability/electrical work remain open. Morning evidence decides actual sleep duration and drain; do not infer them merely from cable removal.

2026-10-08 session stop at the owner's request: archive62 preserves the early USB return from trial61. Normal policy accumulated 943.163696289 seconds of full suspend in two same-boot returns, including one 600-second maintenance wake with the screen and radios still asleep followed by automatic re-entry. SPM8→10, taint0, COMPLETE667, Reborn restored, ext4 errors0. USB insertion ends the second interval; the helper labels that wake unknown. This is interrupted, not a thirty-minute drain pass. Archive63 verifies no test/helper, alarm or backstop remains active and product quarantine is restored. USB/charging work. No further deployment, test, reboot or image work is scheduled tonight.

At the next session, verify identity and live overlays again, qualify the local maintenance correction with a bounded short normal-policy test, then repeat thirty minutes before two hours/overnight. No milestone is closed. The audit below retains all independent reliability and electrical gaps.

During61, a local regression exposes a second maintenance issue: ordinary battery notifications can stale the pre-pause wake count and cancel re-entry. Local-onlyd43b592 refreshes the count after the safety interval without losing Power/RTC or the atomic late-event guard;27 tests pass. It is not deployed. The physical interruption did not reproduce this race: the first maintenance re-entry succeeded. Qualify the corrected source before longer endurance.

Archive59 passes normal automatic sleep with the recorded Python overlay and USB software-budget wake permission disabled: one203.051331-second interval, RTC IRQ+1/Power+0, COMPLETE573, same boot/taint0, Reborn restored, no filesystem errors. Named maintenance and connected-USB refusal also pass. USB reconnect and charging remain functional with physical wake owners enabled. This proves the narrow cause/correction; sealed02 does not contain either fix.

Archive60 deployed the tested endpoint/journal preservation change under both operation locks; first_entry survives the maintenance return in62. The next endurance gate remains at least1800 actual SPM seconds through normal automatic sleep and600-second dark maintenance returns. Preserve fresh post-wake SOC separately from the immediate, possibly stale receipt. Longer endurance, cold C3, electrical minimum and the original reliability defects remain open under the refreshed audit. No release promotion, image rebuild or flash at this boundary.

Archive56 confirms the single-write correction, but automatic requests still abort before SPM: three EBUSY returns and no new sleep. Only y2-usb-input wake counts increase2→8. Its software budget callback raises a wake event during MUSB suspend. Next admitted test removes wake capability from that synthetic supply only, preserving physical charger/RTC/Power wake and all charging limits. The matching kernel source uses no_wakeup_source; the installed runtime policy will be recorded separately. Longer endurance remains unadmitted until normal product completion passes. Existing independent reliability gaps stay open.

Archive53 proves the normal inactivity path requests suspend after a named maintenance owner releases its flock, but product completion FAILS. Buffered Python sysfs I/O replays an initially refused suspend during close, then reports the original EBUSY despite a198.585602-second same-boot RTC return. SPM7/7 and taint0 are preserved. Host and installed-Python fixtures reproduce two stores and the stale error. Next: a single unbuffered control-write fix, targeted error regressions, archived userspace overlay and the same bounded product test. Quarantine remains until that review. Existing intermittent devices and warm display failures, clean C3, electrical and long-endurance gaps remain open; no image/release promotion or electrical-policy change.

Archive50 adds five clean full returns, including305.200562 s RTC residence: total Power2/2, RTC4/4, same boot/taint0, SPM6/6 with no restore failure. Five-minute ADC4.057031→4.052416 V, SOC89→89, no unexpected wake. Next admitted gate is a bounded test of normal automatic sleep with live maintenance ownership and RTC wake. Quarantine is released only for that reviewed test, then restored. Longer drain and the two unresolved reliability defects remain open.

Clean02 Power wake now passes in archive47: five stages, same boot f567b2c3..., taint0, 37.950378 s SPM residence, Power IRQ +2/RTC +0, COMPLETE163. Archive48 adds successful PCM playback/pause and internal/SD fsync/hash checks. Next is the admitted bounded RTC/Power repeat plus five-minute measurement; quarantine remains. One pass does not close the earlier intermittent devices reset or warm-start display fault. No product/endurance or release promotion.

Hardware restart has restored clean boot f567b2c3... on exact 02, taint zero, normal frame IRQs and owner-confirmed player screen; archive45b is complete. Proceed with the admitted clean PM stages and one Power wake, RTC300 fallback, without diagnostic modules. Original intermittent devices and warm-start display defects remain open; product auto-sleep is still quarantined.

Service reboot after archive42 reproduces warm-display startup failure on boot7c416e45 (taint512 before any probe). Archive44 preserves it and the previous RTC COMPLETE seq197. Owner hardware restart is requested; obtain an untainted boot for staged Power wake. No PM trial is armed. Warm-start display and the original intermittent device wait remain open reliability defects.

Archive42 now proves installed 02's built-in suspend path returns from RTC on the same boot: five stages pass diagnostically, 24.529053 s actual SPM sleep, COMPLETE seq197, no restore/filesystem failures and Reborn restored. Observer taint4096 remains. Next remove that observer by archived service reboot and verify the clean, normal path; original intermittent device/warm-display failures and product/endurance gates stay open.

Latest return 40-device-wait-clean-display-return/: freezer and devices complete on boot 22bd28fa..., with no new warning or filesystem errors and working display restoration. Matching stack observer is diagnostic taint4096; the original intermittent devices failure remains open. Next is five staged checks followed by exactly one bounded RTC full-suspend trial, conditional on all non-taint checks, to verify the installed built-in context correction. Product auto-sleep stays quarantined until clean reliability is established.

Current checkpoint (2026-10-08 local): archive 34-device-wait-normal-return/ shows freezer returns but Reborn display restoration fails on warned boot e6b85337...; devices is never reached. No trial remains armed. Restore the display through its normal owner and fix the observed renderer retry failure before continuing original DPM_PREPARED localization. Product sleep remains quarantined. This supersedes the earlier next-action checkpoints below without changing any release, electrical, memory, partition or flashing gate.

Read-only active-display capture now identifies OVL FSM 0x100 (h_wait_w_rst) with no progress across 35 samples. The subsequent stopped-pipeline recovery diagnostic loses transport; normal KMS disable precedes module loading, so the stopping point and whether any reset occurred remain UNKNOWN. Recover its durable checkpoints after the requested owner restart before another test. The pre-test archive is complete; no backstop was armed for this diagnostic, an omission that must not recur. Success must restore actual frame IRQs, not just a modeset return value. This stays a diagnostic gate under the current audit.

Latest physical gate: owner installed corrective DEEPSLEEP-FINAL-02. USB SSH verifies exact Linux 5a32683 / Reborn 5a9f758, root v1.32, boot f350d9dd..., taint zero. USB configuration and charging return, with zero bus-clock failures. Archive 26-successor02-initial/ recovers 01's exact identity and stage-2/-EIO failure evidence. Next: five staged PM checks and one bounded clean Power trial; then RTC/repetition/product policy/endurance if the return passes. The clean-image audit retains all unresolved coverage and release gates. Earlier checkpoints follow.

Clean trial result: freezer passes; the devices stage resets before any real entry. Retained DPM_PREPARED and successful codec/storage/I2C callbacks reproduce the earlier device-phase stall. Archive 28b-clean-power-recovered/, new boot e1df409f.... Next smallest gate is a bounded matching-kernel blocked-task capture during devices-only testing. No Power/RTC pass or endurance admission.

The first verbose-entry device diagnostic returns without reproducing the hang. A refined capture preserves production verbosity through entry. Its replacement requires a service reboot after archive; this also exposes an intermittent DRM startup vblank fault (boot e6b85337..., warning taint 512), retained as an open display/warm-boot defect. Further bounded PM observation is explicitly diagnostic on that warned boot; it cannot qualify production sleep or erase the clean fail.

Current gate: owner-installed DEEPSLEEP-FINAL-01 boots Reborn, but USB enumeration and charging fail. Suspend testing has not begun on it. The owner saved a diagnostic export; USB-only SSH prevents live collection until transport returns. An actual CCF/probe regression proves the new childless gate omits the inherited clock ownership handoff. Correct that after verified idle recovery, retain failed 01 bytes/evidence, and complete software/package checks before owner installation of the corrective image. No charging/current limit or hardware safety bypass. This supersedes the prior candidate readiness below; see the refreshed regression admission.

Correction 02 is now software-validated: Linux 5a32683 / Reborn 5a9f758, 543 production cases, fresh ARM/kernel/config/DT/modules/QEMU/ELF/package. Owner-only installation follows the saved diagnostic export. First recover 01 logs and verify 02 identity, USB and source-backed charging; only then stage PM. No clean suspend or electrical readiness claim advances at this boundary.

Latest boundary: the complete repeated diagnostic collection records 1811.835 s of full suspend in 1847.315 s, four expected RTC wakes/no unexpected wakes, 4.050439→4.013964 V and old SOC 86→86. Total 14 same-boot diagnostic returns on 404ef43c..., with module taint explicit. The earlier device-phase wait, separate poweroff panic, clean repetition, product use and electrical minimum remain open. No two-hour or full-suspend overnight result exists.

One successor is built and software validated from Linux 0c9b78c / Reborn 5a9f758: DEEPSLEEP-FINAL-01/root v1.31, 542 production cases and 332 Reborn workspace cases plus kernel/DT/modules/ABI/QEMU/package checks. Seal BOOTIMG/ Y2ROOT with exact 04 fallback and request owner installation; preserve Y2DATA. This includes context/CIRQ/USB/product fixes and truthful suspend-time SOC/ post-resume UI reporting. No electrical limit or SOC curve changes. Next physical gate is clean installed identity/staged PM/repeated Power/RTC/auto-policy, then longer drain tests. All #16/#27/#28/#29/#31/#32/#33/#34 remain open; closed #30 does not qualify this work. No release promotion or autonomous flash. See the sealing audit. The earlier progress notes below are retained chronologically.

POWER-FINAL-04 is physically installed (Linux 5aeba051, Reborn 8cd365da, root v1.30), verified by locked SSH on boot dbcb3854-d028-4882-94aa-cc86b5af091e. This supersedes older installed-baseline wording below. The boundary audit retains all hardware/recovery/production gates; the owner's new instruction authorizes this investigation, not a milestone promotion or autonomous flash.

1. Archive every surviving incident record, including mounted pstore and both

   rotated battery journals; distinguish the six-hour owner observation from subsequent shutdowns and unsynchronized wall time.

2. Reproduce normal 04 idle without manually enabling C3: retained battery rows

   show zero C3 entries and PERI blocker 0x400. Resolve actual clock ownership.

3. Qualify staged PM and use the installed post-thaw sampler to localize the

   known full-suspend failure. Fix and verify same-boot Power/RTC restoration.

4. Only after reliable full resume, enable owned/inhibited product auto-suspend

   and measure 5 min, 30 min, 2 h and overnight voltage/SOC without calibration hacks.

5. If kernel changes require installation, seal one tested BOOTIMG/Y2ROOT

   successor preserving Y2DATA for owner flash. Software tests cannot satisfy the remaining physical/endurance gates.

Status: IN_PROGRESS / PARTIALLY READY. Tracking #28/#29/#31/#32/#33/#34, historical #30 power work retained despite its closed issue state. Current findings and evidence limits: deep-sleep report.

The first accumulated 30-minute attempt was cut short by USB insertion after two verified RTC segments and a third successful USB-triggered return. RTC auto-disarm is now observed before cleanup; the full measurement is repeating. SOC source review additionally finds suspend time excluded from estimator elapsed time. The narrow BOOTTIME correction preserves all calibration and protection policy, exposes rather than hides loss, and joins the same unsealed successor.

Physical update: normal battery idle after USB detach enters C3 (55.82% CPU0 over 300 s), while the retained cold battery boot had USB bit-10 veto and zero C3. Five staged PM tests pass on untainted 04. One real entry returns through device resume then stops at TASKS_THAWED and resets; the three-second sampler does not run. Early-boot PM/pstore archive now physically verified. Next step is to check the bounded context-ownership diagnostic: the recovered RTC-only trace shows core and platform both saving CPU/GIC context, with the platform save overwriting enabled IRQs after CIRQ masks them. Source removes that second owner and defers CIRQ replay until after core restore. Same-boot completion and Power/RTC attribution remain physical gates. No flash performed.

Latest narrow progress: the context diagnostic completes real RTC suspend and same-boot restoration on 404ef43c..., with original IRQ masks, filesystem progress, Reborn and five-second PCM playback verified. Repeated cycles use normal console suspend next; the intermittent device-phase wait and clean successor qualification remain open. Runtime USB source now also parks its detached monitor on verified absence and wakes on charger cable/source events; ARM compilation and native fault/event checks pass, physical reduction pending.

Current physical count is six complete diagnostic resumes on one boot: five RTC and one independently attributed Power wake (Power IRQ 24→26, RTC fixed at 5). The owner sees normal display return; COMPLETE/error 0 and zero ext4 errors persist. Four RTC plus the Power cycle use normal console suspension. Admit standalone five-minute battery-only suspend measurement under the refreshed audit. External-module taint, the earlier intermittent device-phase wait and clean-image/endurance/product-policy gates remain explicit; no promotion. Product source preparation begins with a database queue barrier after existing scanner/background quiescence, ensuring a pending transaction cannot be hidden behind a paused UI. Installed 04's automatic-sleep guard remains unchanged.

Latest measurement: 7/7 actual same-boot diagnostic returns, six RTC/one Power. Five-minute residence 305.85 s, voltage 3.983422→3.950244 V, SOC 76→76; zero unexpected wakes. Recent charge relaxation and taint remain explicit. The installed ten-minute PCM timer remains for bounded battery observation; product maintenance wake/re-entry must keep display/radios asleep and count all awake gaps. Implement delayed product sleep and owned inhibitors on the corrected kernel capability; installed 04 cannot be enabled by this userspace. Thirty-minute runtime baseline and longer accumulated-suspend tests remain.

The missing 30-minute runtime baseline is now retained: 56.72% C3, 32.30 mV voltage loss and provisional SOC 91→85, with no suspend request. Next physical boundary is 30 minutes accumulated full suspend under the matching diagnostic, with explicit RTC segments/awake gaps; 2-hour progression waits for that result. A post-resume Reborn observation-generation defect also leaves its battery view stale while the safety daemon remains fresh. Source now cancels that abandoned pending request without changing any SOC mapping or sample cadence.

Build admission: freeze the coherent source pair in isolated worktrees for one DEEPSLEEP-FINAL-01/root v1.31 successor, then perform full checks before sealing BOOTIMG/Y2ROOT only. No flash or readiness promotion. The diagnostic 30-minute trial is running independently; source identity must not change with later measurement/report updates. Clean built-in and product-policy qualification remain dependent on owner installation.

USB candidate handoff boundary — 2026-10-04

One preserving USB candidate is sealed at out/y2linux-platform-v1-usb-candidate/. Compiled Linux08019df / Reborn51d477b; BOOTIMG SHA256 b1b9dc1ec000159cb9d9c75327ab6d60aa32f29ffe476ab8450f7850e76fc9c4, Y2ROOT SHA256 cf0a4eccbd30cb556bb28d3ca7d145a8a75e968e24b1fbaf70d31b3dcaed9c52. Package/root/checksum validation and449-record source/license seal pass. The qualification harness adds checked user-partition/SD admission, every before/after hash, process CPU/RSS and actual frequency diagnostics; its6 native guards pass. Later docs/harness revision is explicitly separate from the compiled source pair.

Every standing coverage row retains the preceding software-sealing admission: CONFIRMED bounded UART01 boot/core/display/input/storage/audio/CPU/C1/C2/C3; CONFIRMED retained Baseline02 installed identity only, boot27e6d684…/taint0; PARTIAL new-image/cold-boot/full-suspend/endurance, USB/radio/desktop/performance, power/thermal/electrical/firmware/calibration/backlight/watchdog/wake and RAM/ recovery/update/security/services/time/identity/privacy/release gates (#27/#28/#29/#31/#32/#33/#34); UART crash capture UNKNOWN, host/VBUS PHYSICAL_GATE, FM excluded. MTP adds no block export or protected/memory/charging policy change. USB platform remains MORE_WORK_REQUIRED; all issues stay OPEN remotely.

Shared-lock refusals leave current speed, BEFORE/AFTER throughput and transfer CPU/IRQ/memory unmeasured. No new hardware result is inferred from software checks. Next exact action: owner Download Only of BOOTIMG and ANDROID/Y2ROOT while preserving Y2DATA/protected partitions, then the attached locked transfer harness and desktop/ ten-session/mid-copy/eject/library/playback/PIO acceptance. No flash or push; owner main/parallel subsystem work stays untouched. Full USB report and structured evidence retain the scope.

USB software-qualified candidate sealing admission — 2026-10-04

The source pair Linux 08019df44c8a7fbc249fc49a5a41f7ac906429ba / Reborn 51d477b2f30edfc90418d8d0c3fa5fad8de6f581 builds fresh kernel6.18.0-y2linux-platform-v1-usb, rootv1.22, release1.0.0-usb-candidate.1. 411 production checks (406 pass/5 minimal-host dependency skips with separate 27 filesystem/GIO and3 ALSA checks), 79 focused tests,240 Reborn tests/fmt/clippy, ARM/Cortex-A7/QEMU, config/DT/modules/ ABI/dependencies/inventory/legal checks pass. Final harness guards pass before packaging. No flash/push or physical promotion; detailed USB report.

Every coverage row is refreshed: boot/display/input/storage/audio/CPU/timers/ clocks/C1/C2/C3 remain CONFIRMED only within retained UART01 bounded proof; installed Baseline02 identity is CONFIRMED by another locked session’s saved receipt, boot27e6d684…/taint0; USB product implementation is SOFTWARE_VALIDATED but USB/desktop/large-transfer/PIO-disabling/ten-session acceptance is PARTIAL #27/#28/#31/#33. Radio/firmware/calibration/thermal/battery/charger/backlight/ watchdog/wake and new-image cold boot/full suspend/endurance remain PARTIAL #28/#29/#31. RAM/reservations/recovery/update/security/services/time/identity/ privacy/release remain PARTIAL #28/#32/#33 with unchanged memory/protected-data contracts; measured MTP memory remains missing. UART crash capture UNKNOWN, host/VBUS PHYSICAL_GATE, FM excluded. Existing issues stay OPEN remotely.

Three further scratch-baseline attempts remain NOT_RUN_DEVICE_LOCK_BUSY. No USB speed, throughput, transfer CPU/IRQ or host discovery result is invented. Proceed to seal only BOOTIMG/Y2ROOT with accepted UART01 fallback, source/legal archives, evidence and the single guarded qualification harness. Y2DATA and all protected partitions remain preserved. Subsequent qualification-only docs/harness revisions are separate from compiled identity. After owner installation, exact-image host/file/library/eject/performance acceptance is the next smallest blocker; verdict MORE_WORK_REQUIRED. Parallel owner-main integration work remains untouched.

USB candidate freeze: latest retained installed identity — 2026-10-04

The parallel integration review provides a newer retained locked hardware observation: installed Baseline02, Linux 8584ccd8 / Reborn b71b4688, rootv1.21, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint0. The saved Bluetooth identity receipt hash eb2fd0f0377e25c5a848b6aa02e9cabe3bf6af4a2089ed146670d36f6999f07d matches its evidence inventory. USB work still cannot obtain the physical lock; this is retained evidence, not direct new USB observation or CPU cold-boot proof.

Every coverage row in the preceding admission remains unchanged in classification: boot/display/input/storage/audio and C1/C2/C3 retain bounded UART01 CONFIRMED proof; Baseline02 installed identity is CONFIRMED while new cold-boot/full-suspend/ endurance remain PARTIAL #28; USB/radio/desktop/performance remain PARTIAL #27/#28/#31/#33; firmware/calibration/power/electrical/thermal/wake and security/ services/time/RAM/recovery/update/production remain PARTIAL #28/#29/#32/#33; UART crash capture UNKNOWN, host/VBUS PHYSICAL_GATE, FM excluded. The 64 MiB responder bound and unchanged protected partition/memory/charger policy persist.

The first ARM attempt exposed legacy g_cdc module linkage to unexported boot policy symbols; production now uses only configfs for both selections. The kernel build passes. Supervisor restart/unbind and parent-death FD release, and correct space/read-only error receipts are included before final candidate freeze. Resume the same fresh build workspace for the coherent source pair; publish only one final preserving candidate after all software checks. Parallel owner-main planning edits were read for evidence and remain untouched/unmerged. No flash/push/closure.

USB MTP architecture and candidate-build admission — 2026-10-04

The owner-authorized desktop pass selects MTP + ECM, with disconnected-session ACM + ECM for advanced use. Kernel/root scope changes are now explicit: configfs/ FunctionFS, a bounded pinned C responder, Linux-only user-file views and a narrow PIO progress fix. No MSC, block handoff, filesystem migration, FIFO/memory layout, charging-current or radio-policy change. The USB report records rejected alternatives and host/product limits. Source/native tests cover actual responder transfers/descriptors and data-integrity failures; software verification precedes one fresh ARM/package candidate.

All standing coverage rows are refreshed against retained UART01 Linux e9e8d63 / Reborn b71b4688, boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint0; fresh access remains refused by the shared nonblocking device lock. The retained receipt remains real hardware authority; new protocol/speed/throughput/host behavior is NOT_RUN.

Existing issues remain OPEN and unchanged remotely; no milestone closes. The next smallest boundary is one fresh kernel/Buildroot/Reborn ARM candidate and config/DT/ modules/ABI/source/legal/package checks, only BOOTIMG/Y2ROOT with Y2DATA preserved. Then owner installation enables exact-image desktop, ten-session and integrity/ performance harness acceptance. No flash or push is authorized for this agent.

USB desktop file-transfer activation audit — 2026-10-04

Owner explicitly authorizes implementation/research/performance and a single preserving candidate on codex/platform-v1-usb; no flash or push. Entry source Linux b5aea93 / Reborn b71b4688, isolated sibling worktrees. GitHub #27/#28/#31/ #32/#33/#34 refreshed read-only and remain OPEN; this local branch does not alter parallel work or remote milestone status.

Real hardware authority is retained UART01 Linux e9e8d63 / Reborn b71b4688, boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint 0: physical receipt. Fresh access was refused by the shared nonblocking /tmp/y2-platform-v1-device.lock; availability is retried in bounded batches. No new physical result is inferred.

Next smallest boundary: choose a file-safe protocol with protected partition exclusion and current endpoint budget; implement typed product/library integration, automated failure tests and bounded current-image scratch baselines. New kernel/ root changes require coherent implementation and one ARM/package candidate before owner installation. Y2DATA is preserved, never reformatted and never concurrently mounted as host block storage. No milestone closure or physical-ready verdict.

POWER01 full-sleep failure localization — 2026-10-05

Five staged tests PASS. The actual RTC boundary FAILS: reset-return reaches codec register-cache replay, then the backstop produces boot c884ae1c. Add bounded sleep-only cache/I2C breadcrumbs alongside the measured idle DAC/supervisor corrections. No repeated failed entry, automatic sleep or readiness promotion. Playback after full wake must be tested even if idle-bias off causes the initial resume callback to stay suspended.

POWER01 measured idle follow-up — 2026-10-05

Owner controls/display PASS; detached five-minute runtime C3 62.6446%, 9765 clean returns, same boot/taint zero. The idle boundary audit admits board-local DAPM idle-bias-off for the observed closed-PCM codec Standby and a measured event-driven connectivity supervisor follow-up. Preserve USB02, charging limits, input responsiveness and sole reconnect owner. Exact target playback/rail restoration, full RTC/Power resume and endurance remain open; ordinary autosuspend and platform readiness are not promoted by runtime C3.

Installed POWER01 physical continuation — 2026-10-05

Owner installation is complete: exact Linux fedc005 / Reborn ccf450a, POWER-FINAL-01, boot 69d8df3a, taint zero; read-back runtime idle PMIC slots 5–7 are correct. PC ECM/MTP and SDP 450 mA charging are available with fault zero. Follow the post-install gap audit: healthy display/USB/product smoke, matched idle/radio/audio/charging windows, then safe longer idle and staged/backstopped full resume. Overnight battery, termination, RTC/Power same-boot sleep and battery health remain open. This supersedes OWNER_INSTALLATION_NEXT only; no platform readiness promotion.

Power handoff and continuation — 2026-10-05

The fresh hardware handoff audit records the same WARN-tainted USB02 boot, real incident voltage-floor shutdown, seven power windows, restored original components and a successful 30-minute PC charge trend. The single preserving POWER-FINAL-01 pair is built, software/package validated and source sealed from Linux fedc005 / Reborn ccf450a. Yesterday's USB work and all electrical/data/memory guards remain. Owner installation is the next smallest blocker; the agent does not flash or push. Platform 1.0 remains PARTIAL. Continue with exact installed identity and clean display/PMIC admission, the same A–J matrix, near-full charging, staged and actual same-boot RTC/Power wake, and five-/30-minute/two-hour/overnight acceptance. No automatic suspend, endurance, battery degradation or readiness claim precedes those results.

Power candidate preparation — 2026-10-05

The build-boundary audit retains actual USB02 low-voltage shutdown, charging/C3 slot defect and frozen display evidence. Matched MTP and Bluetooth polling improvements are proved; kernel, display, Reborn idle and endurance need the owner-installed correction. Prepare one fresh preserving POWER-FINAL-01 pair with all software/package gates. No physical/public acceptance follows packaging. Resume with exact identity and warning-free display, then the same A–J power states and staged/full sleep tests. No memory, charging limits, USB transport foundation, DATA or recovery changes.

Runtime PCM slots — 2026-10-05

The slot defect audit records live zero-address slots5–7 and matching stock PCM slot5/6 requests. Checked initialization to the current VPROC bank/existing1.15V floor and a fail-closed C3 prerequisite are admitted. No voltage/current limit or arbitration changes. Physical C3/charge/endurance and full resume remain open until owner installation.

First display takeover — 2026-10-05

The ownership audit admits checked first-only OVL stop after DSI parking, so normal Linux config can reset stale loader state before starting scanout. Display timing and C3 taint guards stay. Native/build validation and owner-installed physical proof are separate. No full-sleep/readiness promotion follows a source repair.

Charging/idle boundary — 2026-10-05

The charging/C3 audit records a protection-register failure with C3 active and a successful five-minute DCP charge with C3 quarantined. Conservative charger/watchdog DORMANT exclusion is admitted; C1/C2, limits and guards stay. Causality and the DRM warm-boot warning still require localization. Software polling fixes undergo matched live trials; full suspend/endurance remain open. No platform readiness promotion.

USB02 overnight power investigation — 2026-10-05

The admission audit binds current hardware to installed Linux0cb7792/Reborn2444e0f, USB02/rootv1.26, and admits isolated power branches from that exact pair. Yesterday's USB work remains the foundation. Voltage decline and orderly low-voltage shutdown are observed; an initial genuine full charge and pack capacity/current are unproved. Current C3 production policy is Enabled/-1 but active radio/USB ownership blocks the observed window. Quantified userspace timed waits and shared-directory notifications are the next localized fixes. Radio-off/detached counters, staged/real same-boot sleep and charging/long endurance remain separate gates. Platform 1.0 remains PARTIAL; no flash, push, main merge or closure follows.

Baseline01 physical pause / USB diagnosis next — 2026-10-04

The standing audit uses exact installed Linux98daf3f/Reborn9cd12bae, original bootef2c8f2f/taint0. Smoke, all-core C1,18hotplug transitions,five OPP/voltage checks,CPU0-only parking, C2 4,857entries/85.4572%/zero restores failed and bounded DATA/SD hashes pass. C3 remains unqualified: owner restart interrupted a deliberately USB-off topology wait; there is no recorded armed/entered cycle or proved C3 crash. The host helper now waits for parking with USB attached and caps offline time at180s;30 targeted host tests pass without changing installed firmware.

Owner's earlier post-flash USB disconnect report stays open. Verified persistent host static-IP/MAC/autoconnect fixes one reachability defect, not all reconnects. Recovered boot9d4d0245 has DRM WARNING/timeouts and automatic C3 safely Skipped. Next smallest gate is exclusive read-only USB/boot diagnosis after the separate wired-audio session finishes, as the owner requests. No new CPU/transfer/full- suspend stress until resolved. Baseline02 RTC FAIL, separate new-image Power/RTC, endurance/electrical, peers/audio/desktop matrices and distribution remain open. Readiness is PARTIAL_INTERRUPTED; no feature, issue or milestone is promoted. Images/233 IDs/OPPs/memory/partitions/fallback remain unchanged; no build/flash/ push/main merge follows from this host-only correction and evidence update. Wired precision is now measured: the MT6582 wired playback path carries at most 16 meaningful bits to the CS43131. The earlier DL1-monitor argument was withdrawn and replaced by direct evidence: absent later-SoC wide-format registers, fixed 16-bit DL1/DL2 memifs, 16-bit interconnect outputs and an on-device capture of the I2S serializer pad (low 16 bits of every 32-bit slot zero in every fixture). Wide S24/S32 hardware output is not available; S16 (44.1/48) is the platform maximum. See the evidence record and the scoped boundary correction. No unsupported format, candidate, flash or other subsystem promotion is admitted.

Baseline01 installed physical admission — 2026-10-04

Owner reports newest preserving firmware installed and authorizes its tests. Read-only pinned USB admission at 18:35 UTC verifies compiled Linux 98daf3f0220f4a1580bb4893094ee9f0e02362a5 / Reborn 9cd12baeeb70b204e08299bd4c91c8c35604c304, kernel 6.18.0-y2linux-platform-v1-baseline-01, root 2025.02.18-platform-v1.24, boot ef2c8f2f-56d9-418a-b8a8-2dfd329d0973, taint 0. Standing audit repeated before activating exact-image physical qualification; all 233 feature IDs, source selection, protected partitions, memory and OPP scope, fallback and open issue gates remain unchanged. Owner confirms wheel, display, audible local track and normal battery display, playback now stopped. Root/DATA/SD ext4 and Wi-Fi COMPLETED/DNS are observed Ready. C1 counters advance; C3 boot policy Enabled/-1/disable0 is now observed on the integrated boot, but C2/C3 entry/restore and after-full-resume remain unqualified. Current screen/radio/USB/topology guards correctly report unmet prerequisites.

Retained UART01 C1/C2/C3 remains historical exact-image evidence. Baseline02 RTC watchdog FAIL remains open; persisted sleep restore_failed refers to old boot 27e6d684, not a new trial. Power and RTC full wake are separate NOT_RUN on this image. Battery 4.039 V/estimated 88%, no current sensor: no battery-life claim. Storage/USB desktop, Bluetooth peer, audio fidelity/rates, graphics/input counted controls, endurance and current-image recovery still require scoped tests. Firmware/codec redistribution, source/notices and collector gaps remain open; owner-local testing does not qualify public distribution or close issues.

Next smallest boundary: smoke geometry/kernel/owner gate, then guarded CPU foundation under the shared nonblocking lease and independent pinned Wi-Fi. Stop mutation on identity/boot/taint/restore/storage failure. Broader USB/suspend follows only a passing gate; no flash, reboot loop, push or main promotion. Private admission and campaign receipts: owner out/platform-v1-physical/. Readiness: PHYSICAL_QUALIFICATION_IN_PROGRESS, overall acceptance PARTIAL.

Independent baseline final closure check — 2026-10-04

+ +Before ending the software-seal boundary, retained real hardware and owner scope +were reviewed again against the canonical copy. No hardware evidence changed: +UART01 C1/C2/C3 is source-bound CONFIRMED, integrated physical/endurance remains +NOT_RUN, Baseline02 RTC FAIL stays open, Power/RTC and electrical benefit remain +separate PARTIAL/UNKNOWN gates. All 233 IDs, active issue states, hardware/memory/ +OPP/DATA scope and distribution limits remain as in the preceding audit. No new +build, flash or unchanged ROM/recovery provenance audit is needed for this +receipt-only documentation update. Canonical software/package/source/checksum +verification PASS; next step remains owner-preserving flash and one smoke-first +campaign. No main, public release or physical milestone is promoted. +

Independent baseline software seal / physical handoff audit — 2026-10-04

Standing boundary audit repeated before canonical baseline promotion, using actual retained hardware receipts rather than integrated source claims. Selected owner source, C01–C34 dispositions and the 233-feature scope remain unchanged. Final compiled pair Linux 98daf3f0220f4a1580bb4893094ee9f0e02362a5 / Reborn 9cd12baeeb70b204e08299bd4c91c8c35604c304 is CONFIRMED by fresh clean build and software/package tests: 475 cases (465 passed, ten native-covered skips), 324 Reborn, 384 ELF files/1411 edges, 106 packages. No runtime correction or expanded hardware, memory, OPP, partition, battery or production subsystem scope was admitted.

Retained real hardware: UART01 e9e8d63/b71b4688, same boot fd955840…/taint 0, C1 WORKING, C2 84.3473%/zero clock restores failed, C3 21 guarded context-loss returns plus 3624 normal entries, total 3647 matching returns/ACK/CIRQ/timer restores and zero faults. Sixteen foundation source files/three idle functions directly match that physical implementation. Automatic new-image policy remains PARTIAL until cold boot and guarded runtime observation. Baseline02 boot 27→88 RTC watchdog recovery is FAIL: DEVICES_RESUMING/wake 0x20/PCM 596/reset 0x59325253, CS43131 resume entry without leave; inner stall UNKNOWN. Power and RTC are separately NOT_RUN on the final baseline. New rollback/substep source is not physical fix evidence. Ordinary product full sleep remains guarded.

Storage/radio/JPEG/DB/performance evidence retains exact earlier image/binary scope; integrated USB desktop, peer/coexistence, controls, wider AFE, recovery, endurance and battery/electrical benefit are PARTIAL / NOT_RUN or UNKNOWN. Firmware/optional codec rights and source/collector limitations remain PARTIAL; public release is NOT_READY, owner-local qualification proceeds separately. Retained issue snapshot #16/#27/#28/#29/#31/#32/#33/#34 OPEN; #30 historical CLOSED does not close full sleep. No remote issue or milestone was changed. No ROM, stock/recovery/source provenance campaign or device connection was repeated.

Next smallest boundary: seal one owner-directory preserving baseline and let the owner flash BOOTIMG+ANDROID only, then run the smoke-first single campaign. Verdict READY_FOR_PLATFORM_V1_PHYSICAL_QUALIFICATION, not physically qualified. No main merge, flash, push or milestone closure. Source/software CONFIRMED; integrated hardware/endurance/public distribution remain open.

Independent baseline finalization admission — 2026-10-04

Owner authorizes verification/reconciliation, fresh build/software qualification and one preserving owner-local baseline. No device access, flash, main merge, push or public release. Standing audit repeated before finalization/production identity change, using retained real hardware evidence. All eleven paired merge bases equal Linux b5aea937/Reborn b71b4688. Master compiled4c72466/dd73dcc; later tips are documentation-only. Owner dirt and all histories are preserved.

All233feature IDs/layout/OPPs/memory/protectedDATA/calibration scope stays fixed. C01–C34 source dispositions checked against blobs/diffs/contracts. Final identity and smoke-gated single coordinator with separate Power/RTC are needed. Retained issue snapshot: #16/#27/#28/#29/#31/#32/#33/#34 OPEN; #30 historical CLOSED, never full-sleep acceptance. No issue/milestone changed or closed. Next: fresh independent build/package/software, then owner flash and exact-image hardware. Finalization PARTIALLY_READY; no integrated physical promotion.

