feat(anan): publish link telemetry and withdraw the PA readout. Principle VIII.

docs/architecture/radio-capabilities-map.md

@@ -204,6 +204,7 @@ the production dialog and checks its displayed peak level and frequency.
 | `hasSupplyVoltageTelemetry` | ✅ | ❌ | ❌ | `MainWindow::applyCapabilitiesToUi` | PA supply-voltage readout in the status bar. Icom: ✅ only when the active model profile provides a calibrated Vd curve; unprofiled models neither publish nor poll Vd/Id |
 | `hasPaTemperatureTelemetry` | ✅ | ✅ | ❌ | `MainWindow::applyCapabilitiesToUi` | PA-temperature gauge and °C/°F selector in Radio Vitals. Icom: ❌ until a model profile declares and implements a PA-temperature meter; the IC-9700 has no such declared telemetry |
 | `hasPaCurrentTelemetry` | ❌ | ❌ | ❌ | `MainWindow::applyCapabilitiesToUi` | Calibrated PA drain-current face in Radio Vitals, used only when PA-temperature telemetry is unavailable. Icom: ✅ only for the IC-9700 profile's documented 0–20 A Id calibration. Flex remains ❌ because its PACURRENT meter is known to clip below real full-power draw |
+| `paTelemetryAudit` (record) | — (absent) | — (absent) | — (absent) | `MainWindow::applyCapabilitiesToUi` | A PROTOCOL audit of the PA telemetry the wire format can carry at all, as opposed to what a session has so far RECEIVED. ENGAGED with `temperatureAbsent` means the payload was read field by field and there is no temperature in it, so the status bar WITHDRAWS its PA temperature row instead of leaving it on `PA --°C` forever. ABSENT (the default) means NOT AUDITED and every consumer keeps its historical presentation — which is what holds Flex, HL2, Icom, Sim and RTL at today's behaviour rather than reclassifying a default `false` above as "proven absent". ANAN: engaged, because Protocol 2's whole analog payload is six fields and none is a temperature. A record and not a bool per #5262 M2, whose boolean count is frozen and shrink-only; the audit can also grow a current/supply field without a second ratchet argument |
 | `hasMainFanTelemetry` | ✅ | ❌ | ❌ | `MainWindow::applyCapabilitiesToUi` | Main Fan gauge in Radio Vitals. All current Icom models are ❌ because the backend does not publish fan-speed telemetry |
 | `hostFrequencyCalibration` | ❌ | ✅ | ❌ | `RadioSetupDialog` (Calibration page), `AutomationServer::doFreqCal` | Shows the Calibration page and enables the `freqcal` bridge verb. Means "**the client** owns the frequency-error correction", not "this radio has an error" — every radio does. Flex is ❌ because it calibrates itself (`radio set cal_freq` / `pll_start`), and that surface stays in the Frequency Offset group on the Receive page. HL2 is ✅ because its 76.8 MHz NCO scale is a `localparam` in the bitstream (`radio.v` M2) and no register in the HPSDR map accepts a correction — see `docs/architecture/hl2-frequency-calibration.md` |
 | `hostDroopCalibration` | ❌ | ❌ | ❌ | `RadioSetupDialog` (Droop Correction page), `AutomationServer::doDroopCal` | Shows the Droop Correction page and enables the `droopcal` bridge verb. Means "the client has measured and can correct a real DDC edge droop on this radio", not "this radio has no droop" — HL2's own DDC decimation chain looks architecturally similar and has never been characterised, so its ❌ is "not yet measured", not "known absent". ANAN: ✅ — the Saturn FPGA's DDC0 CIC/decimation chain has a real, bench-measured sin(x)/x droop near the edges of the displayed span (`AnanDroopCorrection.h`), corrected in-app via `AnanDroopCalibrator` |

src/core/backends/RadioCapabilities.h

@@ -544,6 +544,28 @@ struct RadioCapabilities {
     // because some radios define PACURRENT with an unusable/clipped range.
     bool hasPaCurrentTelemetry = false;
 
