diff --git a/include/virtual_sx1262.h b/include/virtual_sx1262.h
index d412f97..5edf571 100644
--- a/include/virtual_sx1262.h
+++ b/include/virtual_sx1262.h
@@ -214,7 +214,8 @@ typedef struct {
   /* Events the chip had and the firmware never saw, because IrqMask had them
    * masked off. A firmware waiting on an interrupt it did not enable looks
    * exactly like a quiet mesh, and this is the only thing that tells the two
-   * apart. */
+   * apart. One per event, however many of its bits were masked: a header is one
+   * event whether one interrupt or three would have announced it. */
   uint32_t irq_suppressed;
   /* Scans run, and scans that found a carrier. A host comparing a firmware that
    * scans against one that does not needs to know the scans happened and what
diff --git a/src/VirtualSX1262.cpp b/src/VirtualSX1262.cpp
index e9bd58d..fd4d5d2 100644
--- a/src/VirtualSX1262.cpp
+++ b/src/VirtualSX1262.cpp
@@ -149,9 +149,14 @@ void VirtualSX1262::tick(uint64_t nowMs) {
     // The receiver's own preamble length stands in for the transmitter's: a
     // mesh agrees on one, and the chip is told nothing about who is sending.
     // An implicit-header receiver has no header to find.
+    //
+    // HeaderValid alone: SyncWordValid is the GFSK modem's interrupt, and a LoRa
+    // receiver reports its header through HeaderValid (DS_SX1261-2 table 13-29).
+    // Raising both made every header count as suppressed on a firmware that
+    // correctly leaves the FSK bit out of its mask.
     const double headerEnd = preambleSyms_ + kSyncAndSfdSymbols + kHeaderBlockSymbols;
     if (!headerImplicit_ && !headerRaised_ && sinceMs >= headerEnd * symMs) {
-      raiseIrq(kIrqHeaderValid | kIrqSyncWordValid);
+      raiseIrq(kIrqHeaderValid);
       headerRaised_ = true;
     }
   }
@@ -203,7 +208,7 @@ void VirtualSX1262::setChannelBusy(bool busy) {
     preambleRaised_ = false;
     headerRaised_ = false;
     if (stuckIrqMs_ == 0) {
-      irq_ &= (uint16_t) ~(kIrqPreambleDetected | kIrqHeaderValid | kIrqSyncWordValid);
+      irq_ &= (uint16_t) ~(kIrqPreambleDetected | kIrqHeaderValid);
     }
   }
   channelBusy_ = busy;
diff --git a/src/registers.h b/src/registers.h
index 73e0210..285ce6e 100644
--- a/src/registers.h
+++ b/src/registers.h
@@ -164,9 +164,10 @@ namespace {
 // The header cannot be read until the whole preamble, the two sync symbols and
 // the 2.25-symbol SFD have gone by, and it is the first interleaver block after
 // them: eight symbols, always at CR 4/8. A fixed twelve symbols raised
-// HeaderValid 32 symbols early for MeshCore's 32-symbol preamble at SF8 and
-// below (16 at SF9 and up), which shrank the stretch where a firmware that can see
-// preambles answers differently from one that cannot to a third of its real width.
+// HeaderValid 32.25 symbols early for MeshCore's 32-symbol preamble at SF8 and
+// below (16 at SF9 and up). That shrank the stretch where a firmware that can
+// see preambles answers differently from one that cannot, from PreambleDetected
+// to HeaderValid, to eight symbols of its real 40.25: a fifth.
 [[maybe_unused]] constexpr double kPreambleSymbols = 4.0;
 [[maybe_unused]] constexpr double kSyncAndSfdSymbols = 4.25;
 [[maybe_unused]] constexpr double kHeaderBlockSymbols = 8.0;
diff --git a/test/test_interrupts.cpp b/test/test_interrupts.cpp
index 4e97abe..9f18b27 100644
--- a/test/test_interrupts.cpp
+++ b/test/test_interrupts.cpp
@@ -7,6 +7,34 @@
  */
 #include "harness.h"
 
