Nice contribution, @nigelfenton — well-structured PR with clean separation from the existing AG code path. The approach of reusing AntennaGeniusModel for protocol compatibility is solid. A few things to address before merge:

Dangling reference capture in RadioSetupDialog.cpp

In buildPeripheralsTab(), the web UI button handler captures settings by reference:

auto& settings = AppSettings::instance();
// ...
connect(webBtn, &QPushButton::clicked, this, [this, &settings]() {
    QString ip = settings.value("SS_ManualIp", "").toString();
    // ...
});

settings is a local reference on the stack of buildPeripheralsTab(). Capturing it by reference in a lambda that outlives the function (button click handler) is undefined behavior — the lambda holds a dangling reference after the function returns.

Fix: drop the &settings capture and call AppSettings::instance() directly inside the lambda, or capture this only (which you're already doing):

connect(webBtn, &QPushButton::clicked, this, [this]() {
    auto& settings = AppSettings::instance();
    // ...
});

Duplicated ShackSwitch detection logic

The check serial.startsWith("G0JKN") || name.contains("ShackSwitch", Qt::CaseInsensitive) appears in at least 5 locations across 3 files:

• AntennaGeniusApplet.cpp
• MainWindow.cpp (lambda + two connect handlers)
• RadioSetupDialog.cpp (isRealAg, isSsConnected, webBtn handler — 3 times)

The PR description mentions an "isShackSwitch flag on AntennaGeniusModel" but I don't see such a method on the model. Consider adding a static helper or member function on AntennaGeniusModel:

static bool isShackSwitch(const AgDeviceInfo& info) {
    return info.serial.startsWith("G0JKN") ||
           info.name.contains("ShackSwitch", Qt::CaseInsensitive);
}

This avoids the risk of the detection logic drifting between call sites if e.g. a new hardware revision uses a different serial prefix.

Misleading comment on openShackSwitchWebUi

The comment says "Tries Chromium-based browsers with --app flag; falls back to default browser" but the implementation is just QDesktopServices::openUrl(). Either update the comment to match the implementation, or remove the wrapper entirely and call QDesktopServices::openUrl() directly at the call site — the function adds no value as-is.

7-second QTimer::singleShot delay for AG manual connect

In MainWindow.cpp onConnectionStateChanged:

QTimer::singleShot(7000, this, [this, agIp, agPort]() {
    if (!m_antennaGenius.isConnected())
        m_antennaGenius.connectToAddress(QHostAddress(agIp), agPort);
});

A hardcoded 7s delay is fragile — if the ShackSwitch TCP handshake takes longer than expected, or if onConnectionStateChanged fires again during that window (reconnect scenario), this could race. The isConnected() guard helps, but consider:

• Is 7s always long enough for ShackSwitch to complete its connect + enrichment?
• If the user disconnects and reconnects the radio within 7s, the timer from the first connection still fires with stale agIp/agPort values.

A safer pattern would be to cancel the pending timer on disconnect (store the timer ID and call killTimer or use a QTimer member).

Minor: webPort field on AgDeviceInfo

The new webPort field defaults to 0 and is parsed from the UDP beacon's webport key. The beacon validation (info.serial.isEmpty() || info.ip.isNull()) doesn't check webPort, which is correct since it's optional. Just noting that the port validation in the click handler (> 1024 threshold, falling back to SS_WebPort setting, then to 5000) is well thought out.

Everything else looks good — the AppletPanel integration follows the existing AG pattern, the CMakeLists change is minimal, and the model enrichment logic for late UDP beacons is a nice touch. The AgDeviceInfo struct additions (webPort, lastRadioBand getter) are clean and backward-compatible. Thanks for the thorough testing notes and rollback section in the PR description.