+    // A PROTOCOL audit of the PA telemetry the radio's wire format can carry at
+    // all, as distinct from the three flags above, which report what a session
+    // has so far RECEIVED. Engaged means the payload was read field by field and
+    // the absence is a property of the protocol rather than of a radio that has
+    // simply not sent one yet -- so a readout may be WITHDRAWN instead of left
+    // printing a placeholder that can never resolve.
+    //
+    // std::nullopt (the default) means NOT AUDITED, and every consumer keeps its
+    // historical presentation. That is deliberate: it holds a family that has
+    // made no such claim at its existing behaviour instead of quietly
+    // reclassifying a default `false` above as "proven absent" -- the
+    // all-defaults-false trap check_capability_records.py was written to stop.
+    //
+    // A record rather than another bool because RadioCapabilities is AT its
+    // frozen boolean count and shrink-only (#5262 M2); per-feature records are
+    // the sanctioned shape for a new capability.
+    struct PaTelemetryAudit {
+        // The protocol defines no PA temperature field anywhere in its payload.
+        bool temperatureAbsent = false;
+    };
+    std::optional<PaTelemetryAudit> paTelemetryAudit;
+
     // The radio reports main-fan speed as live telemetry. False means the
     // Radio Vitals applet omits the fan gauge instead of presenting an
     // instrument that can never receive a sample. This is independent of PA

src/core/backends/anan/AnanBackend.cpp

@@ -254,12 +254,12 @@ AnanBackend::AnanBackend(QObject* parent)
     });
     connect(m_client, &P2Client::connectionError, this,
             [this](const QString& reason) { emit connectionError(reason); });
-    connect(m_client, &P2Client::dropsUpdated, this, [](quint64) {
-        // No LinkStats wiring in this phase -- see the design plan's
-        // "explicitly not in this commit" list. Connected so the signal has
-        // a receiver rather than going nowhere; a future commit can surface
-        // it through IRadioBackend::linkStats().
-    });
+    // Drops reach the readout through the counters snapshot below, which
+    // carries the same total on a fixed cadence. Not wired per-drop: the
+    // LinkStats tick is deliberately periodic rather than event-driven, so a
+    // link that has gone silent still reports.
+    connect(m_client, &P2Client::linkCountersUpdated, this,
+            &AnanBackend::onLinkCounters);
     // Only DDC0 feeds this DSP. Like ddc0IqReady above, notification and
     // processing run directly on the I/O thread, with the DSP as context.
     // Invalidate its partial FFT before the discontinuous block arrives.
@@ -405,6 +405,20 @@ RadioCapabilities AnanBackend::capabilities() const
     c.hasAgcThreshold = true; // Host receiver DSP implements threshold/off gain.
     c.manufacturer = QStringLiteral("Apache Labs");
     c.model = QStringLiteral("ANAN-G2");
+    // ---- PA telemetry ----
+    // hasSupplyVoltageTelemetry stays FALSE although the radio sends a
+    // supply-rail count (P2Protocol decodes it): no known scale fits it. On a
+    // G2 bench the 5 V-reference scale read a 13.8 V rail as ~40 V, 2.9x high,
+    // and the 3.3 V board's scale differs from it by only 1.5x, so no board
+    // choice fixes it. A volts readout waits for an operator calibration.
+    //
+    // There is NO PA temperature in this protocol. The radio's whole analog
+    // payload is six fields and none of them is a temperature, so the readout
+    // can never resolve and is withdrawn rather than left showing a placeholder.
+    // hasPaTemperatureTelemetry stays false on its own (nothing is received);
+    // this says the stronger thing -- that nothing CAN be.
+    c.paTelemetryAudit = RadioCapabilities::PaTelemetryAudit{
+        /*temperatureAbsent=*/true};
     c.canCreateSlices = false;
     c.maxSlices = 1;
     c.maxPanadapters = 1;
@@ -730,6 +744,12 @@ void AnanBackend::disconnectRadio()
     m_tuneThrottleTimer->stop();
     m_tunePendingApply = false;
     m_pendingBandwidthKsps = 0;
+    // The next session's counters restart from zero in P2Client::start(), so a
+    // cached snapshot from this one would make the first tick of the next read
+    // as a huge backward jump in every total.
+    m_linkCounters = {};
+    m_linkCountersSeen = false;
+    m_lastSnapshotRxBytes = 0;
     // A genuine operator-initiated disconnect always fires disconnected(),
     // even one that lands mid-rate-change -- this is not the zoom case
     // m_rateChanging exists to hide.
@@ -1405,6 +1425,43 @@ void AnanBackend::defineMeters()
     emit meterDefined(d);
 }
 
+IRadioBackend::LinkStats AnanBackend::linkStats() const
+{
+    LinkStats s;
+    // Nothing measured yet: the consumer keeps whatever source it already had
+    // rather than being handed a snapshot of zeros.
+    if (!m_connected || !m_linkCountersSeen)
+        return s;
+    s.reported = true;
+    s.alive = m_linkCounters.rxBytes > m_lastSnapshotRxBytes;
+    s.rxBytes = static_cast<qint64>(m_linkCounters.rxBytes);
+    s.txBytes = static_cast<qint64>(m_linkCounters.txBytes);
+    s.rxPackets = m_linkCounters.rxPackets;
+    s.rxPacketsLost = m_linkCounters.drops;
+    s.localEndpoint = m_linkCounters.localEndpoint;
+    // LEFT AT -1, WHICH THE STRUCT DEFINES AS "NOT MEASURED" AND THE MODEL
+    // HONOURS BY IGNORING. This transport is stream-only: the radio is told
+    // where to send and then sends, with no request/response exchange to time,
+    // so there is no round trip here to measure. Reporting 0 instead would
+    // render as a perfect "< 1 ms" link -- a measurement never taken.
+    // rttMs, jitterMs, gapMs and gapMaxMs all keep their -1 defaults.
+    return s;
+}
+
+void AnanBackend::onLinkCounters(const P2Client::LinkCounters& counters)
+{
+    // A session restart shows as a BACKWARD jump, since P2Client::start() zeroes
+    // its counters; the cache is then a dead session's and is discarded here,
+    // whichever path restarted it. The first snapshot of a session compares
+    // against ZERO: against itself, it would read dead on the first tick.
+    if (counters.rxBytes < m_linkCounters.rxBytes)
+        m_linkCountersSeen = false;
+    m_lastSnapshotRxBytes = m_linkCountersSeen ? m_linkCounters.rxBytes : 0;
+    m_linkCounters = counters;
+    m_linkCountersSeen = true;
+    emit linkStatsUpdated(linkStats());
+}
+
 void AnanBackend::onDspMeter(float dbfs)
 {
     const double dbm = static_cast<double>(dbfs) + kSMeterDbmOffset + m_attenuationDb;

src/core/backends/anan/AnanBackend.h

@@ -92,6 +92,11 @@ class AnanBackend : public IRadioBackend {
     void setKeying(bool key, const AetherSDR::TxCoordinator::Operation& operation, const AetherSDR::TxCoordinator::Completion& completion = {}) override;
     void invokeExtension(const QString& ns, const QString& verb,
                          quint64 requestId, const QVariant& arg = {}) override;
+    // Transport snapshot for the status bar's Network field and Network
+    // Diagnostics. Synchronous, and reads only the cache onLinkCounters() keeps
+    // on this thread -- never the client's own counters, which belong to the
+    // I/O thread. The push half is onLinkCounters(), on the client's tick.
+    [[nodiscard]] LinkStats linkStats() const override;
 
     // ---- pure-function pieces, exposed static for testability ----
     // (matches the WdspChannel/Hl2RxDsp precedent of exposing normally-
@@ -149,6 +154,7 @@ class AnanBackend : public IRadioBackend {
 
 private:
     friend class AnanNoiseBlankerTestAccess;
+    friend class AnanLinkStatsTestAccess;
     void beginDspSetup();
     void finishDspSetup(quint64 generation, bool ok, const QString& error);
     // The "restart P2Client with m_pendingParams, then retune" half of what
@@ -183,6 +189,21 @@ class AnanBackend : public IRadioBackend {
     // Declares SLC:LEVEL to the meter seam; on every connect, before the
     // first reading can arrive. See its definition.
     void defineMeters();
+    // One client snapshot -> one LinkStats push. The FIXED cadence is the
+    // client's timer, not ours: this runs on whatever the client published, and
+    // it must keep arriving after the radio goes quiet, because "nothing came
+    // this second" is the observation the heartbeat's alarm path waits for
+    // (IRadioBackend::LinkStats).
+    void onLinkCounters(const P2Client::LinkCounters& counters);
+    // The last snapshot the client published, cached on THIS thread so the
+    // synchronous linkStats() getter never reads the I/O thread's counters --
+    // see P2Client::LinkCounters for why that read would be a race.
+    P2Client::LinkCounters m_linkCounters;
+    bool m_linkCountersSeen = false;
+    // rxBytes at the previous snapshot, for LinkStats::alive -- a link that is
+    // bound and counting but has gone quiet must read as dead, which a
+    // cumulative total alone cannot say.
+    quint64 m_lastSnapshotRxBytes = 0;
     // One WDSP S-meter reading (dBFS) -> dBm, smoothed, published on a tick.
     void onDspMeter(float dbfs);
     // The same smoother Hl2Backend publishes through, so the two receivers'