diff --git a/CHANGELOG.md b/CHANGELOG.md
index 960cbdc2..d57c72d3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -23,6 +23,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - **Highway arrangement fallback excludes drum arrangements.** When no explicit arrangement is requested, the most-notes fallback now skips drum arrangements (detected via manifest `type: drums` or legacy name keywords) so songs open on a guitar/bass part instead of the drum part. Drum-only packs still fall back to the first arrangement.
 - **Keys Highway 3D: a mastery change no longer resets the run's score.** Hits, misses, streak and best streak now carry over when the mastery slider (or Difficulty Ladder's auto-adjust) rebuilds the playable notes, so the stats posted at song end cover the whole run (#94).
+- **Keys Highway 3D: changing the hand filter mid-song no longer resets the run's score** (#130). Both / LH / RH rebuilds the playable notes but kept calling `_resetScoring()`, wiping hits, misses, streak and best streak — reachable mid-song from Split Screen's per-panel popover. Judged-note keys are `(t, midi)`-keyed so they stay valid across the rebuild and the miss sweep only runs over the visible notes, so only the sweep cursor is re-anchored, mirroring the mastery path (#94).
 - **Docker and Proxmox builds now include the starter feedpaks.** First-run seeding can find all listed sources instead of repeatedly warning that the packs are missing.
 
 - **`gp8_audio_sync` and `gp_autosync` now route GPIF parsing through the shared hardened parser (`lib/safe_xml.py`) instead of each carrying its own local `defusedxml`-or-stdlib fallback.** Both files parse `Content/score.gpif` out of an imported `.gp`/`.gpx` container — attacker-influenceable input — and previously duplicated the exact try/except-ImportError-then-stdlib-`ET.fromstring` pattern `safe_xml.py` centralized for every other untrusted-XML call site. Consolidating removes two divergent copies of security-relevant parsing logic; also corrected a stale `requirements.txt` comment that claimed both modules already delegated to `safe_xml.py`.
diff --git a/plugins/keys_highway_3d/plugin.json b/plugins/keys_highway_3d/plugin.json
index be45ca83..abc39dfb 100644
--- a/plugins/keys_highway_3d/plugin.json
+++ b/plugins/keys_highway_3d/plugin.json
@@ -1,7 +1,7 @@
 {
   "id": "keys_highway_3d",
   "name": "Keys Highway 3D",
-  "version": "0.2.6",
+  "version": "0.2.7",
   "description": "RS+-style 3D falling-note piano highway fed by the Sloppak Notation Format, with Web MIDI input scoring.",
   "type": "visualization",
   "bundled": true,
diff --git a/plugins/keys_highway_3d/screen.js b/plugins/keys_highway_3d/screen.js
index cf221601..00132245 100644
--- a/plugins/keys_highway_3d/screen.js
+++ b/plugins/keys_highway_3d/screen.js
@@ -3384,7 +3384,7 @@
 
         // Switch this instance's hand filter. A no-op when unchanged, so a
         // host re-applying the same value (splitscreen restores on every
-        // provider refresh) can't reset scoring mid-song. Meshes are rebuilt
+        // provider refresh) can't touch scoring mid-song. Meshes are rebuilt
         // only once the scene exists; before that, the next build reads the
         // refiltered `playable`.
         function _setHandFilter(next) {
@@ -3395,9 +3395,19 @@
             if (!notesGroup) return;
             const curTime = _latestTime;
             buildNoteMeshes();
-            _resetScoring();
-            // same as the mastery path: resume the sweep where playback is
-            // now — notes already elapsed must not re-sweep as misses
+            // Same as the mastery path, this is the SAME run: hits / misses
+            // / streak / best streak carry over, so changing the hand filter
+            // during play doesn't wipe the score (and the stats posted at
+            // song end cover the whole run). The judged-note sets are keyed
+            // by noteKey(t, midi), not by list index, so they stay valid
+            // across the new playable set — and keeping them stops a note
+            // already judged from being counted again if a later hand change
+            // brings it back into view. The miss sweep only runs over the
+            // visible `playable`, so a note hidden by the filter can never
+            // accrue a miss while hidden. Only the sweep cursor indexes the
+            // old list, so re-anchor it: playback kept advancing, and notes
+            // the player already passed must not be retroactively counted
+            // as misses.
             _anchorMissSweep(curTime);
         }
 
@@ -3631,13 +3641,12 @@
         // miss sweep at the current playback position instead of the list
         // head. _resetScoring() rewinds _sweepCursor to 0 — right for a fresh
         // chart, wrong here, where playback keeps advancing: a sweep from 0
-        // against a just-emptied _hitNoteKeys/_missedNoteKeys would re-mark
-        // every already-elapsed visible note (including ones already hit) as
-        // a miss and tank the run's accuracy. The floor keeps the tolerance
-        // band around the change retro-free (a "run restart", matching what
-        // the reset already does to hits/streak), and _latestTime is restored
-        // so a note-on landing before the next draw still judges against real
-        // song time, not the reset's 0.
+        // would re-mark every already-elapsed visible note never judged
+        // (including ones the previous filter hid and this change just
+        // revealed) as a miss and tank the run's accuracy. The floor keeps
+        // the tolerance band around the change retro-free, and _latestTime
+        // is restored so a note-on landing before the next draw still
+        // judges against real song time.
         function _anchorMissSweep(now) {
             _sweepCursor.idx = sweepStartIndex(_notation && _notation.playable, now, HIT_TOLERANCE_S);
             if (Number.isFinite(Number(now))) {
diff --git a/plugins/keys_highway_3d/tests/instance_invariants.test.js b/plugins/keys_highway_3d/tests/instance_invariants.test.js
index 92e3ba7e..5f845445 100644
--- a/plugins/keys_highway_3d/tests/instance_invariants.test.js
+++ b/plugins/keys_highway_3d/tests/instance_invariants.test.js
@@ -75,6 +75,19 @@ test('a mastery change keeps the run totals instead of resetting scoring', () =>
     assert.match(fn[0], /_anchorMissSweep\(now\)/);
 });
 
+test('a hand-filter change keeps the run totals instead of resetting scoring (#130)', () => {
+    // Same shape as the mastery path: changing Both / LH / RH mid-song
+    // rebuilds the playable notes, so wiping hits/misses/streak made the
+    // posted stats cover only the time since the change. Judged-note keys
+    // are (t, midi)-keyed so they survive the rebuild; the miss sweep only
+    // runs over the visible playable, so a hidden-hand note can never
+    // accrue a miss while hidden. Only the sweep cursor needs re-anchoring.
+    const fn = src.match(/function _setHandFilter\(next\)\s*\{[\s\S]*?\n {8}\}\n/);
+    assert.ok(fn, '_setHandFilter not found');
+    assert.doesNotMatch(fn[0], /_resetScoring\(\)/);
+    assert.match(fn[0], /_anchorMissSweep\(curTime\)/);
+});
+
 test('updateScene windows its arrays instead of walking every note (feedBack#95)', () => {
     // The window helpers are pure and unit-tested in data_layer.test.js, but
     // nothing there can catch them being exported and never wired up — the