[gdluck]

Requested by A

Review follow-up (GreenGrassBlueOcean, 2026-11-04) — applied in edf9ea8

The rule is understood: no change to decoded values, log strings or accepted builder domains without committed on-wire plant traces. Where the original description below still describes a reverted item, the diff and the CHANGELOG are the truth.

Blocker	Action taken
1. turn_on_with_source	The runtime DeprecationWarning and import warnings removed; the WHAT#MULTIMEDIA_TYPE#AREA vs source ambiguity is an evidence-boundary note in the docstring only (frame, log and validation as on master).
2. turn_off_atomic / route_source / source_address	Everything back to master: no area= parameter or area helpers, turn_off_atomic("3#2#1") -> *22*0#4#0*3#2#1##, route_source default matrix_area=1 and master's log, no ValueError on route_source("2#1", 7) / ("6", 7), source_id / source_address not bounded to 1..9, no raise on 3#12#1.
3. OWNSoundDiffusionEvent	requested_source on WHAT 1, preset, frequency_khz, human_readable_log and the area fallback back to master; the ~20 new decoders / properties (DIMENSION 2, 3, 4, 7, 17, 18, 19, 20, 21; WHAT 2, 22, 36..43, 55, 56) and builders (turn_on_atomic, select_station, request_station, store_station, set_equalizer_preset, request_equalizer_preset, next_preset, previous_preset) removed; master's isdigit() guards on WHERE / WHAT parameters, which the branch had dropped, are back.
4. WHO 16	All runtime changes reverted (set_volume 0..100, volume not None above 31, select_source("20", 2) rejected again, WHERE 100 is_source_event / log as master, the new WHAT properties removed); tests/test_sound.py and tests/test_sound_diffusion.py are byte-identical to master; tests/test_sound_audit_fixes.py deleted (its remaining tests asserted branch wording).

Kept: docstring notes only, on volume_up / volume_down (bare step-1 vs 3#STEP, the bt_multi oracle rows), turn_on_with_source (the ambiguity, written as an evidence boundary, not as a deprecation), send_rds_text (the bt_multi NACK note plus the absence of a client dimension 10 write in the published grammar) and open_paging / start_paging_voice (trace provenance). No crash-guard was kept because the branch added none: the dimension 5 / 11 trailing-empty-field change was a decoded-value change (None -> 101500), not a crash fix, so it was reverted. Every changed line in sound.py is a docstring line; frames, decoded values, logs and accepted domains are master's.

Committed plant evidence. There is no physical sound-plant trace for any of the reverted items beyond what master already pins; the MyHOME oracle_who22_tuner fixtures are emulator-derived, not on-wire. The only new WHO 22 evidence in this round is the bare-power-frame crash, which goes through #163 with its own oracle suite and golden pin. If a physical WHO 22 / WHO 16 capture would be worth having, say which frames; the capture instructions can be prepared for any user with a sound plant.

Summary

Before: in *22*WHAT#4#X*3#A#P## for WHAT 0 / 1 / 2 / 34 OWNd reads the second parameter X as a source, so turn_off_atomic hard-codes 0#4#0 and route_source defaults the target's area to 1 (misrouting area-2 speakers); the dimension 5 / 11 frequency decode applies the FM 10 / 100 kHz heuristic to AM bands and crashes on a trailing empty value; "preset" names the stored station (dimension 6) while the page's preset is dimension 19; several dimensions and WHATs the page defines are logged as unknown; some builders emit frames outside the published grammar without saying so; WHO 16 set_volume accepts 0..100 (page: 0..31, volume 50 refused by all twelve emulated gateways), select_source refuses amplifiers 20..90 (the six WHO 16 gateways accept them), WHERE 100 (all sources) is logged as a zone and the page's WHATs are "received command: N".