Platform 1.0 software exit / physical qualification preparation — 2026-10-04

The repeated boundary audit reconciles the233-feature scope, eleven completed source pairs, C01–C34 contracts, all eight active issues and real hardware evidence before this boundary. One sealed owner-local candidate is READY_FOR_PLATFORM_V1_PHYSICAL_QUALIFICATION. Compiled Linux 4c72466f2f27ddf9f0ca18ba8ddc4d89c52599d2 / Reborn dd73dcc3243045cd33e319dda2f76918ed37ea31; root 2025.02.18-platform-v1.23. Broad software/kernel/ARM/QEMU/ package/privacy/source gates PASS:466cases (456pass,10native-dependency skips, all10actual cases separately passed),324Reborn/fmt/Clippy,384ELFs/1411edges.

Next smallest blocker is the owner's preserving installation of this exact pair, then one A–J campaign. Integrated hardware/endurance are NOT_RUN; Baseline02 full RTC FAIL and UART01 C1/C2/C3 remain source-bound. Ordinary deep sleep/wideAFE/peer/codecs are not promoted. Public distribution is blocked pending firmware/codec permissions, corresponding-source/notices and collector limitations. All existing open epics remain open; historical#30 closure is unchanged and supplies no full-resume proof. No source/memory/layout expansion, main merge, push or master flash/device access.

Platform 1.0 production composition boundary — 2026-10-04

The repeated standing audit compares all233 features, eleven frozen paired sources, issues/owner decisions and exact hardware receipts before production composition. Source units are integrated on paired isolated codex/platform-v1-integration branches;324 Reborn workspace tests/fmt/strict Clippy and affected Linux tests pass. Proceed to one fresh full ARM/kernel/DT/modules/ABI/production/package/privacy/source gate, then seal one y2linux-platform-v1-integration-candidate. Its physical state begins NOT_RUN; existing subsystem candidates/measurements remain separate.

Retain UART01 C1/C2/C3, Baseline02 full RTC FAIL and restored boot88 integrity, temporary069415a performance proof and the native S16-only hardware ceiling. Reconcile F134's implemented MTP component without claiming mass-storage/RNDIS. No memory/layout/protected-data expansion, public distribution waiver or epic closure. Owner alone installs/flashes the reviewed preserving pair. One source- bound qualification coordinator will hold the existing nonblocking device lease, reuse affected subsystem helpers and retain pending human/equipment/endurance lanes. Readiness: INTEGRATION_IN_PROGRESS; product PARTIAL.

Platform 1.0 integration activation boundary — 2026-10-04

All eleven paired owner handoffs are now clean, source-frozen and reviewed. The standing boundary audit compares the233-feature target, exact source pairs, owner decisions/issues and real receipts before activation. Proceed with isolated paired codex/platform-v1-integration branches from Linuxb5aea937/Rebornb71b4688; Reborn main alone omits the known-good scanout release. Owner worktrees remain independently owned. Follow the C01–C30 dependency plan and targeted tests after each coherent source unit, then one fresh broad software, ARM/kernel/package/privacy/source gate and one integrated preserving candidate.

UART01 C1/C2/C3 remains protected. Installed Baseline02 RTC full resume FAIL and boot88 recovery are retained; new suspend/codec source is uninstalled. Performance069415a has a bounded source-bound temporary-app pass on boot88, with stock/data/policy restored. Neither evidence qualifies the future RC. Native wider audio, real peer/coexistence, counted input, endurance/electrical, exact-RC full restoration and distribution remain independent requirements. No memory/partition/protected-data scope expansion or epic closure; no main merge, push or flash. Readiness is READY_TO_INTEGRATE for source, PARTIAL for product.

Platform 1.0 parallel integration observation — 2026-10-04

PARALLEL_WORK_IN_PROGRESS. Eleven paired owners are working from Linux b5aea93 / Reborn b71b468. The master maintains the status, dependency/contract plan and release picture. Follow the current gap audit. Do not merge active owners, promote their candidates or create the final image before reproducible handoffs and conflict decisions. Integration branches will use the protected Baseline02 source pair; Reborn main lacks b71b468 and is not by itself the complete foundation.

New Bluetooth/Wi-Fi owner receipts identify installed Baseline02 on boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0. Older awaiting-installation text below is superseded for identity only. Retained UART01 proves bounded C1/C2/C3: 84.3473% C2, 65.2450% C3 and 3,647 checked returns with zero restore failures. Keep that architecture, five OPPs and automatic CPU0 boot policy. Baseline02 cold-boot/C3 qualification is not established by radio identity reads. Full sleep, high-resolution physical bit/rate proof, real Bluetooth peers, USB desktop behavior, endurance and distribution remain separate requirements.

Next: finish owner handoffs, reconcile overlapping service/scanner/media/Power contracts, repeat the boundary audit, then integrate coherent commits into codex/platform-v1-integration in each repository with targeted tests after each subsystem. After one fresh broad software/package gate, produce one y2linux-platform-v1-integration-candidate and one integrated physical campaign. No main merge, push, flash, hardware/memory scope expansion or epic closure is authorized by this observation entry. The existing 233-feature inventory remains authoritative; append exact integrated-RC evidence after integration without overwriting historical receipts or dropping required feature/distribution rows.

Later owner RTC full-sleep evidence on the same installed Baseline02 is FAIL: watchdog recovery changes boot 27e6d684 to 88b85fa3-2282-4d93-b0b8-818a105ddd0a, taint 0; retained journal stops at DEVICES_RESUMING (wake 0x20, full PCM 596, reset stamp 0x59325253), with no complete/exit proof. Direct scratch hashes and zero ext4 counters after reboot prove recovery only. New suspend source was not installed. Preserve prior UART01 C1/C2/C3 evidence; full resume and automatic C3 remain separate gates. Master archives the raw failure receipts offline and neither repeats the trial nor closes any milestone. Review the source-bound owner handoff before integration.

Read-only GitHub refresh on 2026-10-04 confirms #16/#27/#28/#29/#31/#32/#33/#34 OPEN and historical #30 CLOSED since 2026-09-15. That planning closure does not qualify physical full resume; the new failed RTC receipt keeps this release gate open. No external issue state or comment was changed by master.

Storage Platform 1.0 closing audit — 2026-10-04

This boundary reviews the owner's isolated storage pass before local handoff. Actual installed Baseline02 is Linux 8584ccd85f052f351fe51650b2348c83ca894ed4, Reborn b71b468860233faa0a42b8448ec5777fa952b8e3, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0. Final locked hardware receipts record clean scratch cleanup, mounted ext4 SD at SDR104, eMMC HS200, both clocks gated, zero PM mismatch/error and unchanged transport fault counters. The non-blocking lease was released normally before the reserved suspend lane; that lane's later qualification is separate from this pass. New source remains uninstalled: Linux implementation 2a603c1, Reborn source identity ee1c25a and playback/sidecar guard f726ae4. Affected Reborn app38/platform77/library28/ media22 plus architecture guards, denied-warning Clippy and pinned ARM build pass; final ARM ELF/dependency and QEMU pinned image pixels pass as emulation. See the storage report and compact physical/software receipt.

Fresh read-only GitHub refresh keeps #16/#27/#28/#29/#31/#32/#33/#34 OPEN and #30 CLOSED. No status mutation or milestone closure is warranted. Existing recovery/fallback, Y2DATA/protected partition, memory, supplies and firmware obligations remain unchanged. Final classification is PARTIALLY READY: local source/report handoff is ready, general Platform 1.0 qualification is not. Next boundary is paired Linux/Reborn/native integration with explicit source-key migration/media admission, then owner-supported repeated/absent boot, second-card and physical filesystem matrix. Do not infer hardware authorization from this planning row. No flash, push, autoformat, raw write or mounted-media reset.

Storage playback-source guard audit — 2026-10-04

Final paired-source review finds C19's saved queue contains old Track snapshots, and Runtime::load checks only whether a pathname exists. SourceChanged pauses an unavailable current track, but a later manual resume or queued decoder can reopen a same-named file on a different card. The validated CID-qualified source must be checked at decoder admission, with pinned source directory and regular file descriptors so mountpoint/file-entry replacement cannot redirect the open. Reject symlinks and subordinate mounts while walking relative components; retain the pins through native close. Extension-free file-FD header probing requires direct-versus-pinned metadata/PCM tests across all valid fixtures; sidecar admission uses the same regular-file and pinned-parent boundary. Review found the local artwork loader selects codecs by suffix, so extension-free sidecar FDs need a bounded 12-byte JPEG/PNG/RIFF-WebP signature fallback. Preserve existing named-file behavior, image limits, codecs/protocols and native ABI; validate identical direct/pinned pixels and affected native tests. The app must ship its matching native media library; old installed binaries remain unchanged.

This is within the owner's storage coordination for playback/media replacement and #32/#33; it does not change SQLite, FFmpeg/native ABI, DSP, sink configuration, MMC modes or memory/idle policy. Installed physical hotplug remains bounded old- source evidence; the new paired guard requires host fault tests, app/platform checks and ARM build, then later target acceptance. No old queues are rebound to unproven cards. Update the storage report before closing this review; no broader release/milestone acceptance is inferred.

Storage removable-mount gate audit — 2026-10-04

A source review during the authorized filesystem pass found that `y2-media mount` identifies the card but does not run the existing read-only checker. The owned-host FAT32 test proves that a dirty clean-unmount bit produces a nonzero fsck.fat -n result without changing the image. Current physical media is clean ext4, with successful explicit e2fsck -f -n and remove/reinsert; no dirty user media is created. Adopt a bounded read-only preflight before SD mounts, reject checker faults/timeouts/source changes, and leave automatic repair/formatting disabled. Internal root/data boot gates stay unchanged.

This is a localized production-admission change within #32/#33, not hardware, memory or filesystem-format expansion. Read-only checks may delay mounts or refuse media requiring journal recovery; that availability limit must remain visible. Existing MT6582 modes, fallback, runtime PM and C2/C3 guards remain unchanged. Paired claim/source-ID changes still require Linux/Reborn integration; new source is software-tested, not installed by this pass. Proceed with focused media/checker tests and update the receipt.

Storage Platform 1.0 optimization admission — 2026-10-04

Owner requests an isolated storage/filesystem pass from Linux b5aea93 and Reborn b71b4688 on codex/platform-v1-storage; no main-worktree edits, flash or push. This narrow review precedes implementation and physical measurement. Retained hardware authority: candidate3's 200 MHz SD data CRCs and clean bounded 100 MHz reads (SD characterization); UART01's clean eMMC/SD SHA256 after real C2/C3 returns (physical receipt). Candidate4's stock pad/card-driver changes are included in current source, but its original receipt is software-only. A read-only identity probe saw Baseline02 running at HS200 and SDR104 with zero counters, boot 27e6d684-3794-4490-95db-a489027cd4f4; the lock acquisition raced another session, so no qualification is claimed from that probe and all further physical access requires the non-blocking device lock.

Read-only GitHub refresh: #16/#27/#28/#29/#31/#32/#33/#34 OPEN, #30 CLOSED; no external status mutation or milestone closure. Preserve storage addressing, Y2DATA, recovery/backup/calibration obligations, exact fallback, memory map, regulators and idle guards. No raw writes, autoformat, mounted-media reset or SQLite optimization. Proceed with the smallest justified source fixes, targeted regressions and safe scratch measurement; owner is available for SD interaction. Final decisions and unobserved gates belong in the storage report.

Wi-Fi optimization source closing audit — 2026-10-04

Scoped source pass closes on Linux 7056977 / Reborn 2cdb9cc; physical acceptance remains PARTIAL, with no phase/milestone promotion. Latest Wi-Fi-owned device receipt is Baseline02 boot 27e6d684…, taint 0 before/after baseline. GitHub #28/#31/#34 refreshed read-only at closing, all OPEN. Exact results and sanitized receipt: Wi-Fi optimization.

Next smallest boundary is an available lock for the exact integrated readiness, DHCP action and Reborn after sweep, preserving USB control wake and supplicant socket wake. Isolate the iperf reset, then AP-loss/cold-boot/power/coexistence coverage. No source-test result substitutes for those physical gates. Retained recovery, memory, packaging, calibration and distribution boundaries stand.

Wi-Fi Platform 1.0 optimization admission — 2026-10-04

Owner authorizes isolated codex/platform-v1-wifi worktrees, bounded shared-lock Wi-Fi tests, localized fixes and regression checks; no flash, push, parallel branch merge, memory/partition change or Bluetooth redesign. Starting source: Linux b5aea93, Reborn b71b468. Read-only USB SSH under nonblocking /tmp/y2-platform-v1-device.lock observes installed Baseline02, kernel 6.18.0-y2linux-baseline-02, compiled Linux 8584ccd, Reborn b71b468, root v1.21, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0. This supersedes the earlier awaiting-installation statement for identity only; this read does not qualify all Baseline02 cold-boot or CPU gates. Current lease, route and DNS are ready, automatic policy requests CAM/off, screen is blank. Release inventory agrees with compiled source. GitHub #28/#31/#34 refreshed read-only, all OPEN; no remote status mutation or milestone closure.

Next smallest boundary: capture current baseline and exact power-mode semantics, fix only demonstrated polling/readiness deficiencies, then repeat affected tests and bounded measurements with reversible RAM-only tools. Association, DHCP, route and successful DNS remain separate; unknown status never becomes a real disconnect. See optimization report for the eventual closing receipt. Existing recovery, calibration, source and ROM provenance is unchanged and is not recaptured.

Bluetooth optimization boundary — 2026-10-04

Owner explicitly authorizes the Bluetooth implementation/research pass in isolated codex/platform-v1-bluetooth worktrees from Linux b5aea93 / Reborn b71b468. Read-only locked SSH identifies installed Baseline02, Linux 8584ccd8, Reborn b71b4688, kernel 6.18.0-y2linux-baseline-02, root v1.21, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0. This supersedes the *installed identity* in the older UART01 receipt; it does not constitute the separate Baseline02 cold-boot/C3 qualification. Retained UART01 C1/C2/C3 evidence remains bounded to its own image/boot. No headset is available, as confirmed by the owner.

F102/F103/F104 and #31 remain PARTIAL: the controller/BlueZ/BlueALSA are observable; peer pairing, audio, reconnect and endurance cannot be promoted by this pass. F105/F108/F229 retain peer/audio/coexistence evidence gates. #28/#31/#34 remain OPEN on read-only GitHub refresh. Idle measurement with Bluetooth off shows 50.37 Reborn Bluetooth thread context switches/s and repeated 2s BlueZ snapshots; these justify localized event/wakeup fixes, not a new kernel power architecture.

Proceed with signal-based observation, bounded-intent/recovery fixes, private-bus regressions, ARM userspace builds and temporary, restored runtime tests only when the nonblocking shared device lock is held. Preserve one reconnect owner and FFmpeg → shared DSP → final sink conversion → BlueALSA. No memory, firmware, calibration, protected storage, image packaging, flash, push, merge, release or milestone closure. The next acceptance boundary remains actual peer/audio and coexistence measurements on the final integrated artifact. Findings and final classification: Bluetooth optimization receipt.

Bluetooth optimization handoff — 2026-10-04

Localized control/wakeup/diagnostics implementation is software validated; the closing review and receipt separate physical Baseline02 measurements from uninstalled source. MORE_WORK_REQUIRED: retest changed binaries under the nonblocking device lock, then pair/reconnect/ SBC startup, loss/recovery, screen-off and Wi-Fi scan/transfer with an available headset. M5 #31/F103/F104/F105/F108/F229 stay open/partial; no hardware or release milestone is closed. Preserve CONSYS/BTIF/BlueZ/BlueALSA, shared DSP/final sink conversion, one reconnect owner and current codec policy. No kernel/power redesign or optional codec feature-completion project is activated.

Wired precision source boundary — 2026-10-04

The source-only diagram correction includes live post-conversion f64 volume and integer rounding. Low-bit proof is for unity user gain; decode/DSP/packing CPU benchmarking excludes Reborn's live gain/ALSA pacing. Future wider work must resolve that final quantization placement as well as MT6582 transport evidence. No gain, pacing, format, buffer or hardware policy is changed by this correction.

Completion-pass audit and exact research/implementation report record MORE_HARDWARE_EVIDENCE_REQUIRED. The complete admitted path remains S16_LE stereo16 meaningful bits at44100/48000 in32-bit I2S slots. This is not a proved silicon maximum. CS43131 accepts incoming32; exact MT6582 wider DL1 and interconnect programming is unresolved, and Reborn FLTP separately loses general full32 source precision. No speculative wider format is enabled.

Truthful kernel/platform diagnostics, quantization fixes, WAV/FLAC low-bit fixtures and CPU benchmark are source/ARM/QEMU validated. Installed Baseline02 was read under lock; subsequent lock contention prevents fresh playback and A7 CPU measurements. New diagnostics are compiled, not installed. #29 remains PARTIAL for high resolution; #16/#27/#28/#31/#32/#33/#34 keep their existing gates and OPEN status. Next: exact wide-fetch/interconnect evidence, then locked existing-S16 measurements and an analyzer capture for any admitted wider mode. Source handoff closes no hardware milestone or production release gate.

Wired precision completion admission — 2026-10-04

Owner explicitly authorizes isolated source implementation, exact MT6582/stock research and bounded locked physical audio tests; no flash, push or merge. Linux starts at b5aea93, Reborn at b71b468, on dedicated codex/platform-v1-highres-audio worktrees. Read-only device admission under /tmp/y2-platform-v1-device.lock observes installed Baseline02 kernel, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0 and idle Y2Audio. Exact direct constraint intersections admit only S16_LE stereo 44100/48000; S24_LE/S32_LE and 88200/96000 are rejected. This is driver admission evidence, not silicon maximum or low-bit proof. Raw receipt: out/highres-audio/validation/device-admission.txt. Retained UART01 and native headphone evidence remains authoritative for its bounded tests; new-image general CPU qualification is outside this pass.

