Repository navigation
1.3.0: the settings stream on the app's own channel (ADR-064/065) - #56
Merged
Merged
Conversation
…3 at every session (ADR-064/065) Once per MAESTRO session — at Connect and at every ADR-044 re-open — the Connect sequence sends one SubscribeToSettingsChanges and one ReadSetting 4:13 after SubscribeRuntimeInfo and before GetHardwareInfo, byte-identical to the official app's CAP-051 1177/1178 (channel 21) and CAP-056 1451/1450 (channel 19). Every SERVER_STREAM packet for a field the app reads goes through the same decoders as a read answer and replaces the value with the packet's receive time; setting 13 is the noise-control mode (0 = no mode settable, never shown as a mode; it also feeds the availability). A stream packet for a field the app does not read is logged payload-free (RoutedFrame.UnreadSetting), not an UnidentifiedFrame. Why: a press-and-hold on a bud changes the mode on the Buds, and the Message Stream's Notify reaches only the client holding DLCI 0x04 — in the user with Google Play services about half of the holds at best. The Buds push the same change on the app's own channel to a subscribed client (CAP-051 1717/1755/1802, CAP-038 2170/2205 with no Message Stream open), so the hold and the re-claim of ADR-061/062 are no longer needed; this file also drops them (ADR-065). Cause rules (ADR-065 item 4, the maintainer's choices in chat 2026-10-14): a stream value equal to the app's own Set within ANC_SET_WINDOW_MS (2 s) is the app's change; a differing push is "Changed by the Buds"; a differing read answer (the Get's Notify or the 4:13 RESPONSE) is "Changed since the app last saw it", also across a Connect — the cause is no longer reset at Connect. Nothing is written in reaction to a stream packet; no retry, no polling. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ices setting and the package query (ADR-065) Gone: the hold of the DLCI 0x04 claim while the noise-control tab is shown (ADR-061), the re-claim chain with the "Noise control with Google Play services" setting, its card, "Take it back again" and the hint line (ADR-062), the "Google Play services: found / not found" line, PlayServicesProbe and the manifest's <queries> entry (ADR-063). HoldReclaimMode, ChannelTaken, PlayServicesPresence and HoldReclaimSettingsStore are deleted; the stored key hold_reclaim_mode is no longer read — ignored, never migrated, never removed (DarkModeSettingsStore documents it and gains a DataStore constructor for the JVM test). The ANC (i) ends with the settings-stream sentence (the maintainer's text, hedged: the wear-driven switch stays a hypothesis) and carries the two rewritten cause lines; the Message Stream claim is ADR-032's again: a tap, the tile, Refresh, a pull and the Connect's snapshot, released after the linger. Why: with the settings stream (previous commit) the mode reaches the app on its own channel, so nothing needs the shared channel beyond a tap; the Play-user tug of war of 1.2.1/1.2.2 and its setting lose their reason (DECISIONS.md ADR-065, maintainer-approved in chat 2026-10-14, ai-sessions/0091). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…re-claim tests removed
Repository (FakeBudsTransport, runCurrent/settle): the Connect sequence's bytes
and order (fifteen requests); the ADR-044 re-open subscribing again, end to end
through connect() with a stub android.bluetooth.BluetoothDevice (the SDK stub's
package-private constructor); a session lost at the subscription not retried; a
push shown as changed by the Buds with its time, nothing sent, no claim; value 0
never a mode; a differing read of 13 ("changed since"), the CAP-051 2247 case;
the own Set in CAP-051's order, before the ACK and after a cut-off answer; the
CAP-056 wear sequence; the other fields from the stream with the own-write
mirror keeping its mark; an unread field logged and ignored; nothing held or
re-claimed. The 0089 cause tests rewritten for one rule across a Connect; the
ADR-061/062 blocks deleted. UI: the (i) text, the two cause lines, no card, no
hint, no button, no Play line. A 1.2.2 settings file with hold_reclaim_mode is
read and written without touching that key.