After: the second parameter is exposed as what_area (and requested_source only for WHAT 35), turn_on_atomic / turn_off_atomic / route_source derive the area from a 3#A#P / 4#A / 5#3#A#P target (explicit area= allowed, range 0..9 checked; 2#S / 6 targets keep master's area 0 since no area can be derived); the tuner decode is modulation-aware (AM in kHz marked as an inference) and never raises; station (dimension 6) and preset (dimension 19) are distinct with the old names kept as deprecated aliases; the page's dimensions 2 / 3 / 4 / 7 / 17 / 18 / 20 / 21#x and WHATs 5 / 6 / 22 / 36..43 / 55 / 56 decode with names; undocumented builders say so in their docstrings and source_address accepts 1..9 (published 1..4, 5..9 plant evidence); WHO 16 volume 0..31 (replies above 31 give None), amplifiers [1-9][0-9], WHERE 100 = all sources (is_source_event), and the page's WHATs carry is_* properties while the log keeps the "received command: N (meaning)" text MyHOME pins.

Context and how it was found. A asked for OWNd master to be checked against the OpenWebNet Encyclopedia functional pages (audit of 2026-10-28; 94 rows reproduced by running the code; confirmed by a firmware-oracle replay on the twelve emulated gateways and a whole-Encyclopedia search). This is the WHO 22 / WHO 16 group (rows D1..D7, S1..S4), built on #153. Oracle confirmation: *22*35#4#1#7*3#2#1## produces area-1 bus bytes and 35#4#2#7 area-2 bytes, so a defaulted area 1 misroutes area-2 speakers on nine gateways; WHO 16 volume 50 refused 12/12, 31 out on 6; *16*3*20## accepted by the six WHO 16 gateways; the bare *22*3*W## step sends step 00 on F454 / F459 / MyHomeServer1 and the same bytes as 3#1 elsewhere. The area reading of the second parameter is corroborated by the plant captures in the Encyclopedia's sound-matrix-routing review (*22*2#4#1*5#2#1## = area 1, source 1); the oracle could not distinguish the two readings (gateways copy the parameter verbatim), so turn_on_with_source is not changed, only deprecated.

Changes

• OWNd/message/sound.py, WHO 22 (D1..D7, who-22-sound-diffusion/README.md "WHAT", "Historical touchscreen syntax", "Payloads", "Core parameters", "DIMENSION"): what_area for WHAT 0 / 1 / 2 / 34, requested_source for WHAT 35 only; turn_on_atomic (new) / turn_off_atomic / route_source with _validate_area (derived from the target or explicit, 0..9); turn_on_with_source(where, source_id, multimedia_type=4) keeps master's frame and emits a DeprecationWarning pointing to turn_on_atomic + route_source; modulation-aware dimension 5 / 11 decode; station vs preset (dimension 19) with deprecated aliases; volume_up / volume_down keep master's bare *22*3*W## / *22*4*W## for step 1 (fix(sound): reject non-positive step values in OWNSoundDiffusionCommand.volume_up/down #127 kept that form on purpose) and 3#STEP / 4#STEP above; named decoders for the page's dimensions and WHATs; source_address 1..9; send_rds_text, set_preset(..., freq), open_paging marked as outside the published grammar in their docstrings, not removed.
• OWNd/message/sound.py, WHO 16 (S1..S4, who-16-sound-system/README.md "WHERE", "Dimension values", "WHAT values"): set_volume 0..31, replies above 31 -> None; select_source amplifiers [1-9][0-9]; WHERE 100 = all sources; is_* properties for WHATs 20 / 23 / 30 / 33 / 40 / 50 / 53 / 101 / 102 with the log text "received command: N (meaning)"; module docstring "Sound System".
• CHANGELOG.md: [Unreleased] Added / Changed / Fixed bullets with the frames and sections; OWNd/message/__init__.py exports.