Open issues #16/#27/#28/#29/#31/#32/#33/#34 refreshed read-only. No remote status change is warranted or authorized. High-resolution implementation is ready only after exact fetch/packing/interconnect documentation or stock programming establishes every meaningful bit. If unavailable, retain the safe S16 contract and publish the unresolved hardware boundary without claiming a 16-bit silicon ceiling. Next: exact vendor source and targeted stock analysis, software low-bit fixtures, truthful diagnostics and locked baseline tests.

Bounded S16 test admission: a second locked read verifies the installed Linux 8584ccd85f052f351fe51650b2348c83ca894ed4 / Reborn b71b468, same Baseline02 boot/taint 0, suspended AFE, no PCM owner and headphone disabled at -60 dB. Exact stock HAL width/fetch functions return -ENOSYS; mixed-tree ASoC wide fields are unselected by its MT6582 defconfig and lack compatibility proof. No wider mode is admitted. New kernel diagnostics and S24 software rounding are host/ARM compiled, not installed. Ready for bounded existing S16 DMA/rate/runtime-PM observation and offline three-format A7 CPU measurements under one non-blocking device lock. No source/memory/production scope expansion.

Media stock physical receipt reconciliation — 2026-10-04

The receipt reconciliation audit adds actual stock scanner evidence from the separate storage owner's device slot: Baseline02 Linux 8584ccd8 / Reborn b71b4688, boot 27e6d684-3794-4490-95db-a489027cd4f4, closing taint 0. DATA/HS200 1000-track initial/incremental 5526/393 ms, peak 10656 KiB; SD/ext4/SDR104 5495/401 ms, peak 10936 KiB. Each initial scan completes 1000/zero failures, each rescan reuses all tracks; storage fault deltas remain zero. Original private receipts and hashes are summarized in the media report.

#29/#32/#33 classification: CONFIRMED bounded physical stock scan/reuse, CONFIRMED candidate software, UNTESTED new-candidate physical scan/concurrent playback. Remaining coverage/OPEN epic status and backup/calibration/recovery obligations from the prior closing review are unchanged. The next device slot belongs to suspend qualification; no device operation, build, flash or push occurs in this documentation reconciliation. Integration remains ready with limitations; next media gap is a matched-candidate guarded SD/playback receipt. No physical speedup is inferred from host/QEMU or stock-only measurements.

FFmpeg, metadata and library software closing review — 2026-10-04

The owner-authorized Reborn media subpass is software-ready with limitations, following the closing gap audit. Reborn a22107c + 43a695c retain FFmpeg/float DSP/SQLite ownership while repairing metadata/gapless/artwork correctness and bounding scan work. The complete media receipt records 252 host tests, ARM release, 32+30 QEMU tests, native ownership stress, component/ABI checks and 100/1k/10k benchmarks. Host 10k initial scan median 7408→1084 ms, unchanged 191→130 ms, process peak 66488→53720 KiB.

Retained UART01 Linux e9e8d63 / Reborn b71b4688, boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint 0, governs bounded installed playback/storage/CPU evidence. Baseline02 SDK/QEMU results do not qualify this new media candidate on SD. The nonblocking device lock was busy with the storage lane; no physical device operation, flash, installation or push occurred here. Parallel owner-reported SDR104 results do not substitute for media measurements.

All eight active epics remain OPEN after read-only refresh. No hardware/memory/ partition scope changed and no production waiver is inferred. Software integration is READY; candidate physical qualification is PARTIALLY READY and requires its existing lock/admission controls. PLATFORM_V1_READY_WITH_LIMITATION applies only to this software pass. No engine or SQLite rewrite is justified. The closing documentation audit does not build/flash or repeat unchanged ROM/recovery provenance.

Metadata cache revision boundary — 2026-10-04

The media pass found that existing indexed rows otherwise retain the old AIFF artist mapping, album-artist alias behavior and Opus duration after extractor corrections. Before this localized database extension, repeat the standing review against retained UART01 and Baseline02 evidence. Hardware coverage and image identities remain as recorded in this pass's admission; new code is host/ARM/QEMU qualified only. The shared physical device lock is busy with separate storage qualification, so no physical access occurs here. The owner's parallel SDR104 result is separate evidence, not a measured media-pass result.

Add one default-zero per-file extractor revision column, retaining public SQLite schema/user_version 1 and all existing row IDs/columns. Old rows reparse once; successful rows record the revision, and malformed rows cannot force the rest of the source to reparse forever. No raw artwork/tag corpus is stored, no SQLite architecture replacement, data reset, hardware, memory-layout, partition or package change is authorized or performed. The owner explicitly authorized metadata correctness and incremental ingestion; this is necessary cache invalidation within that scope. Validate migration, one-time reparse and subsequent reuse, alongside the requested final software checks. #29/#32 remain PARTIAL for media acceptance pending the final receipt; #16/#27/#28/#31/#33/#34 retain their independent OPEN gates after read-only refresh. No external status mutation, release waiver, provenance repetition, flash or push.

FFmpeg, metadata and library pass admission — 2026-10-04

Owner explicitly authorizes the Reborn Platform 1.0 media pass in isolated codex/platform-v1-ffmpeg-metadata worktrees. Entry source: Reborn b71b4688, Linux b5aea93. Preserve file → FFmpeg → media abstractions → float DSP → sink conversion; no engine replacement, kernel/hardware/partition change, flash or push. Parallel radio and wired-audio work remains separate.

Retained UART01 real hardware evidence identifies Linux e9e8d63, Reborn b71b4688, boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint 0. It proves bounded playback, storage and C1/C2/C3 behavior; it does not measure large-library metadata throughput. Baseline02 provides ARM/QEMU/software evidence, not new-image cold-boot acceptance. GitHub #29 and #32 refreshed read-only: OPEN; stale historical issue text does not supersede retained newer hardware receipts. No external issue status change is authorized or needed for this software subpass.

The software pass is ready to proceed. The smallest blocker is missing measured metadata/scan evidence, not hardware authorization from a planning epic. Existing backup/calibration/fallback, memory layout and production obligations remain. This documentation audit performs no build/device test or provenance repetition.

Library/database bounded native Y2 closure — 2026-10-04

This closing review supersedes only the earlier library target NOT_RUN label. Actual installed Baseline02 is Linux 8584ccd85f052f351fe51650b2348c83ca894ed4, Reborn b71b468860233faa0a42b8448ec5777fa952b8e3, kernel 6.18.0-y2linux-baseline-02, root 2025.02.18-platform-v1.21, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0. MemTotal 951824 KiB; Y2DATA rw ext4/eMMC. All six private native before/after runs (1k/10k/50k) preserve boot, taint and exact mount generation. The read-only observer after optimized scan 14 reads ext4 errors_count 0 on DATA, SD and root. Driver exits 0 after descriptor-relative owned DATA/SD fixture and temporary binary cleanup; shared device lock released normally. Installed app/database and storage modes are unchanged. No flash, push, service change or production database replacement.

The 50k fixture uses 25k files on DATA + 25k on SD, all SQLite databases on DATA; capacity admission includes bytes/inodes/reserves. The stock filesystem inode limit prevents a safe single-volume 50k fixture; no resize/format is authorized. 1k/10k use only DATA. 50k after artifact is Reborn c161404; before is archived b71b468 production code with current benchmark-only adapters. Transfers are SHA256-verified; checked-in sanitized results retain artifact identity and all median/p95/worst samples. See Reborn evidence/report. The target lacks /proc/self/io; target write bytes are unavailable, not zero.

All active issues/epics refreshed read-only 2026-10-04: #16/#27/#28/#29/#31/#32/#33/#34 remain OPEN. Existing coverage owners and retained failures remain; no external issue or milestone status is changed. Protected partitions, hardware/memory layout, recovery/calibration and standing installation decisions are unchanged. No unchanged ROM/recovery provenance, package or kernel checks are repeated.

The localized library/database pass is ready for source integration, with measured regressions and current 20k queue/8 MiB session limits explicit. The next smallest boundary is combining the separate storage and FFmpeg commits, resolving source ownership/reuse revision interactions, then targeted integration checks. This is not whole Platform 1.0 closure or authorization for flash/push.

Library 50k split-fixture admission — 2026-10-04

Locked read-only admission confirms installed Baseline02: Linux 8584ccd85f052f351fe51650b2348c83ca894ed4, Reborn b71b468860233faa0a42b8448ec5777fa952b8e3, kernel 6.18.0-y2linux-baseline-02, root 2025.02.18-platform-v1.21, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0. Y2DATA is rw ext4 on mmcblk0p7, 770.6 MiB, errors_count 0; RAM MemTotal 951824 KiB. Standalone private before/after 1k and 10k library runs pass. This is not full Baseline02 cold-boot/CPU/platform qualification.

The first 50k-file attempt failed with remote status 1 and cleaned its owned scratch; that wrapper failed to retain stderr, so its exact failing syscall is unknown. The storage owner reports only ~51071 free Y2DATA inodes and ~31982 on the current 512 MiB SD ext4 filesystem. 50k files plus ~2084 release directories cannot safely fit on either filesystem. Do not resize/format/mount/remap anything. This is a fixture capacity boundary, not permission to change production storage.

Proceed, after the queued live session releases the same device lock, with a bounded 50k collection split into 25k new private files on Y2DATA and 25k on SD. Keep all SQLite databases on Y2DATA. Pin both scratch directory descriptors, check bytes/inodes/reserve and mount/boot identity, clean only newly owned dirs. No existing media, installed app/DB, SD mode, services or hardware controls change. The separate storage physical sequence is finished and restored SDR104/HS200. Core media-removal/identity fault qualification remains separate; fixture sources are synthetic and benchmark exact mount generation is checked by the parent. The Reborn report must label this two-source layout and missing target I/O counters (/proc/self/io is absent), rather than claiming zero flash writes. Existing #28/#31/#32/#34 remain OPEN; no whole-platform acceptance is promoted.

Library/database host validation boundary — 2026-10-04

Repeat the admission audit before recording this local software result. Reborn b0b114e implements bounded fingerprint/presence scans and coherent keyset catalog loads; 9152005 preserves instrumentation-only baseline reproduction, eca41fc covers UUID-less CID compatibility. Schema v1, persistent indexes, WAL/NORMAL, 64-row write transactions, single worker, protected/data/memory-layout and installed images remain unchanged. The cache target is explicit 2 MiB; no new RAM layout.

Retained real hardware authority remains UART01 Linux e9e8d63 / Reborn b71b468, boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint 0. No new-image or physical performance pass is inferred from host timing. Baseline02 cold-boot qualification remains PHYSICAL_NOT_RUN. #28/#31/#32/#34 refreshed read-only again 2026-10-04: all OPEN; no external milestone/epic changes or new issue required.

The Reborn pass record and checked-in host samples retain median/p95/worst and observed regressions. The next smallest gate is locked scratch-only Y2DATA before/after timing/memory, without flash, production database or installed binary replacement. Source-ID producer changes belong to storage; library accepts legacy, UUID-less CID and UUID+CID IDs without auto-rebinding old queue snapshots to unproven cards. No unchanged ROM/recovery/source provenance, package or kernel qualification is repeated. This records software readiness, not whole Platform 1.0 closure.

Library/database performance pass admission — 2026-10-04

Owner authorizes isolated host profiling and bounded scratch-only Y2 tests for Reborn SQLite/library Platform 1.0. Reborn base b71b468860233faa0a42b8448ec5777fa952b8e3; Linux base b5aea93. Separate codex/platform-v1-library-db worktrees preserve main trees. FFmpeg metadata parsing belongs to another session. No flash, push, production database replacement, kernel change or milestone closure is authorized.

Retained real hardware authority is UART01: Linux e9e8d63, Reborn b71b468, boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint 0, bounded C1/C2/C3 and storage integrity pass. Baseline02 new-image/cold-boot qualification remains PHYSICAL_NOT_RUN. GitHub #28/#31/#32/#34 refreshed read-only 2026-10-04: all OPEN. No external status changes are needed for this localized pass.

Proceed with measured local library improvements within the existing worker and schema. Application memory is measured, not new RAM-layout authorization. Future physical runs acquire /tmp/y2-platform-v1-device.lock; no unchanged source/ROM/ recovery provenance is repeated. Closing report must distinguish host/ARM evidence from real-device and whole-platform qualification.

Platform 1.0 graphics pass — 2026-10-04

Evidence-boundary audit records the completed isolated pass on installed Baseline02, Linux 8584ccd8 and stock Reborn b71b4688, same boot and taint 0. Mali/Lima/Mesa/480x360 scanout and GPU PM remain preserved. Fresh screen-off has zero frames/GPU-active time, MFG and scanout clocks off; five retained-context wake cycles and brightness 4/16/32 readbacks pass. The temporary changed ARM JPEG decoder reduces median CPU 492.637→75.332 ms and peak RSS by 20,600 KiB for the 4096-square fixture; 68 Linux/238 Reborn tests, strict lint and ARM release checks pass. The tested SDK is FFmpeg 9.0.2, with exact native membrane hashes in the receipt. Select one reduced-JPEG implementation when combining with the FFmpeg/media pass; retain its aggregate-I/O/cancellation guards and qualify the resulting native ABI.

Graphics acceptance remains PARTIAL. Stock paused Home still presents 25 frames/20.138 s with 4.148 s GPU active residence; the parallel performance pass owns event-loop/radio/renderer correction. Integrate and accept that correction before closing static-redraw scope. Source/host and current-boot splash protocol evidence do not replace visual cold-boot/shutdown, long-off content, forced-context loss, electrical savings or endurance gates. Retained UART01 C1/C2/C3 remains scoped to its artifact; C2/C3 did not enter in this attached-SSH observation. Full system suspend belongs to another session; #28/#31/#34 remain OPEN after read-only refresh. The device state is restored and lock released. No flash, push, main change, memory expansion or release acceptance. Exact results and localized changes are in Y2-PLATFORM-V1-GRAPHICS.

Reborn performance pass closing boundary — 2026-10-04

The scoped Product v2 application performance pass is CONFIRMED on Baseline02 boot 88b85fa3-2282-4d93-b0b8-818a105ddd0a, taint0, with preserved baseline and exact candidate 069415a ARM SHA 7871ec67f6678c7f522b1b84da0f1e53ab1646e89bc6abb6b7eafe10c02286c7. See the closing gap audit and Reborn evidence.

Wheel KMS p95 is 59.93→17.52 ms and Now Playing 77.61→17.15 ms; all 1,000 steps match. Frame CPU p95 falls from 5.09–9.94 to 1.65–2.82 ms. Sampled static screens stop redrawing; Home CPU falls 4.96→1.88%, playing 87.95→11.21%. Four warm-start interactive p95 improves 779→499 ms. Supplemental repeated resume/pause, cover caching, private 10k fixtures, stop/reopen and active shutdown pass. Product v2 retains bounded pixel near-equivalence; no photon/analog/power, full-suspend, peer-radio or improved cold-boot qualification is promoted.

Installed b71 Reborn/data/paused screen-off state, original C3/governor policy and rw storage with zero ext4 errors are restored. Device leases are released; no flash, push, main-worktree change, production package or memory/partition change belongs to this pass. Read-only issue refresh keeps existing epics OPEN.

Next boundary: cross-owner source integration is PARTIALLY READY. Reconcile source epochs/prepare and inflight observations with fresh storage/FD admission; compose wired pacing with strict transport/format ownership and use the radio owner's event worker in place of the local powered-off timeout. Repeat relevant combined qualification after integration. The 10k single-album stress layout (34.36 ms), deeper libraries/queues, analog/peer/electrical evidence and the independent failed full-resume gate remain scoped gaps, not implementation or hardware authorization. This update is documentation-only.

Reborn performance matched-run readmission — 2026-10-04

The independent suspend owner released its lease after watchdog recovery, storage integrity checks and normal restoration. This pass independently observes Baseline02 Linux 8584ccd8 / Reborn b71b4688, new boot 88b85fa3-2282-4d93-b0b8-818a105ddd0a, taint 0, stock paused/screen-off Reborn, three tracks, root/data/SD rw with zero ext4 errors and unchanged automatic CPU0 C3/schedutil policy. See the readmission audit.

Earlier performance receipts remain exact old-boot evidence. The next boundary is a fresh preserved baseline followed by candidate b8c05db on the same new boot, then restoration and a closing audit. Full suspend remains PARTIAL and its failed resume is an independent release gate; no additional deep trial or production/flash/push/policy change is authorized here. Existing epics stay OPEN.

The subsequent same-native gain/write probe confirms a tiny partial-write spin in Reborn audio. Next localized step: bounded PCM-capacity waiting, preserving gain, formats, buffers, leases and accepted-frame accounting; validate source and rerun the exact candidate before closure. High-resolution/analog and peer Bluetooth gates remain separate; see the readmission audit for probe scope.

Reborn performance pass admission — 2026-10-04

Owner authorizes measurement and localized Reborn responsiveness fixes on branch codex/platform-v1-reborn-performance, preserving Product v2. Fresh read-only USB SSH under /tmp/y2-platform-v1-device.lock observes installed Y2LINUX-BASELINE-02, Linux 8584ccd8, Reborn b71b4688, kernel 6.18.0-y2linux-baseline-02, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0, and automatic qualified CPU0 C3 policy Enabled/budget -1/disable 0. This supersedes installation-pending information only for device identity and boot-policy observation; it does not substitute for full Baseline02 qualification. Retained UART01 C1/C2/C3 evidence remains scoped to its tested image.

Read-only GitHub refresh confirms #16/#27/#28/#29/#31/#32/#33/#34 OPEN. No remote status mutation, milestone closure, build image, flash, push, memory reservation, partition or production package change is authorized by this pass. Temporary executable/fixture runs are reversible hardware measurements; retain and restore the installed service and owner data. Next smallest boundary is a measured baseline before any optimization. Source/ROM/recovery provenance is unchanged and is not repeated. Report belongs in Reborn docs/validation/REBORN-PLATFORM-V1-PERFORMANCE.md.

