Thanks @jensenpat for tackling this — the DAX ownership filtering and contiguous TCI receiver mapping are exactly the right approach for firmware 4.2.18 compatibility, and the backwards-compat fallbacks are well thought through. The manual WSJT-X validation across both DAX and TCI paths is appreciated.

A few things worth addressing before merge:

1. Duplicated helper functions across three files

parseStatusHandle() and streamStatusBelongsToUs() are copy-pasted identically into both TciServer.cpp and MainWindow.cpp. A third near-duplicate parseStreamToken() lives in RadioModel.cpp. Similarly, tciTrxForSlice() is duplicated between TciProtocol.cpp (taking a QList<SliceModel*>) and TciServer.cpp (taking a RadioModel*).

These should live in one place — either a small shared header or on RadioModel itself. Duplicated parsing logic is a maintenance risk if the handle format evolves with future firmware.

2. Duplicate log lines at registration

In TciServer.cpp around the DAX RX stream registration, the new code adds a qCDebug(lcCat) log immediately followed by the existing qCInfo(lcCat) that logs the same event:

qCDebug(lcCat) << "TCI: registered DAX RX stream"
               << "0x" + QString::number(streamId, 16)
               << "for channel" << ch;
qCInfo(lcCat) << "TCI: registered DAX RX stream" << Qt::hex << streamId
              << "for channel" << ch << "(#1331)";

One of these should be removed — the debug line is sufficient for diagnostics and the info line already existed.

3. const_cast in tciTrxForSlice

Both copies of tciTrxForSlice use const_cast<SliceModel*>(slice) to call QList::indexOf. Since slices() returns QList<SliceModel*> (non-const pointers), and the parameter is const SliceModel*, the cast is technically needed — but it would be cleaner to accept SliceModel* directly (both callers already have non-const pointers) or use std::find to avoid the cast.

4. m_rxAudioPackets incremented but timer never started

TciServer.h adds m_rxAudioLogTimer, m_rxAudioPackets, and m_rxAudioFramesSent, and onDaxAudioReady() increments m_rxAudioPackets. However, I don't see where m_rxAudioLogTimer is started or where these counters are ever read/logged. If the periodic summary log was intended, it looks like it didn't make it into the diff. If it's scaffolding for future work, consider removing the unused members until they're wired up — dead state is confusing.

5. MainWindow.cpp changes are borderline out-of-scope

The MainWindow.cpp diff adds parseStatusHandle/streamStatusBelongsToUs helpers plus rewires sliceAdded → wireSlice to use contiguous TCI indexes. The TCI rewiring makes sense here, but the status-handle parsing helpers duplicate TciServer.cpp. If MainWindow doesn't use them yet (they're only defined, not called from any visible diff context), they could be deferred until actually needed.

6. Minor: m_daxStreamDebug map grows without bound

RadioModel.h adds QMap<quint32, DaxStreamDebugState> m_daxStreamDebug for tracing. The streamStatusRemoved() helper exists but I don't see cleanup code that erases entries from this map when streams are removed. Over a long session with many stream create/destroy cycles, this could grow. Consider erasing entries on removal.

Overall the core logic — ownership filtering, contiguous receiver mapping, per-client audio_start:<receiver> routing — is solid and well-motivated. The compatibility fallbacks are careful. The main ask is to consolidate the duplicated helpers and clean up the unused timer/counter scaffolding. Nice work on a tricky firmware compat area.