State Diagram

stateDiagram-v2
    direction TB
    state "WHO 22 area derivation for turn_on_atomic / turn_off_atomic / route_source" as A {
        [*] --> FromTarget : 3#A#P, 4#A, 5#3#A#P -> area A (0..9 checked)
        [*] --> Explicit : area= given -> used, 0..9 checked
        [*] --> SourceTarget : 2#S or 6 -> turn_on / turn_off keep area 0, route_source needs area=
        FromTarget --> Frame : *22*WHAT#4#AREA*TARGET## / *22*35#4#AREA#SOURCE*TARGET##
    }
    state "WHO 16 volume / amplifier" as V {
        [*] --> Volume : set_volume 0..31 -> *#16*W*#1*V##, reply above 31 -> None
        [*] --> Amplifier : select_source amplifier 01..99 (page WHERE table)
        [*] --> AllSources : WHERE 100 -> is_source_event
    }

Loading

Where are the traces

Subsystem	Capture	Plant	Frames verified by this PR
WHO 22 area vs source	OpenWebNet-Encyclopedia reverse-engineering/sound-matrix-routing.md plant captures; MyHOME#839 F500 traces (H/L4684 6.0.11)	sound plant, F500 tuner	*22*2#4#1*5#2#1## = area 1 / source 1; every OWNSoundDiffusionCommand call of MyHOME#839 keeps its signature and frame
WHO 22 / 16 builders	own-firmware-oracle ownd-audit-frames replay (2026-10-28, emulated, two identical runs) and tests/golden/firmware_oracle.json	twelve emulated gateways	35#4#1#7 vs 35#4#2#7 area bytes, volume 50 refused 12/12, amplifier 20 accepted on the six WHO 16 gateways, bare *22*3*W## step 00 on three

What does the Oracle say

Frame	Result
*22*35#4#1#7*3#2#1## / *22*35#4#2#7*3#2#1##	area-1 / area-2 telegram bytes on 9 gateways (area derived from the WHAT parameter)
*22*0#4#0*3#2#1## vs *22*0#4#2*3#2#1##	both forwarded, parameter copied verbatim: the reading is not distinguishable on the bus, hence the deprecation rather than a change
*#16*21*#1*50## / *#16*21*#1*31##	refused 12/12 / out on 6
*16*3*20##	accepted by the six WHO 16 gateways
*22*3*3#1#1## vs *22*3#1*3#1#1##	step 00 vs 01 on F454 / F459 / MyHomeServer1, identical bytes on the other six (#127 kept the bare form)

Testing

• tests/test_sound_audit_fixes.py (new, 54 tests) plus updates in tests/test_sound.py and tests/test_sound_diffusion.py; pytest.deprecated_call on turn_on_with_source (the project turns DeprecationWarning into errors).
• Full suite: 2855 passed (plus the environment-dependent test_runtime_version_matches_installed_package); scripts/verify_coverage.py 7,297 / 7,297, sound.py 100 % statements and branches (master had 11 partial branches); ruff check, mypy OWNd, scripts/verify_library_standards.py, scripts/sync_documentation.py --check pass.
• MyHOME v2-phase1-architecture suite against this branch: 5187 passed, 1 deselected (the pin-bump doc test).

Precedent check

Kept from #153: store_station 1..5, steps 5001.. / 5101.. and 5#STEP / 6#STEP, bare *22*5*W## = auto-seek, no WHAT 7 / 8, no dimension 10 decode, the ownkb:* ids; from #127: the bare single-step volume frame (the 3#1 change from the first revision was reverted); from #128: volume bounds on WHO 22; profile wording untouched. Changed back after review: derived areas range-checked, source_address 1..9, 2#S / 6 targets do not raise, turn_on_with_source deprecated instead of silently changed; the generic "received command: N" log text kept for MyHOME's pin.