Mutation checks, one at a time: M1 the subscription once per process, M2 0 as
Off, M3 the own Set's stream packet as the Buds' change, M4 the stream to
UnidentifiedFrame, M5 an EQ stream packet ignored — each caught (2/4/1/2/6
failing tests).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…essions/0091) DECISIONS.md: ADR-065 (maintainer-approved in chat 2026-10-14) supersedes ADR-061/062/063 and amends ADR-064 items 2 and 6; their Status lines and pointer Updates on ADR-032 and ADR-064; id_registry.csv accordingly. AGENTS.md §1: the approved pointer (the package query is gone). ARCHITECTURE.md §2, §2.4, §3.1, §3.2 (ANC_SET_WINDOW_MS; HOLD_RECLAIM_* gone), §5a, §6.0b, §7, §9. The CAP-076 skeleton, Group BL, its index and registry rows and APP_TESTPLAN.md §Y rewritten as the 1.3.0 run in both users (holds with the tab closed and open, the own tap, the wear-change re-home, Bluetooth off/on, taps while Play holds the channel, the counts for the verdict). CHANGELOG [1.3.0], README, PROJECT, the release-notes template, versionName 1.3.0 / versionCode 10300, TODO.md, INDEX, the session's RESULT. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…(the 1.3.0 run in both users) The maintainer ran scripts/release.sh 1.3.0 (build commit ff27ee1), installed the APK over 1.2.2 in both GrapheneOS users and recorded CAP-076 in part (only the changed functionality). This prompt lets a fresh session analyse the two recordings, four HCI logs, two exports, two app logcats, the system log, the full logcat and the adb shell log per the project's rules (reading first, resumable phases, no sampling, evidence traceability, approvals in chat), and propose the release verdict (RELEASING.md C3/C4). Why a separate session: the analysis is a CAPTURE task of its own, and this session's context is spent. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ysed (ai-sessions/0092) Why: RELEASING.md C3 — the hardware run of the release build ff27ee1 must be analysed and committed on the release branch after the build commit before the verdict (C4). The run shows the settings stream of ADR-064/065 working in both users: every press-and-hold on the noise-control tab shown within 0.1 s (29/29 with Google Play services holding the shared channel, 12/12 without), the two new requests once per MAESTRO session, byte-identical. The folder renamed to the films' first and last phone times; the films, HCI logs, exports and app logcats as delivered (Git LFS); the two phone-wide logs kept local and committed as the redacted extract (serials removed); the read-only adb checks of 07:52 (the installed APK's SHA-256 = the tested build's) and a sha256sum of the data files. EVENT-NOTES gets the timeline of both parts (the skeleton kept as the appendix); FINDINGS is new. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Why: PROJECT_RULES.md rule 4a — a script behind a count is committed. pwhdlc_crc_table.py (every pw_hdlc frame of DLCI 0x02/0x03 CRC-32 checked, as a table — pwrpc_decode.py does not check the CRC), message_stream_claims.py (the DLCI 0x04 claim inventory of both clients), stream_events.py (the settings stream's field-13 pushes merged into change events — the Buds push each change twice with both buds out) and film_mode_buttons.py (the selected mode button read at 10 fps from a screen recording). Each reproduces the analysis's results byte for byte. No app file changed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…, Known issues Why: the maintainer's decisions on CAP-076 (chat 2026-10-15, ai-sessions/0092): the verdict "release as built + Known issues" (C4 minor issue) and the approved texts. DECISIONS.md ADR-065 Update (hardware-verified, with the numbers and the three minor defects); PROTOCOL.md §4.1 note and §8 row (no status change); ARCHITECTURE.md §5a/§6.0b; CHANGELOG.md [1.3.0] Known issues (the "changed since" line after a reconnect; the tap/collision line reworded); TODO.md (CAP-076 done, the Buds' state, P1–P6 recorded, no FEATURE prompt now); APP_TESTPLAN.md §Y results; TESTPLAN Group BL; the CAPTURE index row and Group BL status; id_registry.csv; PROJECT.md status; the sidebar's new path. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Why: AI_SESSION_LOG_PROCEDURE.md — the session's record (the phases, the checkpoints verbatim, the files read, the deferred items); INDEX row 0092 and the 0091 back-fill "hardware run analysed by 0092". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