+/* Two frames heard under one IrqMask, the way a receive loop takes them: a
+ * carrier long enough to carry its header, the frame, the carrier gone, and the
+ * status cleared before the next. Answers what the chip counted as suppressed. */
+static uint32_t suppressed_over_two_frames(uint16_t mask) {
+  vsx_chip* c = vsx_create();
+  uint64_t now = 0;
+  bring_up(c, &now);
+  /* SetDioIrqParams: the mask under test, and RxDone alone routed to DIO1. */
+  const uint8_t dio[] = {0x08, (uint8_t)(mask >> 8), (uint8_t)mask, 0, 0x02, 0, 0, 0, 0};
+  vsx_spi_transaction(c, dio, nullptr, sizeof(dio));
+  const uint8_t frame[] = {1, 2, 3, 4};
+  for (int n = 0; n < 2; ++n) {
+    vsx_set_channel_busy(c, 1);
+    for (int i = 0; i < 150; ++i) {
+      vsx_tick(c, ++now);
+    }
+    vsx_deliver_frame(c, frame, sizeof(frame));
+    vsx_set_channel_busy(c, 0);
+    vsx_tick(c, ++now);
+    const uint8_t clear[] = {0x02, 0xFF, 0xFF};
+    vsx_spi_transaction(c, clear, nullptr, sizeof(clear));
+  }
+  vsx_counters k;
+  vsx_get_counters(c, &k);
+  vsx_destroy(c);
+  return k.irq_suppressed;
+}
+
 int main() {
   std::printf("virtual-sx1262");
 
@@ -293,5 +321,19 @@ int main() {
     vsx_destroy(c);
   }
 
+  /* ---------------------------------------------------------------- */
+  CASE("a LoRa header is not counted against a mask that leaves FSK out");
+  /* The header was raised as HeaderValid and SyncWordValid together, and
+   * SyncWordValid is the GFSK modem's. No LoRa firmware enables it, so every
+   * header counted as suppressed and a correctly configured firmware looked as
+   * though it were still missing interrupts. The two masks are MeshCore's own:
+   * 1.17 enables PreambleDetected for receive and 1.16 does not. */
+  {
+    check(suppressed_over_two_frames(0x0276) == 0,
+          "1.17's mask misses nothing on a frame it receives");
+    check(suppressed_over_two_frames(0x0272) == 2,
+          "1.16's misses PreambleDetected, once per frame");
+  }
+
   return report("");
 }
diff --git a/test/test_reception.cpp b/test/test_reception.cpp
index d125a7d..ef9e1b9 100644
--- a/test/test_reception.cpp
+++ b/test/test_reception.cpp
@@ -31,6 +31,8 @@ int main() {
     }
     check((irq_flags(c) & IRQ_HEADER) != 0,
           "while the carrier is on the air, HeaderValid is set");
+    check((irq_flags(c) & IRQ_SYNC_WORD) == 0,
+          "and SyncWordValid is not: that is the FSK modem's, and this is LoRa");
 
     const uint8_t frame[] = {0x11, 0x00, 0x33, 0x0a, 0xc2, 0x1f};
     vsx_deliver_frame(c, frame, sizeof(frame));
@@ -41,7 +43,6 @@ int main() {
     check((f & IRQ_RX_DONE) != 0, "the delivered packet raises RxDone");
     check((f & IRQ_HEADER) == 0, "HeaderValid is clear once the carrier has gone");
     check((f & IRQ_PREAMBLE) == 0, "PreambleDetected is clear too");
-    check((f & IRQ_SYNC_WORD) == 0, "SyncWordValid is clear too");
     vsx_destroy(c);
   }
 
@@ -220,7 +221,7 @@ int main() {
   /* ---------------------------------------------------------------- */
   CASE("HeaderValid waits for the preamble, the SFD and the header block");
   /* It was raised twelve symbols into the carrier whatever the preamble, so for
-   * MeshCore's sixteen-symbol preamble it came sixteen symbols early. A
+   * the chip's default sixteen-symbol preamble it came 16.25 symbols early. A
    * firmware that sees preambles and one that does not answer differently only
    * between PreambleDetected and HeaderValid, and that stretch was a third of
    * its real width. At SF8 and 62.5 kHz a symbol is 4.096 ms, so the header ends