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[gdluck]

Fixed the balance magnitude (commit e173498, normal push). A spec-vs-code check flagged that the balance log and CHANGELOG said 030 is "left 30" while this PR's own balance docstring, and the Encyclopedia (who-22-sound-diffusion#tone-balance-and-presets: "remaining digits divided by 3 give magnitude", tests 030 -> left 10, 115 -> right 5), say left 10. balance_level now returns the divided-by-3 magnitude so the log reads "left 10" / "right 5"; the raw string stays on balance ("030"). The touchscreen steps balance in threes, so the division is exact for the values it emits. Tones stay raw 1..63 (the published domain; the raw/3-10 form is the power-amp UI only, per the same page). balance_level and balance are not read by MyHOME. Audit tests 70 passed, full OWNd suite green, MyHOME sound tests 203 passed.

[gdluck]

Correction on D7, following a re-check against the official Legrand WHO 22 v1.1 PDF (commit 06c51a1, normal push). My earlier "not in the published grammar" for the dimension 11 write was wrong: §3.4 "Dimension writing" → 3.4.1.1 "State Memorized Station - Dimension = 11" documents the client write *#22*5#2#S*#11*MOD*FREQ*ST## with an Ack. So set_preset(..., frequency_khz=) builds a published frame; the Encyclopedia functional summary just omits it (worth a note upstream). The docstring, CHANGELOG and the D7 test now cite §3.4.1.1, and keep only the honest caveat that the effect on a physical tuner is not yet confirmed (consistent with the WHO 22-only tuner path noted in MyHOME #839). The genuinely undocumented builders here (RDS-text dimension 10, WHAT 19/20 paging) are unrelated and still marked as such. Sound tests 227 passed, ruff and mypy clean.

[gdluck]

One finding from the 2026-11-04 cross-check that lands squarely in this PR's area. The bare WHO 22 power frame, WHAT 0 / 1 without #mmtype#area (*22*1*WHERE##, *22*0*WHERE##), crashes the sound daemon on F461 020011 and F454 020051 on every target type (source 2#S, amplifier 3#A#P, area 4#A, all sources 6: all nack/crash in the firmware oracle, two byte-identical runs each), while MH202 refuses it without crashing. The published forms, *22*1#<mmtype>#<area>*…## / *22*0#<mmtype>#<area>*…## (WHO 22 PDF 3.1.1 / 3.1.2 / 3.2.1, <area> the output area 1..9), go out (ack/out) on every target.

That also confirms, from the PDF, the reading this PR already takes in turn_on_atomic / turn_off_atomic: the second parameter is the area, not the source. The pre-b13 turn_on_with_source writes the source into that slot, so the fix is to point callers at the atomic builders rather than add an OFF twin of turn_on_with_source.

OWNSoundDiffusionCommand.turn_on / turn_off still build exactly the bare form. I've prepared a single commit on top of this branch (fix/who22-bare-power-guard-on-158) that refuses the bare frame in the send layer the way the WHO 4 dimension 22 and WHO 18 dimension 52 crash frames are refused, deprecates turn_on / turn_off (they keep their frame for callers that pin it, with a DeprecationWarning pointing to the atomic builders), and adds no new builder. OWNCommand.parse stays permissive. Full suite and MyHOME's conformance suite pass.

MyHOME is unaffected: it powers audio zones through WHO 16 (OWNSoundCommand.turn_on -> *16*3*WHERE##, turn_off -> *16*13*WHERE##) and never builds the WHO 22 form. Happy to push it onto this PR, or open it as a follow-up stacked on #158 — your call.

[GreenGrassBlueOcean]

Thanks for the WHO 22 / WHO 16 audit and for carefully recording where the firmware oracle could and could not distinguish parameters (ORACLE_CONFIRM D1, NOT DISTINGUISHED).

Under our standing rule — if we don't have real, committed on-wire plant traces from physical hardware, we don't change behaviour — virtually all runtime changes in this PR rest on WHO 22 / WHO 16 specification tables, archived touchscreen code (libqtdevices), or empty-bus gateway forwarding rather than real physical sound-system captures, and several break existing master contracts and tests.