Platform 1.0 input source handoff — 2026-10-04

Owner ended device waiting. Reborn e51a335 / Linux 95ed509 source and software validation pass; ARM userspace candidate exists, never installed/executed. MORE_WORK_REQUIRED for physical acceptance:100CW/100CCW,20/button, input->frame percentiles, candidate idle/wake/recovery and owner feel. First read-only Baseline02 capture had0events; subsequent input IRQ deltas0, while current four-core topology rejected C2/C3. Preserve qualified older UART01 CPU evidence without promoting this candidate. Lock released, helper removed, input reservation cancelled; no flash/push/main-worktree changes. End-boundary audit. Full input report. No milestone closed.

Platform 1.0 input software candidate — 2026-10-04

Reborn e51a335 / Linux input tooling 99087ba: source and software checks complete (257 workspace tests,37 input/40 UI, fmt/strict Clippy/ARM,5 Linux input cases). Userspace candidate out/platform-v1-input-candidate/ is sealed; kernel/DT/images unchanged. Boundary review.

Retained218 physical wheel pairs have no same-direction gaps below8ms; keep existing wheel filtering and acceleration. Current100CW/100CCW accuracy, 20/button, input->frame and owner feel remain pending a coordinated window. One read-only Baseline02 window (boot27e6d684, taint0) recorded zero events; owner replied after it ended. Idle wheel/I2C0/Power/keypad IRQ deltas all0; current four-core topology rejected C2/C3. CPU policy and installed binary unchanged, temporary helper removed, lock released for storage. No flash, push or physical milestone closure; #28/#34 remain OPEN. Full input receipt.

Platform 1.0 input pass — 2026-10-04

Input implementation and bounded physical observation are active only in the isolated codex/platform-v1-input Linux/Reborn worktrees (bases b5aea93 / b71b468). Admission audit preserves Product v2 semantic controls, the IRQ-driven APT32F driver and retained UART01 C1/C2/C3 evidence. #28/#34 remain OPEN after read-only refresh.

Next: map timing/recovery ownership, reproduce concrete failures, capture current installed events under /tmp/y2-platform-v1-device.lock, fix justified bugs and run targeted Linux tests plus Reborn fmt/clippy/workspace/ARM checks. Physical counts, latency and owner feel remain separate acceptance gates; the initially busy lock does not authorize concurrent access. Build one coherent software candidate if needed. No flash, push, main-worktree edits, CPU-policy changes or full-suspend implementation. Input receipt.

Full suspend failed-return boundary and codec correction — 2026-10-04

Verdict MORE_WORK_REQUIRED. The installed Baseline02 one-shot 12-second RTC trial records SPM EINT 0x20, full PCM length 596 and reset-vector stamp 0x59325253, but no COMPLETE/EXIT. Previous journal sequence 175 stops at DEVICES_RESUMING; phase-7 enter 7407 for CS43131 1-0030 has no matching leave through final sequence 7422. Wheel and both MMC callbacks do leave; their parallel progress does not prove codec completion. The inner codec stall is unknown. Boot changes from 27e6d684-3794-4490-95db-a489027cd4f4 to 88b85fa3-2282-4d93-b0b8-818a105ddd0a, with a retained 30-second backstop. This is failed return/reset recovery, never successful resume; decoded RGU cause was not captured. Power and repeated entries are stopped after this first trial.

A narrowly scoped codec PM source correction replays cached state/XTAL mask after supply-disable refusal without double rail references, and records force/rail/ delay/cache/XTAL substeps in the existing retained ring. It does not claim to fix the unidentified inner stall. Eleven focused fault/PM trace tests and a corrected isolated ARM Image/zImage/modules build PASS; audio/policy has 12 pass + one DTB skip. Prior 244 Reborn workspace tests/fmt/Clippy/ARM proof remains unchanged. No AFE/high-resolution audio, C1/C2/C3, voltage, clock, memory or protected scope change. New source is uninstalled; no flash/push/merge/issue update or milestone closure. Fixtures/listener cleaned, pm_test none/backstop 0/alarm empty, stock software/policies preserved, lease released normally to performance. Next is a coherent owner-installed instrumented candidate, then one checked RTC return, Power wake and full subsystem/cycle qualification only after success. Exact report and bounded evidence.

Full suspend physical entry boundary — 2026-10-04

The retained device lease now admits installed Baseline02 Linux 8584ccd8 / Reborn b71b4688, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0. Freezer on USB and battery devices/platform/processors/core PM tests PASS: helper return, kernel COMPLETE/error 0, unchanged boot, zero ext4 errors and no SPM entry. Charging devices test intentionally refuses at DPM_PREPARED with -EBUSY and returns awake; its guard is preserved. One awake RTC alarm delivers one child IRQ and clears; it does not prove suspend wake. Wi-Fi observation survives radio quiesce/restore. Owner disconnects USB; actual Discharging and USB presence 0 are verified before battery tests.

Read-only issue refresh: #28/#31/#34 OPEN; #30 CLOSED since 2026-09-15 despite its retained checklist. This pass changes no issue status and cannot infer full-suspend acceptance from that administrative closure. Runtime C3 evidence remains separate. Full Power/RTC wake, same-boot deep return, actual subsystem restoration and repetition are still NOT_RUN. Source kernel guards and the new Reborn protocol are built/tested but not installed; installed-image tests cannot qualify their execution. No milestone closes, charger bypass, register change, flash, push, protected-partition or memory-layout change is admitted. Next is one short RTC full trial with existing one-shot backstop and retained journaling; any failed checked return stops subsequent entry trials pending diagnosis. Raw receipts: out/platform-v1-suspend/pm-*-battery/, pm-freezer-charging/, pm-devices-charging/, rtc-awake.json. Full implementation and qualification.

Platform 1.0 full suspend implementation admission — 2026-10-04

Owner explicitly authorizes implementation and bounded whole-device qualification on dedicated codex/platform-v1-suspend worktrees (Linux base b5aea93, Reborn b71b468), with no flash or push. A non-blocking shared device lock governs every physical operation. First locked read observes Baseline02 kernel, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0, active charging and no current retained suspend journal. Installed source identity needs a complete receipt; this new image does not inherit blanket physical qualification from UART01.

Retained UART01 evidence proves C1/C2/C3 in its bounded runtime scope. Retained Fix02 evidence proves SPM/RTC/reset-vector/CPU/device return followed by a stall before kernel EXIT and watchdog recovery, not successful same-boot resume. Current staged PM, Power/RTC wake and all full-suspend subsystem restoration remain PARTIAL/NOT_RUN. Existing #28/#30/#31/#34 and production gates remain open; no external issue change or milestone closure. Preserve C1/C2/C3, exact MT6582 PCM and register contracts, charger refusal, storage and memory layout. Next: trace the installed image, run safe staged PM before one RTC and one Power wake, then bounded mixed cycles only after checked same-boot restoration. Source work continues while the shared device lock or charging prevents physical entry. See implementation receipt.

Baseline02 source integration into main — 2026-10-03

Owner-authorized integration carries cpu-idle-completion at acb5bb3 into remote main at 131ee46 by fast-forward, following the integration gap review. The current source/documentation line is ready for publication. The sealed Baseline02 still identifies compiled Linux 8584ccd8 / Reborn b71b4688; later publication/documentation revisions do not change package bytes.

Retained UART01 hardware evidence proves bounded C1/C2/C3, natural parking, timers, five guarded OPPs and affected I/O behavior. Baseline02 is software/package validated; new-image and automatic cold-boot qualification remain PHYSICAL_NOT_RUN. The next CPU boundary is still owner installation followed by exact identity, boot-policy readback and the guarded one-plus-twenty C3 trials/runtime regression. Full system suspend, wider endurance/thermal conditions, electrical battery measurements and public distribution remain independent gates. #28/#31/#34 remain OPEN after read-only refresh; this integration closes no hardware milestone.

Baseline02 documentation synchronization — 2026-10-03

This is a documentation-only boundary review using the already retained Baseline02 seal and UART01 hardware receipt; no build, flash, device test, remote issue update or repeat of unchanged ROM/recovery provenance occurs. The current state, Baseline02 receipt and UART01 physical report govern the current source/runtime distinction.

The C1/C2/C3 implementation and bounded UART01 hardware gates are closed. UART1 ownership and the exact MT6582 sleep ACK no longer block C3. Baseline02 is sealed and software-tested from Linux 8584ccd85f052f351fe51650b2348c83ca894ed4 and Reborn b71b468860233faa0a42b8448ec5777fa952b8e3. Owner-selected once-per-boot automatic CPU0 C3 policy is included, with unchanged dynamic guards. New-image/cold-boot physical qualification remains PHYSICAL_NOT_RUN.

Next CPU boundary: owner-preserving Baseline02 installation, exact identity and automatic boot-policy proof, guarded first C3 reset-and-return plus 20 cycles, then runtime/wake/storage regression. No C1/C2 redesign or generic CPU audit is needed. Full system suspend/Power+RTC wake, wider endurance/thermal conditions, electrical battery measurement, native high-resolution audio and public release permissions retain their independent gates. Planning epics #28/#31/#34 remain open; this documentation review does not close or authorize their remaining work.

Historical roadmap and boundary reviews

The following text retains earlier dated decisions and results. Its “current”, “next” and candidate labels belong to those sessions; the Baseline02 summary above takes precedence. Historical failures and seal-time NOT_RUN receipts remain evidence for their exact images.

Platform v1 software completion roadmap

Baseline02 sealed candidate boundary — 2026-10-03

Y2LINUX-BASELINE-02 is built, software-validated and sealed. Compiled Linux 8584ccd85f052f351fe51650b2348c83ca894ed4 / Reborn b71b468860233faa0a42b8448ec5777fa952b8e3; kernel6.18.0-y2linux-baseline-02, rootv1.21, release1.0.0-baseline-candidate.2. Fresh kernel/Buildroot/Reborn ARM, config/DT/modules/ABI,391 production tests, 125 focused source tests,27 native filesystem/GIO and3 ALSA tests,238 Reborn tests/fmt/lint, Cortex-A7 QEMU, installed ARM tools, ELF/dependencies, legal/source collection and preserving package validation pass. Five minimal-environment checks skip dependencies separately covered by the native/ARM checks. Final package: out/y2linux-baseline-02-candidate/; receipts: out/baseline-02-validation/final-seal.json. Source/current-main inclusion and preserved implementation are recorded in preserved-architecture.json.

Owner-selected C3 startup now selects budget-1/CPU0 enable once after static foundations pass. Live ARM control/readback and isolated fault tests pass; manual stop/recovery overrides and all dynamic kernel guards remain intact. New-image cold-boot and physical qualification NOT_RUN until owner flashes. Closing read-only SSH still verifies installed UART01e9e8d63, unchanged boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint0,3647 actual C3 returns, no timeout/ context/clock restore fault. C1/C2 and C3 entry/restore implementation are preserved; SPM comment/localversion and userspace boot admission differ.

Only BOOTIMG and ANDROID/Y2ROOT are payloads. Y2DATA and protected partitions remain preserved. Exact UART01 rollback pair remains in fallback/; source/license archives and fresh validation/harness are attached. No flash/push or milestone closure. #28/#31/#34 stay OPEN; full suspend, additional wakes, endurance and electrical battery evidence remain separate gaps. Six owner documentation edits remain preserved. Next boundary: owner-controlled Download Only of the two new images, then exact-identity/automatic-boot-policy and guarded physical harness. Baseline02 handoff record.

Baseline02 automatic C3 policy source check — 2026-10-03

The owner-selected once-per-boot CPU0 policy passes isolated foundation, identity, manual ownership, readback/rollback, failed receipt and stop tests. Live ARM execution on installed UART01 also passes: quarantined budget0/state-disabled then budget-1/CPU0 enable readbacks, repeated-start refusal, same boot/taint0. Original already-qualified runtime controls are restored and owned /run test files removed. This verifies the control transaction on the existing image; new Baseline02 cold-boot execution is still NOT_RUN. C1/C2 and kernel admission/ UART ACK/context/clock/timer/restore logic are unchanged. Only the SPM explanatory comment changes. The harness now verifies the actual automatic boot receipt and controls before altering qualification policy. Freeze this coherent source pair, then build fresh kernel/Buildroot/Reborn ARM and preserve-data package. No flash, push, milestone closure or protected/memory/data change. Raw control receipt: out/baseline-02-validation/live-policy-control.json.

Baseline02 qualified C3 boot-policy admission — 2026-10-03

Owner selects Enable qualified C3 automatically for this new baseline. This supersedes the default-off packaging choice in the admission below, not its hardware evidence. Same installed UART01 source/boot,3647 real C3 returns and clean checked restores underpin the decision. Add a once-per-boot userspace policy using existing CPU0 cpuidle disable and SPM budget controls. Kernel starts quarantined, then normal startup verifies identity and static context/timer/CIRQ/ SPM foundations before budget-1/CPU0 enable. Leave dynamic screen/workload/ radio/USB/topology/OPP/deadline/clock/UART ACK guards intact. No forced parking, frequency/radio change or clock gate. Retain y2.deep_idle=off/CPU-safe recovery and manual stop; repeated service startup must not undo an owner's disable.

Targeted isolated fault tests and a bounded live policy-control/readback test precede source freeze. Then one final fresh integrated baseline build and full candidate checks. The interrupted default-off draft produced no package; it is not a delivered candidate. New persistent cold-boot behavior remains NOT_RUN until owner installs Baseline02. Full suspend/endurance/electrical evidence and #28/#31/#34 dependencies remain OPEN; no new hardware/memory/partition scope, flash or push. Current boundary receipts: out/baseline-02-validation/.

Baseline02 integrated candidate admission — 2026-10-03

Owner requests a fresh integrated baseline after UART01 physical qualification. Read-only authorized SSH reconfirms installed Linuxe9e8d63/Rebornb71b4688, kernel 6.18.0-y2linux-cpu-c3-uart-01, rootv1.20, unchanged boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint0. C3 still has3647 real entries/resumes/successes/UART ACKs, zero timeout/context/clock/CIRQ failures. The completed foundation and21 bounded C3 trials plus60s runtime observation remain authoritative; see UART01 physical evidence. This is a packaging boundary, not a general CPU audit or new hardware phase.

Prepare one fresh BOOTIMG/Y2ROOT baseline from current committed Linux and Rebornb71b4688, retaining all integrated storage/USB/audio/power/radio/product work and current UART C3 architecture. Fresh kernel, Buildroot/Reborn ARM, production regressions, QEMU/ELF/config/DT/module/ABI and data-preserving package checks are required. Preserve Y2DATA, protected partitions, guarded five OPPs, C1/C2 and default-off C3 policy; include latest guarded qualification harness. New image physical qualification remains NOT_RUN until owner installation. Fallback is the unchanged physically qualified UART01 BOOTIMG/Y2ROOT pair.

#28/#31/#34 remain OPEN: persistent C3 boot policy, full-system suspend, additional wake sources, endurance and electrical battery evidence are distinct unclosed coverage. No memory/layout/data-schema change, flash, push or public redistribution authorization. Six preexisting owner documentation edits remain separate. Boundary receipts: out/baseline-02-validation/; new baseline record.

UART01 C3 hardware closing audit — 2026-10-03

C1 WORKING; C2 WORKING; C3 WORKING_AND_REPEATEDLY_OBSERVED on installed Linuxe9e8d63/Rebornb71b4688, kernel 6.18.0-y2linux-cpu-c3-uart-01, rootv1.20, unchanged boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint0. Fresh physical foundation checks pass: four-core C1,18 normal hotplug transitions, all five OPPs, GPT6/GPT4/PPI29/13MHz/highres/NO_HZ, natural owned parking, C2+6971 entries/+50.654676s in60.054881s (84.35%), zero clock restore failures. C3 proves21 budget1/RGU/SRAM checked reset returns, then+3624 normal-policy entries/+39.204662s in60.088344s (65.25%). Final entries=resumes=successes= UART attempts=ACKs=CIRQ transactions/flushes=timer saves/restores=3647. No UART timeout, restore/context/clock/media/IRQ fault or reboot. eMMC/SD, five OPPs, screen/workload/playback, USB and Wi-Fi postwake regressions pass.

UART1 remains clocked: conditional exact MT6582 PIO ownership and global SPM request/ACK resolve admission without forced gating. No kernel/image changes after installation; locally committed c8ae170 fixes observer fsync sequencing and accounts safe guard refusals separately from actual entries. 33 targeted tests pass. Original failed observer verdicts are retained and independently rechecked against stronger UART/SRAM/context/media guards. Authoritative joined proof: out/cpu-c3-uart-physical/20261003T183259Z-uart01/final-verdict.json; hardware record.

F027/F028/F030 may advance to this bounded same-boot physical qualification. The UART C3 blocker is closed. No new flash is required. C3 is intentionally runtime-enabled/budget-1 for THIS BOOT, RGU disarmed and original radios, screen, schedutil598–1300MHz and coordinator restored. Fresh boot still uses image default-off/budget0; no persistent default-on policy is claimed or changed. Full system suspend, new wake-source coverage, endurance, electrical battery measurement and other #28/#31/#34 dependencies remain OPEN. No general milestone closure, memory/partition/protected-data change, flash or push. Six preexisting owner documentation edits remain preserved separately.

UART01 physical trial admission — 2026-10-03

Owner confirms the new candidate installed and authorizes the prepared hardware qualification. Read-only SSH verifies Linuxe9e8d63/Rebornb71b4688, kernel 6.18.0-y2linux-cpu-c3-uart-01, rootv1.20/buildY2LINUX-CPU-C3-UART-01, boot fd955840-35d9-47db-83e0-ff47d6bb2d2b, taint0. Checkout55db6d3 is a later receipt, not installed source. C3 remains disabled/budget0 and actual dormant/ UART handshake counters0. UART0 sleep enable1 and UART1 conditional deferred bit0x20000 are now physically observed. UART1 is no longer a mandatory gate prerequisite; global SPM ACK and context return remain unobserved. Current normal Wi-Fi/USB/1300MHz adds legitimate dynamic blockers, not a new settled UART diagnosis. Existing candidate02 C1/C2 qualification stays authoritative until regression checks on this image complete. Raw evidence: out/cpu-c3-uart-physical/20261003T183259Z-uart01/.