1.3.0 (
ai-sessions/0091, 2026-10-14): the app learns the noise-control mode — and every other setting it shows — from the Buds' settings stream on its own channel (DLCI 0x02), so a press-and-hold on a bud shows on any tab, with or without Google Play services. BuildsDECISIONS.mdADR-064 and the new ADR-065 (both maintainer-approved in chat 2026-10-14).SubscribeRuntimeInfoand beforeGetHardwareInfo: oneSubscribeToSettingsChangesand oneReadSetting 4:13, byte-identical to the official app'sCAP-0511177/1178 (channel 21) andCAP-0561451/1450 (channel 19). No retry, no polling; nothing written in reaction to a stream packet. The only new bytes on the wire are these two requests.SERVER_STREAMpacket for a field the app reads (2, 4, 7, 11, 12, 13, 15, 16, 17, 19, 22, 27, 28, 29) goes through the same decoders as a read answer and replaces the value with its receive time; setting 13 = the mode (0 = no mode settable, never shown as a mode; it feeds the availability). A packet for a field the app does not read is logged payload-free, not anUnidentifiedFrame.Setwithin 2 s → the app's change; a differing push → "Changed by the Buds at HH:MM:SS: the Buds reported a new mode — a press-and-hold on a bud, or a change they made themselves."; a differing read answer (theGet'sNotify, the4:13RESPONSE) → "Changed since the app last saw it (read at HH:MM:SS): …", also across a Connect. The ANC (i) ends with the maintainer's text (the wear-driven switch stays a hypothesis).Removed (ADR-065 supersedes ADR-061/062/063): the hold of the shared channel while the noise-control tab is shown, the re-claim chain, the setting "Noise control with Google Play services" with its card, "Take it back again", the "Google Play services: found / not found" line,
PlayServicesProbeand the manifest's<queries>entry. A storedhold_reclaim_modeis ignored (never migrated). The Message Stream is claimed only for a tap, the tile, Refresh, a pull and the Connect's snapshot, as before 1.2.0.Evidence (real bytes, in the fixtures):
CAP-0511177/1178/1225/1717/1755/1802 and the own tap 1527 → 1529 → 1530/1533 → 1534 (269 ms);CAP-0561451/1450/1468/1696/2175/2259/2324/4025/4028;CAP-0382170/2205 (no Message Stream open);CAP-0452033/2277 (value 0);CAP-0211089/1842;CAP-027923;CAP-0692502;CAP-0152116;CAP-0361517/1519.Checks
cd android && ./gradlew assembleDebug testDebugUnitTest test lint --max-workers=2: BUILD SUCCESSFUL, 0 failing tests (:data1672,:hardware64,:domain42,:ui92 per variant), 7 lint warnings, no error;python3 scripts/lint_docs.pyexit 0.UnidentifiedFrame(2), M5 an EQ stream packet ignored (6) — all caught; the sources restored and verified.<queries>, noINTERNET; no build-script or dependency change (:app:dependencies --configuration releaseRuntimeClasspathchecked for network/telemetry SDKs: none); the AGPL header on every new and changed Kotlin file; no hidden API; every new log line payload-free, no MAC or serial.bb7a843feat,7756e55refactor,152c5b3test,8a72572+391b0c5docs. The gate ran on the final tree.Known issues of 1.3.0 (
CHANGELOG.md)Release checklist (
RELEASING.md)A. Prepare
feature/settings-streamfrom an up-to-datemain(c1d898b).versionName = "1.3.0",versionCode = 10300.CHANGELOG.md[1.3.0](no date yet),README.md,scripts/release_notes.template.lint_docs.pyexit 0 (above).B. Build and verify — from the pushed branch tip
scripts/release.sh 1.3.0on the branch tip (2026-10-15) — build commit:ff27ee1; APK SHA-25603272c6f954316d76c3ef6d1d020abd8bdb97755d7f8e165c84d3f16a9c6a96c; 136 third-party libraries.a7530f5c…d8dcb=README.mdline 19 /SECURITY.mdline 40 (checked 2026-10-15).cp -a dist/1.3.0 ~/opencontrol-1.3.0-tested.C. Test this exact APK
CAP-076(Group BL, the rewritten skeleton undercaptures/): holds with the noise-control tab closed and open in both users, the own tap and the tile, the wear-change re-home, Bluetooth off/on, taps while Play holds the channel; the counts for the verdict.ai-sessions/0092, commitsc32b297…b06181e(CAP-076-FINDINGS.md).CAP-076-FINDINGS.md§12): every hold on the noise-control tab shown (29/29 with Play, 12/12 without); Known-issues lines inCHANGELOG.md[1.3.0]and the keptrelease-notes.md.D. Merge and publish — D1 merge commit, D2 signed tag on the B1 commit, D3 the draft release from the kept copy, D4 the phone. E. After publishing — E1–E4 by a small pull request of its own.
Session record:
ai-sessions/0091_FEATURE_RESULT_2026_10_14.md.🤖 Generated with Claude Code