What can stay

1. Docstring evidence-boundary notes: Recording the #127 / bt_multi step-1 vs. 3#1 oracle notes on volume_up / volume_down, the bt_multi NACK note on send_rds_text, the paging trace provenance notes on open_paging / start_paging_voice, and the WHAT#MULTIMEDIA_TYPE#AREA (ownkb:claim:c002727) vs. turn_on_with_source ambiguity in docstrings.
2. Pure parser crash-guards (if any malformed/truncated frame previously raised uncaught ValueError / IndexError during OWNMessage.parse), without changing decoded values on valid frames.

---

Blockers — Behaviour changes without committed plant traces (please revert)

#	Symbol / File	Behaviour change without real plant trace	Required action
1	OWNSoundDiffusionCommand.turn_on_with_source in OWNd/message/sound.py	Emits a runtime DeprecationWarning (warnings.warn(...)). Because OWNd and downstream suites run with filterwarnings = ["error"], emitting DeprecationWarning is a breaking runtime change — and the docstring itself notes ORACLE_CONFIRM D1, NOT DISTINGUISHED (gateways copy the nibble verbatim onto the bus and cannot settle area vs. source without physical sound hardware).	Remove warnings.warn(..., DeprecationWarning); keep the explanation in the docstring only.
2	OWNSoundDiffusionCommand.turn_off_atomic, route_source, & source_address in OWNd/message/sound.py	Changes turn_off_atomic("3#2#1") wire output (*22*0#4#0*3#2#1## -> *22*0#4#2*3#2#1##), changes route_source("3#2#1", 7) default matrix_area (1 -> 2) and human_readable_log, newly raises ValueError on route_source("2#1", 7) / route_source("6", 7) without area=, restricts source_id / source_address to 1..9 (source_address(10) now raises), and raises on multi-digit areas like 3#12#1 — all without multi-area physical sound-plant traces confirming speaker/matrix behaviour (ownkb:claim:moc-c007654 explicitly cautions that libqtdevices does not validate every WHAT/WHERE combination).	Revert turn_off_atomic, route_source, and source_address wire output, parameter defaults, validation bounds, and logs to master.
3	OWNSoundDiffusionEvent existing properties (requested_source, preset, frequency_khz, human_readable_log) & ~20 new speculative builders/decoders in OWNd/message/sound.py	Drops requested_source to None on WHAT 1, repurposes preset to DIMENSION 19 equalizer preset, decodes AM frequencies in kHz (explicitly noted as "an inference, the page gives the bands but no AM unit"), and adds speculative builders/decoders for equalizer/tone/balance/loudness/3D/play-status (DIMENSION 2, 3, 4, 7, 17, 18, 19, 20, 21, WHAT 22, 36..43, 55, 56, turn_on_atomic, select_station, set_equalizer_preset, etc.) with zero physical sound-plant traces (DIMENSION 19 is not even in the firmware oracle).	Revert OWNSoundDiffusionEvent changes and hold back unverified WHO 22 builders/decoders until backed by real sound-system bus captures.
4	OWNSoundCommand & OWNSoundEvent (WHO 16) in OWNd/message/sound.py	Narrows set_volume from 0..100 to 0..31 and drops OWNSoundEvent.volume to None above 31 (forcing edits to tests/test_sound.py where volume == 37 and set_volume("21", 40) were pinned!), loosens select_source("20", 2) (reversing an explicit pytest.raises(ValueError) assertion in tests/test_sound.py), changes WHERE 100 is_source_event / subject log, and adds speculative WHAT 20/23/30/33/40/50/53/101/102/2001..2115 properties (forcing test_unknown_sound_command_has_diagnostic_log to change from *16*30*22## to *16*60*22##), all with zero physical WHO 16 plant traces.	Revert all WHO 16 runtime changes in OWNSoundCommand and OWNSoundEvent to master.

[gdluck]

Applied in edf9ea8 (pushed; the description now carries the same follow-up table at the top).

