Thanks for this — really thorough docs, and a useful contributor reference for the audio code. Spot-checked the concrete claims against the source and they hold up:

• Resampler::processStereoToStereo() does average L/R and duplicate (src/core/Resampler.cpp:127).
• PCC codes 0x03E3 / 0x0123 / 0x8005 match src/core/PanadapterStream.h:126-128.
• Default TX chain Gate → EQ → DeEss → Comp → Tube → Enh (PUDU) → Reverb matches defaultChain() in src/core/AudioEngine.cpp:1771.
• Final limiter ceiling default -1.0 dBFS matches src/core/ClientFinalLimiter.h:86.
• Opus 24 kHz / 10 ms / 240 frames matches src/core/OpusCodec.h:48-51.
• m_rxBufferCapMs default 200, clamped 50..1000, matches src/core/AudioEngine.h:114,599.
• RADE TX 24→16 kHz mono → encode → 8→24 kHz stereo and RX 24→8 kHz mono → decode → 16→24 kHz stereo match src/core/RADEEngine.cpp:59-62,142,191,212,276.
• The note that applyClientRxDspFloat32() is currently a dispatcher stub matches the TODO bodies in src/core/AudioEngine.cpp:1688-1701; the live RX strip is indeed inlined in writeAudio().

A few small notes (suggestions, not blockers):

• The PR body and audio-pipeline.md Executive Summary explicitly call out three current edge cases (one-sided stereo (L+R)/2 loss in processStereoToStereo(), left-channel-only PC mic meter, DAX radio-native one-sided loss). These are real and worth tracking — consider opening a follow-up issue so they don't get lost in the docs.
• tx-audio-signal-path.md renames "PooDoo™ Audio" to "AetherialAudio" — please double-check that's the canonical user-facing name in current GUI strings; if the menu/tab still reads "PooDoo" anywhere, the doc rename will read inconsistently.
• Minor: "RADE TX/RX branch points" wording in the architecture-pipelines.md ASCII diagram is slightly cramped — readable, but a future pass could split the RADE/DAX boxes onto separate sub-flows for clarity. Not worth churning this PR.

Docs-only, no risk to runtime behavior, and the path/stage table at the bottom is going to save a lot of code-spelunking on future audio changes. Nice work.

Claude here, reviewing on Jeremy's behalf.

Verdict

Strong merge. This is a genuinely useful piece of reference documentation that the project has been missing — comprehensive enough that future contributors won't have to grep four files to understand where audio averages, where it duplicates, and where it stays stereo. The path/stage table and the downmix/format-change table are the most valuable artifacts.

Disclosure: AI-generated via OpenAI Codex GPT-5.5 Pro, reviewed and tested by @jensenpat. I spot-checked four of the load-bearing technical claims against the actual code:

All four match the doc verbatim. The three "current known behavior" bullets at the top (one-sided stereo 6.02 dB loss, left-channel-only PC mic meter, DAX radio-native L+R averaging) are real bugs documented honestly rather than papered over — exactly the disclosure pattern reference docs should set.

What this adds that didn't exist

• Comprehensive path/stage table — file/function, input format, output format, sample rate, channel handling, side effects, in one place
• Downmix/duplication/resampling/format-change table — every stereo↔mono transition in the audio code annotated with which helper does the work
• Meter tap table — pre/post stage, channel policy, units for every local and radio-side meter
• Mermaid diagrams for each path — RX speaker, PC mic voice TX, RADE TX/RX, DAX TX routing — that any contributor can read without opening a debugger

The diffs to architecture-pipelines.md and tx-audio-signal-path.md correctly link out to the new doc and trim down stale summaries. The PooDoo™ → AetherialAudio rename in tx-audio-signal-path.md matches the current code (the AetherialAudioStrip class, the v0.9.6 unification, and the v26.5.1 CHANGELOG framing). Good consistency pass.

Follow-up worth filing as separate issues

The doc surfaces three current bugs as "known behavior." Each deserves its own GitHub issue so they get tracked rather than living only in the doc:

1. Stereo mic 6.02 dB loss in Resampler::processStereoToStereo() — one-sided stereo input loses 6.02 dB when averaged with a silent channel. Affects PC mic capture path at non-24kHz rates.
2. PC mic voice meter left-channel-only read — AudioEngine.cpp:3656 i += 2 stride. A right-only mic device shows silent or incorrect levels.
3. DAX radio-native TX 6.02 dB loss — AudioEngine.cpp:4152 averages stereo to mono with the same one-sided-stereo penalty.

These are independent fixes; the doc doesn't need to wait on them.

Minor

• The doc is 618 lines of new content. The auxiliary tables are the most valuable bit — recommend they stay current as the audio code evolves. Worth a CI hook eventually that flags PRs touching src/core/AudioEngine.cpp or src/core/Resampler.cpp for a "did you update docs/audio-pipeline.md?" reviewer prompt. Not in this PR's scope, just a thought.
• The TCI / VirtualAudioBridge / PipeWireAudioBridge sections look accurate by class-name spot-check, but I didn't trace every code path inside them. If something there is subtly wrong, the table format makes it easy to fix in a follow-up.

CI

• build SUCCESS
• check-paths SUCCESS
• check-windows SKIPPED (correct — docs-only)
• No analyze (cpp) since no source files changed
• CodeQL skipped (correct per the doc-only PR optimization landed in #2397)

Recommendation

Merge. Then file follow-up issues for the three documented bugs so they're in the tracker as actionable work, not just doc footnotes.

Thanks @jensenpat — the diagnostic discipline applied to documentation (tracing every L/R collapse, every resample boundary, every meter tap) is the same care you bring to fix PRs. The contributor notes at the end ("Treat Resampler::processStereoToStereo() as a mono-collapse helper") are the kind of "future-self warning" that prevents repeat mistakes.

73, Jeremy KK7GWY & Claude (AI dev partner)

Approved — docs accurately reflect code, three documented bugs are real and now tracked separately.