Skip to content

ai-sessions/0101: the companion APK for CAP-079's questions — per-bud wear, stream lifecycle, write route, SetWallclock, Find negatives (docs only) - #75

Merged
tedsluis merged 3 commits into
mainfrom
crosscheck/0101-apk-wear-streams
Oct 24, 2026
Merged

tedsluis merged 3 commits into
mainfrom
crosscheck/0101-apk-wear-streams

Conversation

@tedsluis

Copy link
Copy Markdown
Owner

Documentation only — a read-only reverse-engineering session on the companion app v1.0.955078536-10253511 (ADR-017 boundary), cross-checked against all 117 capture logs. No code, no ADR, no FACT promotion, no new wire traffic. Branch from main (98ad4bf); independent of feature/0100-fixes (PR #70) — merge order is the maintainer's.

What this answers (REVERSE_ENGINEERING.md, four entries of 2026-10-18, every claim file:line)

  • (A) Which bud is in an ear. The official app's per-bud on/off-head state is the stream maestro_pw.Maestro/SubscribeToOobeActions (gav.java), bound for device type 6 (GetHardwareInfo field 1 = 6 in 82 logs). It is read only by the eartip fit test (hmw/hmv), the OOBE music demo, the OOBE-mode lifecycle (setting 6) and the OTA screens — no settings, status or battery screen; the call is open only while such a screen observes it. Runtime-info 7.3 = "Low latency mode"; 7.1/7.2 gate the live-dB subscription per bud. Wire: the method id occurs in CAP-078 only; there the call ended (status 14) 65 ms before the push of the bud-out and 16 s before the Buds' DISC — at the wear change (🟡 open only with setting 6 or 14 on).
  • (B) The life of a stream. The app ends every stream with a CLIENT_ERROR CANCELLED (nps.java:65–113) on a route change, when its last observer leaves and just before its own DISC, and answers packets of a call it no longer holds with FAILED_PRECONDITION. Wire: 374 such packets in 117 logs, 0 from OpenControl; CAP-037 cancel → DISC +7 ms. 🔴 whether the Buds stop a running stream on a cancel with the channel open.
  • (C) Field 21's core. Every maestro_pw.Maestro request goes to the primary route = the announced core; only the Dosimeter calls go per core. Wire: WriteSetting 1,044/1,044 and ReadSetting 8,852/8,852 on the announced channel; CAP-058's field-21 writes likewise. Nobody but OpenControl's F4 ever wrote to the other core (🟡 refused, 4/4; reads 🔴).
  • (D) SetWallclock and the day index. Sent once per connection, first thing, on the primary route (144 requests, all on the announced channel). The app never converts the day index to a date; it sums the seven entries up to the Buds' current index. Wire: the index runs in three regimes (bases 2026-07-29/30, 2026-02-13, 2025-06-30) while the Buds' clock was real time — 🔴 what it counts.
  • (E) Find. No ring/find method in the pw_rpc catalogue; goq.CASE (channels 18/23) is OTA transfer only and never on any wire; no write site for qhr 23/30/31/34–39; FE2C1238/109b862f absent; pbpctrl README:52 re-read. ADR-027 stands.

Documents changed (maintainer-approved in chat 2026-10-18, Checkpoint D)

  • PROTOCOL.md: five notes, no status change (§2.2a, §4.3 Option F, §4.5.5, §4.5.8a, §6) and a §8 row.
  • REVERSE_ENGINEERING.md: the four entries above.
  • TODO.md: the per-bud item answered (next: experiment E1); the CAP-079 questions → the notes and experiments E2–E4; the stream-cancel rule as a TODO line (the maintainer's choice, no draft ADR); the Find line and the ADR-060 candidate.
  • ai-sessions/: the 0101 RESULT (complete) and its INDEX row.

Verification

  • python3 scripts/lint_docs.py exit 0 (the "dead filename" lines for ai-sessions/0054 … 0091 are informational and present on main too); scripts/ensure_footers.py up to date.
  • Censuses reproducible with scripts/pwrpc_decode.py per log (commands quoted in the entries); the scratch scripts are not committed.

Details: ai-sessions/0101_CROSSCHECK_RESULT_2026_10_18.md.

🤖 Generated with Claude Code

tedsluis and others added 3 commits October 18, 2026 18:23
…s: per-bud wear, stream lifecycle, write route, SetWallclock, Find negatives

Why: CAP-079 (ai-sessions/0099) left five questions that the captures alone could not answer; the maintainer chose a read-only
reverse-engineering session (ADR-017 boundary) before the next feature round. Every finding is "this code exists and does this"
with file:line, cross-checked against all 117 capture logs; nothing is promoted, no ADR, no code change.

- REVERSE_ENGINEERING.md: four entries (2026-10-18) — the on-head state per bud (SubscribeToOobeActions behind gao/gav, read by
  the fit test, the OOBE wizard and the OTA screens only; runtime-info 7.3 = "Low latency mode"), the life of a server-streaming
  call (CLIENT_ERROR CANCELLED, FAILED_PRECONDITION replies, the using() disposers, the cancel-then-DISC order), the route of a
  write and the wall clock (the primary route; SetWallclock once per connection; the day index never converted, three regimes
  on the wire), the Find negatives (no ring method, goq.CASE = OTA only, no write of qhr 23/30/31/34–39, no FHN UUIDs).
- PROTOCOL.md: five notes, no status change (§2.2a, §4.3 Option F, §4.5.5, §4.5.8a, §6) and a §8 row — maintainer-approved
  in chat 2026-10-18.
- TODO.md: the per-bud item answered (next: experiment E1); the CAP-079 questions → the notes and experiments E2–E4; the
  stream-cancel rule as a TODO line (the maintainer's choice, no draft ADR); the Find line and the ADR-060 candidate updated.
- ai-sessions: the 0101 RESULT (complete) and its INDEX row.

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>
@tedsluis
tedsluis merged commit 425a3de into main Oct 24, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant