diff --git a/.changeset/settle-fast-path.md b/.changeset/settle-fast-path.md
new file mode 100644
index 000000000..80e372346
--- /dev/null
+++ b/.changeset/settle-fast-path.md
@@ -0,0 +1,5 @@
+---
+"@solidjs/signals": patch
+---
+
+The seam (`settle`) no longer pays the effect-queue merge on a plain flush — nothing parked, no lane reveal, no landing: this flush's runs are the queue as they stand instead of a `concat` copy, a parked flush stashes its runs in place, and the per-flush list resets run only when a list is non-empty. About 100 ns less fixed cost per flush (−11% on a 20k-key store commit, −45% on a one-signal flush), with the effect order of every other seam unchanged.
diff --git a/packages/signals/src/core/scheduler.ts b/packages/signals/src/core/scheduler.ts
index 5a92ab804..215447e49 100644
--- a/packages/signals/src/core/scheduler.ts
+++ b/packages/signals/src/core/scheduler.ts
@@ -786,8 +786,7 @@ export class GlobalQueue implements IQueue {
         )
           recompute(zombies[i]);
       // The unchanged passes' tails stay linked (A30, #3469): their inputs
-      // are held, and the committed frame still derives from them.
-      heldTrims.length = 0;
+      // are held, and the committed frame still derives from them — no trim.
     } else {
       commitPendingNodes();
       // The flush committed: an unchanged pass's stale tail goes now. Not
@@ -799,8 +798,12 @@ export class GlobalQueue implements IQueue {
         const t = heldTrims[i];
         if (t._x?._error == null && !(t as any)._modified) trimStaleDeps(t);
       }
-      heldTrims.length = 0;
     }
+    // The list is the flush's either way. Reset only when there is something
+    // to reset (`stagedReaders` below likewise): `length = 0` is a runtime
+    // call even on an empty array, and the plain flush — no unchanged pass,
+    // no lane work — has nothing in either.
+    if (heldTrims.length) heldTrims.length = 0;
     // Lanes: a blocked one parks its frame (its own flight is up); an
     // unblocked one reveals this round's work. Then land every transaction
     // no frame is waiting on — its lanes end with it. Backwards: a landing
@@ -813,7 +816,7 @@ export class GlobalQueue implements IQueue {
     // writes if it parked: the lane seam re-derives them on the committed
     // world (a lane's runs wait this round).
     GlobalQueue._laneSeams?.(t !== null ? stagedReaders : null);
-    stagedReaders.length = 0;
+    if (stagedReaders.length) stagedReaders.length = 0;
     this._queues = [[], []];
     for (let k = transactions.length - 1; k >= 0; k--) {
       const u = transactions[k];
@@ -848,8 +851,15 @@ export class GlobalQueue implements IQueue {
       holding = true;
     }
     for (let i = 0; i < 2; i++) {
-      if (parked) u!._queues[i] = u!._queues[i].concat(own[i]);
-      this._queues[i] = lanes[i].concat(parked ? [] : own[i], this._queues[i]);
+      if (parked) append(u!._queues[i], own[i]);
+      // The plain flush — no reveal, no landing, not parked — has nothing to
+      // order around its own runs: they are the queue as they stand (the
+      // same array, not a copy; `run` takes it whole). Every other seam
+      // builds the ordered queue above.