Blocker	Action taken
1. turn_on_with_source	The runtime DeprecationWarning and import warnings removed; the WHAT#MULTIMEDIA_TYPE#AREA vs source ambiguity is an evidence-boundary note in the docstring only (frame, log and validation as on master).
2. turn_off_atomic / route_source / source_address	Everything back to master: no area= parameter or area helpers, turn_off_atomic("3#2#1") -> *22*0#4#0*3#2#1##, route_source default matrix_area=1 and master's log, no ValueError on route_source("2#1", 7) / ("6", 7), source_id / source_address not bounded to 1..9, no raise on 3#12#1.
3. OWNSoundDiffusionEvent	requested_source on WHAT 1, preset, frequency_khz, human_readable_log and the area fallback back to master; the ~20 new decoders / properties (DIMENSION 2, 3, 4, 7, 17, 18, 19, 20, 21; WHAT 2, 22, 36..43, 55, 56) and builders (turn_on_atomic, select_station, request_station, store_station, set_equalizer_preset, request_equalizer_preset, next_preset, previous_preset) removed; master's isdigit() guards on WHERE / WHAT parameters, which the branch had dropped, are back.
4. WHO 16	All runtime changes reverted (set_volume 0..100, volume not None above 31, select_source("20", 2) rejected again, WHERE 100 is_source_event / log as master, the new WHAT properties removed); tests/test_sound.py and tests/test_sound_diffusion.py are byte-identical to master; tests/test_sound_audit_fixes.py deleted (its remaining tests asserted branch wording).

Kept: docstring notes only, on volume_up / volume_down (bare step-1 vs 3#STEP, the bt_multi oracle rows), turn_on_with_source (the ambiguity, written as an evidence boundary, not as a deprecation), send_rds_text (the bt_multi NACK note plus the absence of a client dimension 10 write in the published grammar) and open_paging / start_paging_voice (trace provenance). No crash-guard was kept because the branch added none: the dimension 5 / 11 trailing-empty-field change was a decoded-value change (None -> 101500), not a crash fix, so it was reverted. Every changed line in sound.py is a docstring line; frames, decoded values, logs and accepted domains are master's.

Committed plant evidence. There is no physical sound-plant trace for any of the reverted items beyond what master already pins; the MyHOME oracle_who22_tuner fixtures are emulator-derived, not on-wire. The only new WHO 22 evidence in this round is the bare-power-frame crash, which goes through #163 with its own oracle suite and golden pin. If a physical WHO 22 / WHO 16 capture would be worth having, say which frames; the capture instructions can be prepared for any user with a sound plant.

[GreenGrassBlueOcean]

Pull Request Review: PR #158

Verdict: REQUEST CHANGES / RECOMMEND CLOSING
Resolution on core question: Does this solve a real problem? No.

---

1. Executive Summary & Problem Assessment

This PR does not solve an active production problem.

• In its current state: It is a hollowed-out PR. Following the reversion in edf9ea8, the diff touches only two files, introducing 42 lines of docstring commentary across four methods in OWNd/message/sound.py and a 3-line changelog addition. Furthermore, it sits in a broken merge state (mergeable_state: dirty) due to an unresolved conflict in CHANGELOG.md.
• In its original state: It did not solve a problem either; it introduced multiple regressions. The initial commits attempted a speculative rewrite of WHO 22 (Sound Diffusion) and WHO 16 (Sound System) based solely on reading documentation tables. Merging those changes would have broken Home Assistant volume sliders, raised unexpected ValueErrors in automations, and degraded parser sanitization.
• The one genuine issue unearthed during this investigation: The discovery that bare WHO 22 power frames (*22*1*WHERE## / *22*0*WHERE##) crash gateway daemons on 8 out of 12 firmware models is a real issue. However, that fix does not live in this PR—it was properly isolated and implemented in PR #163 with dedicated oracle test suites.