Dependencies #28/#31/#34 stay OPEN; no milestone closure or hardware-limit claim. Proceed with normal C1/C2/timer/hotplug/storage/five-OPP regressions, then natural quiet CPU0/low OPP/radios-off/USB-detached topology. One budget1 RGU/SRAM/timer-wake trial must show real request/ACK and checked same-boot return before20 further bounded cycles. Failure disables C3 and stops trials; observe the precise stage, retain independent regressions and fix that cause. Stop paused playback through the existing authorized owner control. No forced UART gate, arbitrary MMIO/probe, memory/partition change, flash or push; preserve six preexisting owner documentation edits. No general CPU redesign, full-system suspend or electrical battery claim is authorized by this narrow trial.

The installed UART01 regression pass now proves four-core C1, natural parking, C2 +6,971 entries/+50,654,676us over60.0549s with zero restore failures, all five OPPs, hotplug, timers, eMMC/inserted SD, screen/workload/playback and USB/Wi-Fi integrity. Same boot/taint0, no filesystem/IRQ faults. Its first C3 window stopped at legitimate MSDC0 bit12 (0x1000), not UART1: the harness fsynced its durable arm receipt after the clean preflight, waking eMMC. Actual UART requests/ACK attempts, CIRQ transactions, CPU context entries and returns all remain0. Therefore this is an observer sequencing defect, not a failed deep-idle experiment. Preserve the kernel/storage guard. Targeted harness fix saves before a read-only two-second stable admission window, checking the real MMC gates, all exposed prerequisites and owned physical CPU0 topology; no storage writes occur between that window and enabling C3. A C3-only continuation requires the exact clean same-boot regression receipt and zero actual UART/CPU entries both in that receipt and on-device immediately before mutation. Any real UART/context refusal stops repeats; first checked entry must still precede 20 further cycles. Thirty targeted harness/CPU/UART tests pass. Scope remains unchanged; dependencies stay open pending real hardware ACK/context results.

UART01 FIRST REAL DORMANT RETURN is now physically proved on the same boot. Corrected observer sequencing reached global UART request/ACK1, exact POWER1 0x15820→0x15821→0x15820, R13 ACK0x8140000, one CPU entry/resume/success, 8,065us SPM/9,363us cpuidle residency, GPT wake0x10, CIRQ clone/flush1/1 and local timer context save/restore1/1. SRAM records UART_REQUEST→UART_ACK→ DORMANT_CONTEXT→DORMANT_FINISH→DORMANT_RETURN→DORMANT_COMPLETE with actual reset-resume assembly marker0x59325253; no restore/context/media/IRQ fault, taint0 and unchanged fd955840 boot. The harness incorrectly rejected the otherwise successful trial because spm.c counts every admission before the budget check: +3 admissions included +2 safe -EACCES budget refusals after the single physical entry. Fix measurement to require EXACTLY one physical entry/ return/success/ACK, account all extra admissions as aborts, require final budget refusal and every original restore guard plus SRAM reset marker. Original raw FAIL receipt remains intact; independent first-entry recheck is attached. 33 targeted tests pass. Continue ONLY20 further budget1 guarded cycles after rechecking the same-boot first-entry and original regression receipt hashes; no retry of any actual hardware failure. C3 stays disabled/budget0 between trials. Full repeatability/policy qualification and dependency closure remain pending; no kernel build or image changes are needed for these harness fixes.

Repeatability now physically proves18 UART ACKs/CPU reset entries/resumes/ successes on unchanged fd955840 boot, positive residency and zero UART/CIRQ/ timer/clock/context/media faults. Sixteen further cycles passed directly; cycle17 also genuinely succeeded, but the checker overrequired the LAST admission to be budget-refused when its two safe guard refusals preceded the successful entry. Correct verdict permits last resumed/result0 as well as budget/-EACCES, still accounting every refusal and requiring exactly one new physical entry/ACK/checked reset return and every restore. Independently rechecked all18 raw cycles (original FAIL receipts retained with their SHA256). Continue only the remaining3 bounded cycles, then near-deadline fallback and 60s normal-policy residency. Any real failure still stops; ordinary policy may stay enabled for THIS BOOT only after all21 checked cycles and postwake regressions pass. Fresh boots retain image default-off/budget0; no persistent policy/image/default-on change is implied. No general CPU or suspend milestone closure; release claims may advance only to the actual bounded idle evidence.

C3 UART sealed candidate boundary — 2026-10-03

Exact MT6582 PIO-owner admission/global UART request-ACK integration is fixed; C1/C2 and all unrelated clock/context/timer/OPP architecture are preserved. Candidate out/y2linux-cpu-c3-uart-candidate/ at Linuxe9e8d63/Rebornb71b4688 is freshly built/sealed. Production375/source109/native host/Reborn238 tests, ARM/QEMU/config/DT/modules/ABI/ELF and preserving package checks pass; five minimal-host dependency skips have separate native/ARM coverage. Sources and collector limitations are attached; package remains owner-local.

Fresh closing SSH still verifies installed candidate02/unchanged boot, C3 disabled/budget0/entries0; C1/C2 working evidence remains authoritative. C3 has never entered. #28/#31/#34 and CPU idle acceptance remain open until owner-installed UART01 gives real ACK and bounded same-boot context/timer wake. No hardware-limit proof, full suspend/electrical battery claim, protected data change, flash/push or milestone closure. Owner alone installs BOOTIMG+Y2ROOT with the preserving scatter; harness is ready for first entry then20 further cycles. Six preexisting owner doc edits remain separate. Exact source, hashes, checks and owner action are in the C3 UART handoff.

C3 UART software freeze audit — 2026-10-03

Targeted stock-contract implementation and100 focused tests are complete. Installed candidate02/source/boot and actual C1/C2 results remain unchanged; C3 has never entered. The copied default UART1 mask lacked stock PIO-owner integration. Conditional UART1 sleep adoption, exact MT6582 console driver sleep capability, bounded global SPM ACK/refusal, checked unwind, retained stages and first-entry/20-further-cycle harness replace that missing contract. All unrelated clock guards, C1/C2/timers/parking/OPPs/media architecture remain. Exact source/ownership/unknown RTL limits are documented in the C3 UART record.

Before production handoff, build one fresh kernel/Buildroot/Reborn ARM pair and validate config/DT/modules/ABI, regressions, ELF/QEMU and preserving package. Candidate scope is BOOTIMG+Y2ROOT at out/y2linux-cpu-c3-uart-candidate/, API1/data/layout unchanged, no protected payload, no flash/push. No hardware promotion: #28/#31/#34 and CPU completion remain open pending real ACK and bounded same-boot DORMANT wake after owner installation. No new general audit, ROM/recovery provenance pass, electrical measurement or full suspend claim.

C3 UART stock-contract admission — 2026-10-03

Owner narrows work to the candidate02 UART1 C3 blocker. Preserve physically working C1/C2 and their architecture; do not conduct another general CPU audit. Fresh targeted SSH verifies installeddb0234e/b71b4688, kernelidle-02/rootv1.19, buildY2LINUX-CPU-IDLE-COMPLETION-02, same boot f155196b-64d4-45a4-88b3-27755a1a8926. UART1 remains clocked with IRQ/DMA/LSR0; actual dormant calls0, C3 disabled/budget0. Restored normal radio/OPP policy legitimately adds other current preflight reasons; the prior settled receipt still isolates UART1 alone. Expected prior diagnostic taint4096 is retained. New read-only admission: out/cpu-c3-uart-physical/20261003T164546Z/.

Exact MT6582 d53dd75c mt_idle default PERI mask includes UART1 bit17; mtk_uart_startup enables DPIDLE for non-DMA ports and enables UART_SLEEP_EN, while DMA ports remain blocked. The global SPM R7 request/R13 ACK follows PCM fetch and precedes PCM register/power/run/WFI programming. Existing Y2Linux has the handshake but lacks UART's conditional mask/sleep-owner integration. Therefore merely dropping bit17 is unauthorized by the hardware evidence. The next implementation boundary is conditional PIO sleep adoption with bounded ACK/refusal, exact unwind and diagnostics, preserving all unrelated blockers, CCF ownership, C1/C2 and existing timer/context/CIRQ safeguards. No forced gate, FIFO/data/baud/reset writes, invented PCM, memory/partition expansion or live C3 trial on the old kernel. Normal console ownership must provide its sleep capability as required by the same global handshake.

Existing #28/#31/#34 clock/idle dependencies remain open; CPU completion is not closed or hardware-impossible. User authorizes one fresh preserving candidate out/y2linux-cpu-c3-uart-candidate/, BOOTIMG+Y2ROOT only, with owner-only flash and no push. Software validation and a prepared first-entry/20-cycle harness must precede handoff; physical DORMANT success remains unobserved until then. Unchanged ROM/recovery provenance, protected data and six owner doc edits stay separate. Full-system suspend/electrical battery benefit remain outside scope.

Candidate02 physical closing audit — 2026-10-03

Owner-installed candidate02 is verified at Linuxdb0234e/Rebornb71b4688, kernelcpu-idle-02/rootv1.19/releasecandidate.2, unchanged boot f155196b-64d4-45a4-88b3-27755a1a8926. Repo HEAD at admission76567e5 is a later receipt, not the installed source. Fresh raw SSH/source/issue evidence: out/cpu-idle-completion-physical/20261003T144857Z-candidate02/. Blueprint/owner scope remains real C1/C2/C3 with no fake guards/counters, no flash/push and unchanged memory/partition/recovery contracts. The existing #16/#27/#28/#29/#31/#32/#33/#34 refresh remains OPEN; no epic is activated/closed.

Main qualification/settled preflight taint0. Two bounded exact-kernel probes then record UART1 reset/sleep/IRQ/DMA/LSR0, DLL1/DLH0, exact LCR restore and balanced CCF refs; they do not prove safe gating. Expected out-of-tree taint4096 remains truthfully until owner reboot, with no resident probes or new critical errors. No C3 trial under diagnostic scope. Final same-boot controls: C3 off, budget0/RGU disarmed, schedutil598000–1300000/coordinatorY, USB device, original screen/radio policy restored, ext4/MMCerrors0, temporary observers/files removed. Playback stays stopped because its initial paused fixture was already deleted.

The harness now waits for normal parking again after synced storage, checks owner and both power copies for20s, and keeps progress saves sparse. Fresh locked CPU90+slow-idle/suspend8+workload2 cases pass;19 native probe guard cases pass. No production kernel/root changes or new image; sealed02 hashes still match. Preserve the six preexisting owner documentation edits and unchanged provenance. The physical pass is complete, but CPU idle completion/release acceptance is blocked by UART1, not closed. Current candidate02 receipt supersedes its historical NOT_RUN/owner-flash instruction below. See physical record and source/harness record.

Candidate02 bounded UART1 divisor snapshot admission — 2026-10-03 15:34 UTC

The first nonresident exact-kernel probe retains UART1 gate0 before/after normal CCF borrow/release, reset0/0, sleep0, FCR_RD0, ACTIVE_EN0, IRQ/DMA0, LSR0; source-pinned non-destructive operands are captured in the current physical run. This disproves a held PERI reset or enabled sleep-control explanation; it does not establish transmitter-idle or a hardware limit. Same installed pair/boot, C3 disabled/budget0, expected external-module taint4096, no resident probe.

The next existing-UART observation uses the exact MT6582 BSP mtk_uart_save/mtk_uart_cal_baud LCR.DLAB select/read DLL+DLH/restore sequence. It changes only the saved register-bank selector and checks its restoration; never writes baud, FIFO, data, IRQ, DMA, GPIO, or reset. Host disables the normal coordinator, waits for all four cores, and switches radios off through their owners; both physical power copies must have MD/CONN off. Module requires CPU0, all four cores, UART1 already clocked, LCR0, and IRQ/DMA disabled before its bounded IRQ-disabled bank transaction. This excludes concurrent unused-clock handoff (its worker requires enabled coordinator/one core). Unknown/live/banked states are skipped. Exact ARM build and11 actual-function guard/cleanup tests pass; module returns EAGAIN and is not resident. The existing expected4096 taint is retained truthfully. Radio/coordinator settings restore in finally. No C3 entry, guard relaxation, live-media reset, memory/partition change, flash/push, external milestone activation/closure, or unrelated release promotion. Raw evidence remains out/cpu-idle-completion-physical/20261003T144857Z-candidate02/; #28/#31/#34 remain the clock-owner dependencies.

Candidate02 UART1 read-only diagnostic admission — 2026-10-03 15:20 UTC

Fresh settled preflight in boot f155196b-64d4-45a4-88b3-27755a1a8926 at Linuxdb0234e/Rebornb71b4688 proves CPU0-only, owner0xe, both secondary power copies0, domains0,747.5MHz, CIRQ/timer/context/PCM prerequisites available; only clocks fails, PERI0x00020000/UART1. SPI0 is correctly released. LCR/IER/ DMA/LSR remain0 after repeated clocked rechecks. No C3 call occurs. Supported USB/radio/DVFS/screen/coordinator settings restore, same boot/taint0/ext4errors0; new live admission verifies C3 disabled/budget0/backstop0 and exact installed pair.

The existing #28/#31/#34 clock-owner dependency needs UART reset/sleep operands, which the installed handoff diagnostic does not expose. A bounded, source-pinned read-only kernel probe is admitted for this existing subsystem. It accepts only the exact Y2 CCF UART1 handle and fixed MT6582 register windows, requires its gate already on, borrows/releases through normal CCF, skips alternate banks and live IRQ/DMA LSR, and never writes UART/GPIO/IRQ/DMA/reset controls. All eight actual probe guard/ref-cleanup fault cases and its exact installed-kernel ARM build pass. It returns EAGAIN after cleanup, leaves no resident module, and truthfully leaves the expected external-module taint until the next owner boot. Qualification results obtained beforehand retain their original taint0 receipt. No C3 entry will be attempted under this diagnostic scope. This is not production activation, hardware-limit proof, a blocker-mask relaxation, memory/partition expansion, flash/push, or an unrelated release closure. Full-system suspend remains excluded. Raw source/build/fault/live receipts stay in the candidate02 physical session.

Candidate02 post-flash hardware admission — 2026-10-03 14:50 UTC

Owner reports installing the sealed follow-up02 and requests physical tests. Fresh authenticated USB SSH verifies Linux db0234e7519c559031da6f427869ab683ffe0f3c, Reborn b71b468860233faa0a42b8448ec5777fa952b8e3, kernel 6.18.0-y2linux-cpu-idle-02, root 2025.02.18-platform-v1.19, release 1.0.0-cpu-idle-completion-candidate.2, and boot f155196b-64d4-45a4-88b3-27755a1a8926. All manifest identity checks match; taint0 and mounted ext4 error counts0. Raw commands/exit codes/timestamps: out/cpu-idle-completion-physical/20261003T144857Z-candidate02/. Read-only issue refresh retains #16/#27/#28/#29/#31/#32/#33/#34 OPEN.

The prepared durable device-side runner uses supported owners and restores screen/radio/DVFS/coordinator settings. Playback state is recorded before normal stop. It does not force busy UART/DMA/IRQ gates, weaken masks, reset live media, perform userspace MMIO, flash, push, change memory/partition scope, or close external milestones. C3 remains disabled on any failed prerequisite or wake; normal enablement requires repeated physical success and remaining regressions. The physical phase is admitted, not accepted. Preserve owner documentation edits and the sealed packages; unchanged ROM/recovery provenance needs no recapture.

CPU idle hardware closing / candidate02 owner-flash boundary — 2026-10-03

Fresh final SSH receipt after build/sealing retains installed01 Linux3dfb5f5/ Rebornb71b4688, boot e0eb2f36-7c1f-4007-99e2-81d2eae047d4, taint0, mounted ext4 errors0. Screen off, Wi-Fi online/Bluetooth off, USB device, original DVFS/ coordinator restored; C3 disabled/budget0; temporary observer removed. Active issues#16/#27/#28/#29/#31/#32/#33/#34 refresh read-only and remain OPEN.

Candidate02 source db0234e/b71b4688 is freshly built and sealed, kernelcpu-idle-02/ rootv1.19. CPU owner/fault100, production366 with native skips covered, Reborn238/ fmt/clippy, fresh ARM kernel/root, config/DT/modules/ABI, QEMU/dependencies/ inventory/legal-info and preserving/root/source/checksum checks pass. C3 category SOFTWARE_READY_NEEDS_NEW_FLASH means ready for bounded qualification, not proof both clocks clear. One guarded entry and then19 more only on first success remain mandatory. Latest package:out/y2linux-cpu-idle-completion-02-candidate/; installed01 package remains unchanged. No reused root image; all affected/local ARM components rebuilt after final source cutoff. Exact Hardware02 fallback, Y2DATA/protected partitions, memory ABI and unchanged ROM/recovery provenance remain preserved. No unsafe MMIO/IRQ/DMA/reset forcing, guard-mask reduction, flash/push or unrelated release closure. Owner-controlled installation is the remaining boundary; see source/hashes/action and physical result.

CPU idle candidate hardware result and follow-up admission — 2026-10-03

Fresh candidate01 run at Linux3dfb5f5/Rebornb71b4688, boot e0eb2f36-7c1f-4007-99e2-81d2eae047d4, completes with CPU idle acceptance FAIL: C1 passes all four cores; natural parking reaches CPU0/owner0xe; C2 adds 7,040 entries and49.674049s/60.047965s (82.72%), restore failures0. Both mounted media pass16 fsynced checksum cycles each across the two storage phases, ext4 counts0. Three hotplug cycles, all five OPPs before/after preflight, screen wake, real playback and bidirectional USB/independent Wi-Fi integrity pass. Same boot, taint0, owner screen/radio/DVFS/coordinator restored; C3 off/budget0/backstop0.