+      this._queues[i] =
+        parked || lanes[i].length || this._queues[i].length
+          ? lanes[i].concat(parked ? [] : own[i], this._queues[i])
+          : own[i];
     }
     if (deferredZombies.length !== 0) {
       // Survivors are the displayed frame of a held pass: they rerun for the
@@ -864,8 +874,8 @@ export class GlobalQueue implements IQueue {
     GlobalQueue._boundarySeam?.();
   }
   run(type: number) {
-    if (this._queues[type - 1].length) {
-      const effects = this._queues[type - 1];
+    const effects = this._queues[type - 1];
+    if (effects.length) {
       this._queues[type - 1] = [];
       runQueue(effects, type);
     }
diff --git a/scripts/size/floor-caps.json b/scripts/size/floor-caps.json
index 4fab2f1a5..9fe0f5ba7 100644
--- a/scripts/size/floor-caps.json
+++ b/scripts/size/floor-caps.json
@@ -1,7 +1,7 @@
 {
   "signals: core floor (createSignal/Memo/Effect/Root/flush)": "7.33 KB",
-  "app: render + one signal (the simple-app floor)": "9.81 KB",
-  "app: hydrating (no stores) with Show/For/Loading/Errored/lazy": "17.64 KB",
+  "app: render + one signal (the simple-app floor)": "9.83 KB",
+  "app: hydrating (no stores) with Show/For/Loading/Errored/lazy": "17.66 KB",
   "page: base server components (hydrating + dynamic + frames + sf reference)": "44.05 KB",
   "page: live server components (base + live/GET + action + isPending/latest)": "47.67 KB",
   "server: floor (getRequestEvent + isServer)": "1.34 KB",
diff --git a/scripts/size/scenarios.js b/scripts/size/scenarios.js
index a3c12f639..c9aae16b5 100644
--- a/scripts/size/scenarios.js
+++ b/scripts/size/scenarios.js
@@ -1266,6 +1266,19 @@ module.exports = [
     // -3 B minified). The same parking-gate change as the `+ isPending/latest`
     // note; brotli layout turns the -3 B minified into a saving here. Not
     // lowered (frozen caps only fall on a deliberate ratchet).
+    // Size-Exception (settle fast path, 2026-10-04): 9.81 -> 9.83 KB, measured
+    // at 9,812 B against `next` @ 924d909d9's 9,801 (+11 B; 2 B over the cap;
+    // +5 B minified). The seam's effect-queue merge is skipped on the plain
+    // flush — nothing parked, no lane reveal, no landing — and the per-flush
+    // list resets run only on a non-empty list: ~100 ns less fixed cost per
+    // flush (one signal + one render effect + flush: 240 -> 133 ns; a 20k-key
+    // store commit: 844 -> 748 ns). Brotli layout: the same +5 B minified
+    // measures -16 B on `+ createStore`, 0 B on the CSR app, +28 B on
+    // `+ every store primitive family` and +17/+21 B on the two pages (both
+    // within their caps); 44 equivalent encodings measured, none under every
+    // cap at once, this one the smallest minified. Accepted by the
+    // maintainer (2026-10-04). The cap is frozen again at 9.83 KB
+    // (measured + 10 B).
     limit: floorCaps["app: render + one signal (the simple-app floor)"],
     alias
   },
@@ -1573,6 +1586,12 @@ module.exports = [
     // above is gone (no heap refusal, no `_manualWriteTime`, no
     // `deleteFromHeap` in the setter). Reverts the #3740 raise. The cap is
     // frozen again at 19.69 KB.
+    // Size-Exception (settle fast path, 2026-10-04): 17.64 -> 17.66 KB,
+    // measured at 17,650 B against `next` @ 924d909d9's 17,632 (+18 B; 10 B
+    // over the cap; +4 B minified). The same seam change as the simple-app
+    // floor note; brotli layout over the larger bundle. Accepted by the
+    // maintainer (2026-10-04). The cap is frozen again at 17.66 KB
+    // (measured + 10 B).
     limit: floorCaps["app: hydrating (no stores) with Show/For/Loading/Errored/lazy"],
     alias
   },
@@ -1975,7 +1994,14 @@ module.exports = [
     // Lowered (#3774, 2026-10-04): 30.82 -> 28.78 KB, measured at 28.77 KB by CI
     // at 22c3d3e14 — the hold model (L2): the signals core rebuilt from its
     // rulings. Lowering is the ratchet; raise only with a Size-Exception.
-    limit: "28.78 KB",
+    // Size-Exception (settle fast path, 2026-10-04): 28.78 -> 28.80 KB,
+    // measured at 28,785 B against `next` @ 924d909d9's 28,757 (+28 B; 5 B
+    // over the cap; +5 B minified, all of it the signals core). The same
+    // seam change as the simple-app floor note; brotli layout over this
+    // bundle (before #3791 the same +5 B measured +2 B here). Accepted by
+    // the maintainer (2026-10-04). The cap is frozen again at 28.80 KB
+    // (measured + 10 B).
+    limit: "28.80 KB",
     alias
   },
   {