fix: auto-call CW gap started before the message had finished

Reported: a 6 s auto-call gap gives clearly less than 6 s between the end of one
CQ and the start of the next.

The wait after sending watched only the keyer's busy signal. That signal travels
from the keyer to the window as an event, and the code gave it one second to
appear before giving up — if it had not arrived and cleared by then, the wait
ended immediately and the gap ran from the START of the call. A 6 s gap after a
4 s CQ becomes about 2 s of silence, which is exactly what was heard.

The message's own duration is now a FLOOR: the wait runs at least the estimated
sending time, and the busy signal only extends it (slow link, long buffer),
capped as before. The estimate can only be approximate, but it cannot be skipped
by a late event, and erring long costs a slightly wider gap rather than a call
that steps on the previous one.

changelog.json

@@ -13,7 +13,8 @@
       "The Recent QSOs counter labels each of its three numbers — shown, matching the filter, and in the log — with thousands separators. \"Showing 10000 of 23683 matches · 28648 total\" read as if the filter had found more QSOs than the log holds.",
       "Switching the CAT backend back to a Yaesu no longer risks leaving the serial port held. Each reconnect opened a new handle without closing the previous one, and Windows opens a serial port exclusively, so the next attempt could fail with \"port busy\".",
       "A Yaesu that is switched off is reported as such. Opening the serial port used to be logged as \"connected\", with the model name left over from the previous session, so a radio that was simply powered down looked like a software fault.",
-      "Yaesu: moving RX and TX together to the SUB receiver is no longer shown as split. That is simplex on the sub VFO, but the rig reports its split flag anyway, so OpsLog now works split out from which VFO transmits and which receives."
+      "Yaesu: moving RX and TX together to the SUB receiver is no longer shown as split. That is simplex on the sub VFO, but the rig reports its split flag anyway, so OpsLog now works split out from which VFO transmits and which receives.",
+      "Auto-call CW now waits the gap you set AFTER the message, not from the moment it starts. With a WinKeyer or serial keyer the wait could end before the call had finished, so a 6 s gap gave about 2 s of silence after a 4 s CQ."
     ],
     "fr": [
       "Cliquer sur un spot du cluster en écoutant sur le VFO SUB ne ramène plus la radio sur le VFO principal. La commande utilisée pour s'accorder agit par définition sur le VFO principal ; sur SUB, c'est désormais le VFO secondaire qui est écrit, sans toucher à la sélection.",
@@ -26,7 +27,8 @@
       "Le compteur des QSO récents nomme chacun de ses trois nombres — affichés, correspondant au filtre, et présents dans le log — avec des séparateurs de milliers. « Showing 10000 of 23683 matches · 28648 total » se lisait comme si le filtre trouvait plus de QSO que le log n'en contient.",
       "Repasser le backend CAT sur un Yaesu ne risque plus de laisser le port série occupé. Chaque reconnexion ouvrait une nouvelle poignée sans fermer la précédente, et Windows ouvre un port série en exclusivité : la tentative suivante pouvait échouer avec « port occupé ».",
       "Un Yaesu éteint est signalé comme tel. L'ouverture du port série était journalisée comme « connecté », avec le nom de modèle hérité de la session précédente : une radio simplement hors tension ressemblait à un défaut du logiciel.",
-      "Yaesu : passer RX et TX ensemble sur le récepteur SUB n'est plus affiché comme un split. C'est du simplex sur le VFO secondaire, mais la radio lève quand même son indicateur de split ; OpsLog le déduit désormais du VFO qui émet et de celui qui reçoit."
+      "Yaesu : passer RX et TX ensemble sur le récepteur SUB n'est plus affiché comme un split. C'est du simplex sur le VFO secondaire, mais la radio lève quand même son indicateur de split ; OpsLog le déduit désormais du VFO qui émet et de celui qui reçoit.",
+      "L'appel automatique en CW attend désormais l'intervalle réglé APRÈS le message, et non depuis son début. Avec un WinKeyer ou un keyer série, l'attente pouvait se terminer avant la fin de l'appel : un intervalle de 6 s ne laissait qu'environ 2 s de silence après un CQ de 4 s."
     ]
   },
   {

frontend/src/App.tsx

@@ -2602,11 +2602,21 @@ export default function App() {
           if (isRig) {
             await waitMs(Math.round(estimateCwMs(resolved, wkWpm)) + 600);
           } else {
-            // WinKeyer: watch the busy echo, capped by the estimate (+slack) since
+            // WinKeyer / serial keyer: the message takes AT LEAST its own length to
-            // over a remote/serial-over-IP link that echo can lag tens of seconds.
+            // send, so the estimate is a floor and the busy signal only extends it.
-            const capMs = Math.round(estimateCwMs(resolved, wkWpm) * 1.4) + 2500;
+            //
-            for (let i = 0; i < 20 && !wkBusyRef.current && !aborted(); i++) await sleep(50); // ≤1s to start
+            // Waiting on busy alone made the auto-call gap start early: busy travels
-            const deadline = Date.now() + capMs;
+            // from the keyer to this window as an event, and if it had not arrived
+            // and cleared within the one-second start window the wait simply ended.
+            // A 6 s gap then ran from the START of the CQ — about 2 s of silence
+            // after a 4 s call, which is what an operator hears as "clearly not 6".
+            const estMs = Math.round(estimateCwMs(resolved, wkWpm));
+            const started = Date.now();
+            const capMs = Math.round(estMs * 1.4) + 2500;
+            const deadline = started + capMs;
+            // The message's own duration, first — this is the part that must not be skipped.
+            while (Date.now() - started < estMs && !aborted()) await sleep(50);
+            // Then let a still-sending keyer finish (slow link, long buffer), capped.
             while (wkBusyRef.current && !aborted() && Date.now() < deadline) await sleep(50);
           }
         }