---

2. Protocol & Architecture Review

The "Spec Audit" Anti-Pattern

This PR began as an audit against OpenWebNet documentation tables and manufacturer PDF specifications. In OpenWebNet development, the epistemic hierarchy must always be:

$$\text{Spec PDF} < \text{Disassembly / Daemon strings} < \text{Firmware Oracle (Emulated)} < \text{Committed On-Wire Plant Traces}$$

Manufacturer documentation frequently contains draft definitions, copy-paste errors, and syntax that never shipped to physical hardware. Modifying runtime decoding, builder parameters, or domain validation without committed on-wire bus captures violates our empirical standard.

Why the Runtime Changes Had to Be Reverted (edf9ea8)

Commit edf9ea8 correctly rolled back the speculative changes after review. Specifically:

1. WHO 22 Area vs. Source Ambiguity (WHAT#4#X):
   The PR originally asserted that the second parameter of *22*1#4#X*WHERE## is strictly an output AREA. However, firmware oracle analysis shows that gateways (bt_multi) copy that nibble verbatim into bus telegrams ($0490018213), making emulator behavior indifferent to semantics. In real-world single-area installations, traces show this field carrying the source ID. Deprecating turn_on_with_source and forcing derived area routing would have broken working single-zone setups.
2. WHO 16 Volume Clamping (0..31 vs. 0..100):
   Restricting set_volume to <= 31 and returning None for reports > 31 violated downstream client contracts. Home Assistant's media_player expects a 0..100% / float scale. Any physical gateway or software component operating on a percentage basis would cause the entity state to collapse to None or throw exceptions on volume changes.
3. Domain & Route Restrictions:
   Enforcing strict 1..9 boundaries and raising ValueError on source targets in route_source added runtime failure modes to existing scripts without evidence of bus errors.
4. Parser Sanitization:
   The initial revision dropped existing isdigit() checks on WHERE/WHAT parameters, weakening parser resilience against malformed frames.

---

3. PR Hygiene & Git Integrity

1. Severe Description/Diff Desynchronization:
   The PR title (fix(sound): WHO 22 area parameter and area-derived routing, modulation-aware tuner decode, station vs preset; WHO 16 volume 0..31, amplifier range, named WHATs...) and the PR body describe 15+ substantive features, new properties, and state diagrams that are completely absent from the actual diff. Merging this PR under its current title and description would severely degrade git history and changelog traceability.
2. Unresolved Merge Conflict:
   Commit b0b5efeb introduced an unmerged conflict marker in CHANGELOG.md. The branch cannot be merged cleanly.

---

4. Breakdown of Claims vs. Reality

Claimed Fix	Does it solve a real problem?	Analysis
WHO 22 Area vs. Source routing	❌ No	Speculative. Conflates theoretical documentation with physical multi-area bus behavior. Reverted.
WHO 16 Volume clamped to 0..31	❌ No	Causes downstream breaking changes in Home Assistant media_player. Reverted.
Station (Dim 6) vs. Preset (Dim 19)	❌ No	Synthetic API churn with no reported integration defect. Reverted.
AM Tuner trailing field handling	❌ No	Changed decoded values without crash evidence; was not a parser panic fix. Reverted.
Bare Power Frame Crash (*22*1*W##)	⚠️ Yes (Real Bug)	Not resolved here. Extracted and properly addressed in PR #163 with dedicated oracle suites.
Retained Docstrings (Current Diff)	❌ Trivial	Clarifies documentation discrepancies, but does not warrant keeping an open, conflicted 10-commit PR.

---

5. Recommendation & Next Steps

1. Close this PR without merging.
2. If the docstring notes detailing the evidence boundaries on volume_up, turn_on_with_source, send_rds_text, and open_paging are desired in the codebase, open a clean, single-commit PR titled:
   docs(sound): add evidence boundary notes to WHO 22 sound diffusion docstrings
3. Direct testing and review efforts to PR #163, which correctly protects gateways from the daemon crash without breaking API contracts or downstream consumers.