C3 is not physically working: preflight rejects PERI0x02020000, exact UART1 bit17 and SPI0 bit25, after normal radio-off/USB-role-none, CPU0-only topology, 747.5MHz cap and an armed RGU backstop. Every other reported static predicate passes; no actual cpu_suspend call occurs. Boot warnings and CCF summary show both retained, zero references, no Linux consumer. Installed diagnostics omit raw handoff operands; do not infer UART's exact busy field or fake C3 success.

Pinned exact MT6582 d53/3be93a68/krillin SPI source identifies STATUS1[0]=1 as idle; the installed predicate instead accepts0/rejects1. UART platform source identifies DMA_EN bit2 as timeout-counter metadata, unlike RX/TX bits0/1. Correct those source-backed decisions, retain unknown/live engines, capture read-only clocked handoff operands, and preserve aliased UART banks. Partial NAND/PWM handoff must retain its gate without aborting the shared CCF provider. No blocker masks, timer/context/PCM/CIRQ/OPP algorithms or memory ABI change.

The unused-clock owner also rechecks retained UART1–3/SPI0 through the existing CPU0-only, screen-off, lease-free deferrable worker, only with both radios physically off. It borrows/releases through CCF, keeps IRQ/DMA/unknown engines, never reads gated or aliased windows, avoids live UART clear-on-read LSR, and quarantines a failed gate readback. A transient loader operation no longer pins its clock forever. No new timer, coordinator-policy/hysteresis change or reset. Native fault tests cover real refs, still-busy engines, unavailable domains, unclocked skips, allocation/borrow and gate-readback failures.

Existing#28/#31/#34 track this next owner dependency; all refreshed issues remain OPEN. The original autonomous completion instruction authorizes safe source fixes and one coherent follow-up preserving candidate. Candidate02 gets a fresh kernel/root build and affected/full production verification; no reused root image, no flash/push, no external closure, no unrelated qualification claim. Current candidate01 hardware evidence stays attached to its actual source pair; candidate02 will remain physically unqualified until owner flash. Full-system suspend and electrical battery improvement remain separate unresolved gates. Physical record, private raw out/cpu-idle-completion-physical/20261003T130848Z-candidate/.

CPU idle candidate physical admission — 2026-10-03 13:13 UTC

Owner reports installing the sealed CPU idle BOOTIMG/Y2ROOT and explicitly requests hardware qualification. Fresh authorized USB SSH verifies actual Linux3dfb5f5/Rebornb71b4688, kernel6.18.0-y2linux-cpu-idle-01, root2025.02.18-platform-v1.18/releasecpu-idle-completion-candidate.1, boot e0eb2f36-7c1f-4007-99e2-81d2eae047d4, taint0; all identity axes match the candidate manifest. Raw admission/status/health/dmesg/timer/IRQ receipts: out/cpu-idle-completion-physical/20261003T130848Z-candidate/. Active issues#16/#27/#28/#29/#31/#32/#33/#34 are refreshed read-only; remain OPEN.

Pause/session state is recorded before normal playback stop. This pass uses the prepared durable device-side SSH harness with independent Wi-Fi observation, supported radio/USB/CPU controls, normal-policy parking, failure quarantine and owner-setting restoration. Ordinary C3 is retained only after repeated bounded success and the remaining regressions. No guards are weakened, no userspace MMIO or live-media reset, no flash/push, no memory/partition/calibration change, no unchanged ROM/recovery recapture or external milestone closure. The physical phase is admitted; acceptance is pending fresh evidence. Preserve the preexisting owner documentation edits and exact Hardware02 fallback.

CPU idle owner-flash boundary — 2026-10-03 11:07 UTC

Repeated standing gap audit uses final current-boot receipts21–23 and read-only GitHub receipt24, alongside the complete CPU evidence and the fresh committed build pair. Installed baseline remains Linux131ee462/Reborn7f9df397, boot cee326c1-a5b0-447a-8eb0-dc3f39f7e2c2, taint0. C1 is WORKING on all four cores. Natural CPU0 parking plus C2 adds7182 entries/50.763741s in60.404076s; controlled 30-second enabled/disabled/re-enabled windows add25.023/0/25.016s SLIDLE, respectively. eMMC/inserted SD ext4 errors remain0. Owner screen/radio/DVFS/USB policy is restored; the temporary owned Wi-Fi observer is stopped.

Fresh clean paired source 3dfb5f5/b71b4688 produces kernel6.18.0-y2linux-cpu-idle-01 and root2025.02.18-platform-v1.18. Kernel/config/DT/modules/ABI, fresh Buildroot/ Reborn ARM, production regression364 cases (five locked-host prerequisites covered by25 packaging-host cases), Reborn238 host cases/clippy/fmt, installed ARM/QEMU,383 ELF files/1403 dependency edges and release inventory/legal-info complete. Software evidence is not current-candidate hardware acceptance. Preserving-package/root/composition/checksum validation and final hash seal pass; owner-flash handoff is ready. No reuse-userspace.

Active issues#16/#27/#28/#29/#31/#32/#33/#34 remain OPEN; no external status is changed. CPU work maps to existing#28, owner dependencies to#27/#31/#33/#34. The next phase is software-ready, physically unqualified until owner flash. Recovery/fallback uses the exact Hardware02 BOOTIMG/Y2ROOT pair. Memory, ROM/ recovery provenance, partition layout, factory/calibration protection and Y2DATA preservation remain unchanged. No hardware limitation is inferred from an untested prerequisite. No milestone is closed, no default-on C3 patch, no flash or push. See implementation and physical record.

CPU idle completion admission — 2026-10-03

CPU idle dependency audit — 2026-10-03 08:05 UTC

Current-boot receipts04–11 now physically close C1 and C2 on the installed baseline: four-core WFI counters/residency and timer continuity pass; natural parking reaches owner0xe/CPU0; C2 adds7182 entries and50.763741 seconds residency in60.404075502 seconds (84.0%), with restore failures0. Eight uncached, fsynced SHA256 cycles per eMMC/SD medium pass in each radio-off run, ext4 counts0. Three hotplug cycles and all five guarded OPPs pass. C3 remains unentered. Live screen-off still owns OVL/RDMA/COLOR/DSI and32-kHz mutex references; inherited AFE clocks remain enabled with zero references. This new evidence supersedes the admission's pending C1/C2 observations, not unrelated gates.

Exact d53dd75c/3be93a68 MT6582 dpidle_condition_mask requires DISP0 0x7ffff and DISP1 0xf clock quiescence, rather than unconditional DISP-domain power-off. CPU completion therefore includes the existing Reborn DRM master's screen-sleep lifecycle, balanced shared display/audio clock ownership, and read-only checks for safely retiring inactive loader clocks. Active or unknown engines remain blockers. This dependency is within the owner's explicit CPU clock/wake scope; it authorizes no new display feature, arbitrary power-domain shutdown, DMA reset, MMIO userspace access, memory layout, partition, flash or push. Existing GPU and radio ownership and active performance ceiling remain required regressions.

Owner explicitly authorizes an autonomous CPU/idle implementation and physical qualification pass on installed y2linux-baseline-candidate, with one fresh y2linux-cpu-idle-completion-candidate preserving BOOTIMG/Y2ROOT package. Only the owner flashes; no push, protected-partition or Y2DATA payload is admitted.

Read-only SSH verifies Linux 131ee4621cd583c955182f994adac5a594fb823f, Reborn 7f9df397ab3809d52e2f1073ca93a305246e0c3a, kernel 6.18.0-y2linux-baseline-01, root 2025.02.18-platform-v1.17, release 1.0.0-baseline-candidate.1, boot cee326c1-a5b0-447a-8eb0-dc3f39f7e2c2, taint0, matching the sealed package. This supersedes installed Candidate4 identity only; retained Fix02 passes/failures remain historical evidence. Private receipts: out/cpu-idle-completion-physical/20261003T074026Z/.

Initial current coverage: GPT6/GPT4/PPI29/CNTFRQ13MHz/highres/NO_HZ admitted; all five guarded OPPs admitted; CPU0-only coordinator-owned topology observed. C1 registered/enabled with accumulated residency, fresh continuity qualification pending. C2 registered/enabled but entries0, live APDMA/BTIF blockers while Wi-Fi runs; bus rejection0 and MMC runtime gating intact. C3 registered/default-off, entries0; current frequency, domains and clocks reject its read-only preflight. No new idle-state pass is claimed. Existing storage tuning remains Candidate4.

GitHub active issues refreshed read-only: #16/#27/#28/#29/#31/#32/#33/#34 OPEN; CPU/timers/idle map to #28 and storage/radio/GPU interactions to #33/#31/#34. Their older descriptions do not override current source or hardware receipts. No external issue or milestone is closed or changed. Every unrelated hardware, audio/distribution/endurance gate retains its existing classification.

Scope: characterize C1 then freeze its proven architecture; decode/fix actual C2 clock/device ownership; resolve every source-backed C3 topology/deadline/CIRQ/PCM/ context prerequisite, without masking blockers or undocumented register writes. Bounded existing runtime controls and workload/storage/network regression tests are authorized, with safe abort and state restoration. First C3 entry requires all real prerequisites; no repeated hammering of a failed deep path. Preserve GPT6/GPT4/PPI29, schedutil/QoS/thermal, hotplug, all guarded OPPs, exact memory/ partition layout and Hardware02 fallback. Full system-suspend acceptance remains a separate gate. Coherent source fixes and fresh software/package qualification precede owner flash and current-candidate physical acceptance.

Track implementation and physical results in CPU idle completion and physical qualification.

USB clock dependency audit — 2026-10-03

Current-boot receipt12 uses supported USB role none over the verified Wi-Fi observer. MUSB changes active ->suspended and returns active on restoring device; same boot, no source/partition changes. PERI USB0 bit10 nevertheless remains active in the exact C3 blocker mask. Source confirms the glue adopts that gate without a CCF consumer. The MT6582 d53 USB PHY's usb_enable_clock explicitly owns MT_CG_PERI_USB0 through enable_clock/disable_clock.

This CPU completion boundary now admits a balanced USB0 CCF consumer and MUSB runtime save/gate/enable/restore hooks, preserving attached sessions, DMA/IRQ refusal, PHY sequencing and ordinary role-based detach/reconnect. No blocker bit is removed. Only the existing controller is modeled; no host, VBUS, PHY-retune, power-domain, memory, partition or flash scope is added. Fresh USB regressions and same-boot role recovery are required. The next C3 harness may intentionally release USB through the existing role interface while its durable device-side job runs, then restore transport. Radios and USB may both legitimately prevent C3 until their owners have quiesced.

CPU idle source/build boundary audit — 2026-10-03

Fresh same-boot receipts04/10/11/12/16/17 now establish four-core C1, useful C2, physical secondary hotplug, all five OPPs, the USB0 ownership leak, actual44.1-kHz playback with PlaybackNormal demand/all-core restoration and screen-off/on userspace transitions. Taint0; mounted eMMC/SD errors0; C3 has not entered. The source pass preserves the working timer/C1/C2/MMC/DVFS/parking algorithms and repairs the exact C3 clock/context/deadline/CIRQ/PCM prerequisites. Targeted native fault tests and fresh Reborn host/clippy checks pass; full fresh software and package qualification is next. This crosses a software build boundary only: no new hardware/memory/partition scope, milestone closure, flash or push. All unrelated full-suspend/storage-ceiling/audio/distribution gates remain open. One coherent CPU idle preserving candidate and its automated20-cycle harness are the next deliverables. CPU hardware completion remains pending owner flash.

Integration build boundary — 2026-10-02

Handoff closure: fresh source 02eb735 / 72e94b2 is now sealed as Y2LINUX-FEATURE-COMPLETION-01,350 production and238 Reborn tests passed, ARM/QEMU/package checks passed. READY FOR HARDWARE DECISION RUN only. Installed Candidate4 and retained Fix02 physical evidence below are unchanged; no new physical qualification, public distribution or milestone closure.

The read-only connected-device receipt establishes installed Candidate4 Linux dcbd7d1 / Reborn b92d312, boot 159f7c81-ab89-4440-bc89-df827dd8bc34, taint0. Current storage negotiates HS200/SDR104 near200MHz; this narrow idle observation is not endurance acceptance. PM records are invalid/no-attempt; C2 has no observed entries and C3 remains disabled. Fix02 remains the broad physical qualification, with full-resume/C2/loaded-USB failures unresolved physically.

The authorized fresh build uses Linux 02eb735 and Reborn 72e94b2, including source-backed tuning recovery, truthful PM evidence, codec integration, Productv2 and tester/release safety. No partition/memory/voltage scope expansion, physical qualification or milestone closure. Only one preserving owner-local integration candidate is admitted; all existing hardware and public-distribution gates remain. No flash or push. The implementation pass records final software/build outcomes.

Feature completion implementation admission — 2026-10-02

The owner now explicitly authorizes the ten-lane implementation campaign and one fresh preserving candidate; the earlier planning-only limit is superseded. Starting Linux 9ac2c0b7018ce707ec15f8845daf58ce38ec18a0, Reborn eb64b8b21058eacb1ef183f7570cca8100cbbb46. Existing dirty planning documentation is preserved. The 233-feature inventory remains the work queue; no fresh broad audit or unchanged ROM/recovery provenance acquisition is substituted for implementation.

Standing boundary review uses the retained Fix02 real hardware result (awake timers/OPPs/parking/MMC/staged PM pass; full resume, C2 and loaded USB fail), candidate3 storage measurements (200 MHz CRCs, bounded 100 MHz read pass), and candidate4 source/304-test receipt (not hardware acceptance). Boot polish source is newer than the planning snapshot; its build is unfinished, not a sealed receipt. GitHub #16/#27/#28/#29/#31/#32/#33/#34 refreshed read-only and remain OPEN. Every existing coverage classification is carried forward: no new physical pass or milestone closure follows from this activation. Scope maps platform to #27/#28/#33, audio/radio to #29/#31, product/release to #16/#32/#34.

No memory-map, partition, voltage or clock-limit change is admitted. Preserve BOOTIMG/Y2ROOT-only installation, Y2DATA and exact Hardware02 fallback. Current SSH identity is attempted read-only; absence of access does not invalidate retained evidence. Parallel source lanes continue across independent blockers. Source-ready changes require targeted tests, then fresh integrated software/build/package checks. Hardware acceptance and public redistribution remain separate gates. No flashing, push, destructive device operation or external issue mutation. Track findings in the implementation pass.

Current beta planning boundary — 2026-10-02

The owner's later release-target correction supersedes the initial minimal-beta proposal. Use the complete factual audit and feature-completion plan: three passes (platform; audio/Bluetooth; Reborn/release), then one integrated RC, physical qualification and endurance. Native meaningful24-bit/rates where supported, viable Classic codecs, C2/conditional C3, same-boot deep sleep, admitted high OPPs and fastest robust storage are release targets. Lower modes are fallbacks, not scope waivers. Verdict: FEATURE COMPLETION IN PROGRESS. Candidate4 is sealed/software-validated, unflashed; candidate3 latest reported installed; Fix02 latest broad physical evidence. The standing gap audit is refreshed for planning only. No implementation, build, flash, push or milestone activation.

Current scope, 2026-10-01: See platform state and the Fix02 physical result. The owner-flashed Fix02 qualifies the awake CPU platform, but fails acceptance on full-suspend resume completion, SLIDLE and loaded USB upload. One Fix03 correction candidate is built for the owner's next run; it is not a release. This roadmap keeps earlier Platform v1/Hardware Final/CPU Final boundary decisions as history. Their "ready" and "next" language belongs to those earlier packages.

CPU Final Fix03 — single correction candidate, 2026-10-01

The implementation boundary implements exactly the Fix02 handoff batch. It adds timed resume-window marks up to pm_suspend exit, with the decoded RGU reset cause. It reconciles the stock bus-DCM baseline for SLIDLE and deep idle, makes the USB storm guard progress-aware (the Fix02 storm was legitimate throughput), and fixes three policy defects: wake escalation, SD identity and the DVFS label. The output is one preserving out/y2linux-cpu-final-fix03-candidate/ BOOTIMG/Y2ROOT package with the exact Hardware02 fallback; no flash or push. Next: the owner's single Fix03 run, then one targeted correction for whichever resume call the timed ring names.

CPU Final Fix02 — single correction candidate, 2026-09-28

Implementation boundary implements the Fix02 plan without restarting the CPU architecture: MSDC runtime clock ownership, sustained-idle parking and removal of the recurring screen-off background work, reconciled PWRAP readiness for 1196/1300 MHz, awake-proven retained PM diagnostics with a warm-reset backstop, charger-state discrimination and attributed USB transport faults. One preserving out/y2linux-cpu-final-fix02-candidate/ BOOTIMG/Y2ROOT package with the exact Hardware02 fallback; no flash/push. Next: the owner's single Fix02 run.

CPU Final Fix01 physical sweep — 2026-09-27

Admission audit records exact already-flashed runtime identity through USB SSH, taint0. Owner authorizes one observation/stress sweep using the existing harness, automatic independent continuation and bounded C3/suspend only after prerequisites. No implementation source edits, builds, flashes or pushes. The observation sweep is now **COMPLETE at safe physical-test scope, FAILED for CPU Final Fix01 acceptance**. See the physical report and closing audit. GPT6/GPT4/PPI29/highres/NO_HZ and all real QoS producers pass. Conservative OPPs/hotplug/WFI/playback/thermal authority remain functional. SLIDLE stays blocked by retained MMC runtime clocks; automatic parking is not observed. High OPPs now pass selector-mode recognition but fail PWRAP readiness. First devices-stage request loses recovery and SRAM does not identify the kernel boundary; later USB stress loses connectivity while UI remains usable. Two owner restarts occur, not demonstrated spontaneous resets. C3/full RTC/Power wake remain NOT_TESTED. No whole-platform promotion or external epic closure. Preserve exact fallback.

Next smallest boundary is one coherent Fix02 plan: prove early retained PM diagnostic safety, resolve actual PWRAP readiness, correct MMC runtime ownership without changing card rails/storage safety, reassess sustained-idle coordination, and fix the captured loaded USB recovery defect. The plan is not a new build, flash, memory-layout change or architecture workstream authorization.

CPU Final Fix 01 — focused implementation, 2026-09-27

Admission audit activates the owner's single correction batch using completed physical qualification. Fix GPT6/GPT4/PPI29 admission, complete Reborn QoS writes, decoded clock ownership and useful SLIDLE eligibility, actual Y2 PMIC selector mode, full suspend/wake and persistent stages. Preserve conservative paths and bounded CONSYS retry. Build one fresh out/y2linux-cpu-final-fix01-candidate/ preserving BOOTIMG/Y2ROOT package with exact fallback; no flash/push. No hardware acceptance or external epic closure.

Implementation now reaches the single-candidate build boundary: timer ownership, Reborn/worker leases, clock lifecycle/coordinator, both legitimate PMIC banks, persistent suspend stages and clean charger unwind have focused passing tests. Proceed with fresh broad software qualification; next physical run alone may accept highres/NO_HZ, actual SLIDLE/high OPP entry and same-boot RTC/Power wake.

Fresh software qualification is complete for runtime Linux0de6e95/Reborn36db186: 215 locked passes plus12 host native tests,192 Rust passes/fmt/strict Clippy, ARM/QEMU,382 ELF files/1396 dependency edges and105 pinned source packages. Seal one preserving BOOTIMG/Y2ROOT package with the exact Hardware02 fallback; physical acceptance remains pending the one SSH qualification sequence.

COMPLETE at software-candidate scope: one coherent CPU Final Fix01, BOOTIMG82b38fd3 and Y2ROOT26aa01ce, preserving validation/exact fallback PASS. See the implementation/validation receipt. Close this correction pass before owner-controlled flashing. Highres/NO_HZ, SLIDLE/high OPP entry, full RTC/Power wake and SRAM retention remain physical checks; no epic, automatic sleep or C3 qualification is promoted.

CPU Final physical qualification — 2026-09-27

Closing boundary audit and physical report record COMPLETE observation sweep, FAILED CPU Final acceptance. Exact flashed pair ff586df/afcf9ff matches; lower OPPs/schedutil/hotplug/WFI/kernel QoS, native silent playback and USB/WLAN data work. New GPT6/GPT4/PPI29 admission fails, highres/NO_HZ are inactive, Reborn device hint writes fail, high-bin voltage admission is gated, SLIDLE remains blocked and dormant/deadline/CIRQ entry unqualified.

Freezer and battery devices/platform/processors/core pass. Quiescent MUSB resume avoids storm; isolated CONSYS retry succeeds after a captured mandatory-STP failure. First full RTC suspend never returns; loss of USB/Wi-Fi recovery ends risky testing, owner restart restores clean filesystems and baseline runtime. No same-boot Power/RTC full wake or full device-restoration pass. Keep existing suspend qualification/C3-default-off controls, retain fallbacks, address the one prioritized fix batch. No source fixes/build/flash/push or external epic closure.

CPU Final candidate boundary — 2026-09-27

Boundary audit records completed CPU Final runtime source Linux ff586df / Reborn afcf9ff. Fresh kernel/config/modules, Buildroot/ARM, DT/BOOTIMG/memory, 205 locked regressions plus 12 native dependency tests, 189 Reborn tests/fmt/strict Clippy, ARM/QEMU, ELF/dependency and pinned source inventory pass. The DT validation-only correction 3f86af2 enforces CIRQ/MCU context resources and changes no compiled runtime code. COMPLETE at software-candidate scope: out/y2linux-cpu-final-candidate/. The local package/data/exact-Hardware02-fallback checks pass; no new hardware qualification. C1/C2/C3, stock-bin automatic DVFS, QoS, timer/context/CIRQ, suspend/device restore, diagnostics and recovery options are implemented. C3 remains experimental/default off, CPU0 with other cores physically off; the exact source report records all preflight/default limits. Retained hardware receipts and qualifications remain separate, with no new physical campaign, flash or push.

CPU Final — implementation pass, 2026-09-27

Admission audit uses retained real Hardware02 receipts. Owner authorizes complete CPU/timer/DVFS/ cpuidle/dormant/CIRQ/GPT/suspend software, Reborn semantic hints, bounded fallback and one fresh out/y2linux-cpu-final-candidate/ BOOTIMG/Y2ROOT package. Preserve Hardware Final work and Y2DATA; no flash/push. Implement and source-test independent paths before full build/package validation. Physical qualification stays independent; no long qualification campaign or hardware evidence acquisition is part of this pass.

Hardware Final — single campaign, 2026-09-27

The activation audit confirms installed Hardware 02 (Linux 76bc822 / Reborn 95747e0) through real USB SSH evidence. This owner direction supersedes Hardware 03/04/05 and older per-batch build/flash sequencing. No final capability is inferred from availability.

1. Complete Hardware 02 physical census and broad bounded measurements.
2. Investigate CPU/power/battery, storage/USB, radio/audio and recovery in parallel;

   implement all independently justified corrections on hardware-final.

3. Collect required physical actions while independent work continues. Preserve

   truthful electrical/calibration blockers and bounded fallbacks.

4. Audit accumulated changes, run complete current-source host/ARM/image checks,

   and create one y2linux-hardware-final-candidate preserving Y2DATA.

5. Stop for one owner installation/qualification cycle; destructive root OTA

   application requires explicit approval. Long endurance and unperformed physical cases remain explicit pending evidence, never assumed passes.

Campaign, capability ledger, and qualification own the new scope. Existing epics remain open. No intermediate image is justified at entry.

Active hardware capability ceiling campaign — 2026-09-26

Current boundary: Hardware 02 candidate handoff. Concrete package is built and release validated from Linux 76bc822 / Reborn 95747e0: source-backed SDR steps, Inventra DMA, GPT6/PPI29 local timer, tested Reborn sink repair, bounded failure handling and diagnostics. It is ready for one preserving owner flash with the exact installed Physical 01 pair as fallback. No new mode is physically accepted.

Recovered suspend evidence proves freezer/devices/platform/processors return, including CPU3/2/1; USB overflow and radio timeout prevent full restore. A fresh Power Menu reboot restored pinned SSH and clean storage/taint. RTC standard UTC write/read/ticking now pass; automatic-write policy, retention and alarm remain separate gates. Ordinary DNS works while one transient Wi-Fi observer timeout remains. Post-flash regression and measured 25→50 MHz qualification precede any capability promotion. Older notes retain their historical boundaries.

The owner's platform-wide ceiling request supersedes incremental feature bring-up. Fresh entry audit binds the work to installed Physical 01 (Linux 198fa7c, Reborn feb530f, candidate.3, boot 4991ba2c-0571-4649-9f7c-9e0318abb952). Exact versions match; IOS reads eight-bit eMMC and four-bit SD at 13 MHz/legacy, taint and ext4 error counters are zero. The initial regression gate passes, including checked Wi-Fi DNS/TCP, owner-confirmed wired audio and clean replacement-card workloads. Physical 01 is the working baseline; no final ceiling or freeze is accepted. Earlier current-target descriptions below are historical.

Continuation receipts 17–49 separate storage/query performance, identify stock CPU table 0 and physically prove GPT6/PPI29. The bounded probe restored controls and normal reboot returned taint 0. Reborn's PCM metadata repair passes host regression and temporary ARM SBC playback; owner confirms play/pause. Original binaries/session were restored. Receipt 49 supersedes the incomplete SSH-stream interpretation: CPU3/2/1 restart works, while USB and radio restore fail. The recovered unit is healthy after a confirmed fresh reboot; core/SPM entry remains unproven. Advance future suspend stages only after local and host recovery checks.

1. Retain the verified Physical-01 regression baseline; complete matched library,

   RAM/storage transfer, lifecycle and endurance measurements.

2. Investigate all hardware ceilings against retained stock/vendor and current

   physical evidence before selecting changes; preserve conservative fallbacks.

3. Hardware Batch 2: supported eMMC/SD timing, USB transfer/DMA, working local

   timer and observation. Hardware Batch 3: core/idle/suspend/RTC dependencies.

4. Coherent power, radio and audio batches follow their measured prerequisites:

   meter/pack evidence; stable SBC; real DL1 bit transport before higher rates.

5. Build/validate each coherent batch, then one owner flash and broad regression.

   Root OTA/rollback and endurance follow stable storage/power/core. Freeze only after the final report distinguishes qualified ceilings from evidence gates.

The owner confirmed menu, clean wired/AirPods tones and a known-good replacement SD card; Reborn AirPods play/pause works by owner report. No external USB meter is available presently. The ceiling report tracks these limits. Independent investigation continues. Existing issue owners remain; historical closed #30 does not mean physical power qualification.

Active physical bring-up campaign — 2026-09-26

The owner's complete campaign supersedes the earlier passive qualification scope. The entry audit reconciles current Telemetry 01 hardware evidence with every coverage area. Order: broad bounded census → root-cause clusters → coherent fixes → host/ARM/ candidate validation → necessary owner installation → broad regression; repeat. Independent tests continue after ordinary failures. Only the named safety and recovery conditions stop all hardware work. No planning row authorizes guessed SPM, storage clocks, charger limits, VBUS or audio formats.

Current physical target: Linux userspace 814c2f3, kernel/base d04b95a, Reborn 1556083, Telemetry 01; newer repository heads and original candidate are distinct. All historical failures and limited passes below remain evidence, with old authorization restrictions superseded only by this new explicit owner scope. The bring-up report and master issue table track progress. Existing issues stay open; no platform/endurance acceptance is declared.

The Fix Batch 1 admission audit refreshes all coverage after the broad real-device census. Implement the measured Wi-Fi checksum, legacy storage-width, benchmark/PSS, library query and Bluetooth pair/connect fixes as one candidate, then regress the whole baseline. Stock Y2 evidence supports eight/four data lines at the existing storage clock; higher clocks, timer/SPM activation, charger changes, USB VBUS and wider audio remain gated. The removed inconsistent FAT card stays isolated. Physical SBC pairing diagnosis is deferred at the owner's request and the Pairable window is closed; no pass is inferred from ACL Connected alone.

The packaging boundary audit refreshes all areas with receipts through 56. Linux 198fa7c / Reborn feb530f compile as one candidate; verify final host/ARM/package receipts and the exact running-image fallback before owner handoff. The new UI distinguishes a link, saved bond and audio readiness. Nine decoder suites also pass at the existing frequency caps, with original policy restored; this does not qualify sustained playback or justify governor changes. No milestone is closed.

2026-09-23. Owner-authorized software implementation, local commits only; no physical-device access or flashing. This roadmap does not close hardware epics. Entry revisions: Linux 5f6b4468fb43605ca1da679420823afa72cea73f, Reborn d9ba0549e6b34eff029bd13f7c33a491fee09e66.

Software freeze: built pair d04b95a / 6c8aa12, following the final boundary audit. All implemented target contracts below now have final current-source ARM/image receipts. Host orchestration/inventory tools remain host tools. No row has current candidate physical or endurance qualification. Optional hardware gates remain explicit and do not prevent this bounded software candidate.

Read the entry audit and completion ledger. Evidence levels are independent: IMPLEMENTED, HOST_TESTED, ARM_BUILT, IMAGE_VALIDATED, PHYSICALLY_QUALIFIED, ENDURANCE_QUALIFIED. Workflow labels never imply a higher evidence level.

Stock preloader/LK, BOOTIMG rescue, root/data split, protected partition policy, reserved memory, driver/CONSYS ownership, ALSA/DRM/evdev, FFmpeg, SQLite's single writer, BlueZ/BlueALSA and bounded workers remain authoritative. No new hardware or memory scope is authorized by this document. Advance each boundary only after updating the audit using retained physical evidence; do not repeat unchanged ROM/recovery provenance. Existing #30 acceptance is preserved, with later adverse suspend evidence and unqualified charger changes explicitly retained.

Reborn UI v1 application work — 2026-09-24

The owner now authorizes Reborn UI implementation against the frozen Platform v1 contract before physical platform acceptance. The activation audit retains every physical gate and issue status above. Work order: design/input and navigation; native music/queue screens; service-backed connectivity/system/advanced observations; focus/error/visual tests; fresh host/ARM/package validation; preserving UI candidate for owner review. No platform architecture or hardware expansion. Candidate creation and closure require another evidence audit. No flash.

The software candidate handoff audit records the final native previews, current-source host/ARM checks and preserving root package in out/y2linux-reborn-ui-v1-candidate/. Application implementation is complete at software-candidate scope. Platform v1 remains physically unqualified; the next boundary is owner Session A and the UI qualification checklist, then the relevant platform sessions. No installation or hardware scope is activated here.

The bounded graphics memory review permits a shared 4 MiB application font atlas and bounded collection-art worker, cache and texture. Target RSS/PSS, navigation latency and audio coexistence remain owner qualification items. Platform reservations and hardware gates are unchanged.

The UI candidate admission audit authorizes fresh ARM application compilation and preserving root packaging after host checks. Final image and runtime receipts remain required. Platform v1 compiled identities remain distinct from the application/package review heads.

Physical qualification entry — 2026-09-25

The owner identifies out/y2linux-reborn-ui-v1-candidate/ as the installed and intended qualification package. Authenticated versions metadata matches Linux d04b95aaff713edf943042d97a4c6134ca19fc24 / Reborn 155608393f6acfc6657f2f2e23cd07d0533479c6; local package integrity passes. The entry audit retains all hardware gates and the original plain-candidate identity mismatch. The stop audit records Session A FAIL; B/C/D NOT_TESTED. Health CLI dispatch crashes and status observation blocks on wakeup_count; retained kernel warnings trigger the owner's stop rule. Narrow root/data identity, clean counters, first-frame/main menu and USB SSH observations are retained in the physical qualification report. Next boundary: separately authorized corrective telemetry work and warning review, followed by Session A with the exact resulting installed identity. No production code/build/policy changes, flash, long unattended run or Session E/F/G operation is authorized by this entry; no capability is promoted.

Telemetry repair — 2026-09-25

The owner authorizes the focused repair: health CLI dispatch, bounded wakeup-counter observation, targeted regressions, warning review and a preserving root-only candidate. Flashing remains prohibited. Reborn, kernel, hardware policy and all physical acceptance gates remain unchanged. Next handoff requires host/ARM userspace and root-image preservation receipts; the installed candidate's failed qualification record remains historical evidence.

The candidate handoff audit now records 67 host and 5 packaged ARM/QEMU passes, exact five-file image changes, filesystem/scatter/inode preservation and the original UI fallback. Source 814c2f3, release 1.0.0-candidate.2, is packaged at out/y2linux-platform-v1-telemetry-01-candidate/. The warning review and repair report retains unresolved timer/cache/hardening and physical gates. Nothing was flashed; the installed image and its qualification status remain unchanged. Next is the separately authorized installation/Session A boundary, with stop rules intact.

The owner subsequently reports installing Telemetry 01. The physical re-entry audit resumes identity/baseline verification under the original test limits. Software receipts do not pre-approve the installed image or exempt existing warnings.

Fresh physical baseline evidence now verifies exact Telemetry 01 metadata/nine file hashes and both telemetry repairs on boot 3194fa9d-2dee-4105-86f0-4021580bf8d0. Status/health return and no wakeup reader remains. Root/data counters are clean and the owner sees the main menu with normal warmth. The retained kernel warning rule stops Session A; Wi-Fi health is FAILED/DEGRADED, so A FAIL; B–D NOT_TESTED remains the current session result. No platform physical/endurance acceptance; next resolve warning and Wi-Fi readiness gates without assuming new hardware or source authorization.

Read-only follow-up confirms no configured Wi-Fi networks, a responsive supplicant, and the existing 100 Hz periodic broadcast timer configuration. Intermittent Wi-Fi query delay remains unresolved. These findings refine the warning/readiness diagnosis without authorizing a build or changing acceptance.

The owner subsequently connected to their router and requests the narrow connection check. Association/IP/route/DNS and bounded router reachability may be observed without advancing the full qualification sessions or changing the network configuration.

That check now observes WPA2 association, a DHCP address/default route, 5/5 router and 3/3 Internet-IP ping replies. DNS queries for two hostnames fail; the platform observer also retains intermittent supplicant_unavailable. Wi-Fi coverage is PARTIAL with DNS FAIL, not Online or full Session C acceptance. Next diagnose DNS/readiness; no network settings or production code were changed.

Bluetooth optimization boundary — 2026-10-04

Owner explicitly authorizes the Bluetooth implementation/research pass in isolated codex/platform-v1-bluetooth worktrees from Linux b5aea93 / Reborn b71b468. Read-only locked SSH identifies installed Baseline02, Linux 8584ccd8, Reborn b71b4688, kernel 6.18.0-y2linux-baseline-02, root v1.21, boot 27e6d684-3794-4490-95db-a489027cd4f4, taint 0. This supersedes the *installed identity* in the older UART01 receipt; it does not constitute the separate Baseline02 cold-boot/C3 qualification. Retained UART01 C1/C2/C3 evidence remains bounded to its own image/boot. No headset is available, as confirmed by the owner.

F102/F103/F104 and #31 remain PARTIAL: the controller/BlueZ/BlueALSA are observable; peer pairing, audio, reconnect and endurance cannot be promoted by this pass. F105/F108/F229 retain peer/audio/coexistence evidence gates. #28/#31/#34 remain OPEN on read-only GitHub refresh. Idle measurement with Bluetooth off shows 50.37 Reborn Bluetooth thread context switches/s and repeated 2s BlueZ snapshots; these justify localized event/wakeup fixes, not a new kernel power architecture.

Proceed with signal-based observation, bounded-intent/recovery fixes, private-bus regressions, ARM userspace builds and temporary, restored runtime tests only when the nonblocking shared device lock is held. Preserve one reconnect owner and FFmpeg → shared DSP → final sink conversion → BlueALSA. No memory, firmware, calibration, protected storage, image packaging, flash, push, merge, release or milestone closure. The next acceptance boundary remains actual peer/audio and coexistence measurements on the final integrated artifact. Findings and final classification: Bluetooth optimization receipt.