diff --git a/system/public/aesthetic.computer/lib/visit-tracker.mjs b/system/public/aesthetic.computer/lib/visit-tracker.mjs
index 323f6f2746..dccc5906f5 100644
--- a/system/public/aesthetic.computer/lib/visit-tracker.mjs
+++ b/system/public/aesthetic.computer/lib/visit-tracker.mjs
@@ -30,6 +30,14 @@ export function startVisitTracker(win = window, doc = document) {
     visibleMs = 0; lastSent = ""; lastTick = win.performance.now();
     wasVisible = visible();
   };
+  const syncRoute = () => {
+    if (path === win.location.pathname) return;
+    path = win.location.pathname;
+    // Room/round URLs can change automatically. They are one page visit,
+    // not new arrivals. Only crossing a private boundary resets measurement.
+    const allowed = visitSurface(path) !== null;
+    if (allowed !== Boolean(state)) reset();
+  };
   const send = () => {
     if (!state || disabled() || stopped || win.performance.now() < retryAt) return;
     // Markers can be installed after the initial module by a render harness.
@@ -54,7 +62,7 @@ export function startVisitTracker(win = window, doc = document) {
   };
   const tick = () => {
     if (stopped) return;
-    if (path !== win.location.pathname) reset();
+    syncRoute();
     const now = win.performance.now();
     // Cap suspension gaps: a sleeping laptop is not engagement time.
     if (wasVisible) visibleMs += Math.min(2000, Math.max(0, now - lastTick));
@@ -72,7 +80,7 @@ export function startVisitTracker(win = window, doc = document) {
   };
   const action = name => {
     if (!VISIT_ACTIONS.includes(name) || !visible() || disabled()) return false;
-    if (path !== win.location.pathname) reset();
+    syncRoute();
     if (!state?.interacted) return false;
     if (!state.actions.includes(name)) state.actions.push(name);
     send(); return true;
diff --git a/system/tests/visit-tracking-browser.test.mjs b/system/tests/visit-tracking-browser.test.mjs
index fa4e0d3330..9307edefad 100644
--- a/system/tests/visit-tracking-browser.test.mjs
+++ b/system/tests/visit-tracking-browser.test.mjs
@@ -40,6 +40,9 @@ test("browser funnel, automation, privacy opt-out, private SPA routes and duplic
     assert.ok(received.some(row => row.interacted && row.activeSeconds >= 10));
     const ids = new Set(received.map(row => row.id));
     assert.equal(ids.size, 1);
+    await page.evaluate(() => history.replaceState({}, "", "/some-round"));
+    await page.clock.runFor(2000);
+    assert.equal(new Set(received.map(row => row.id)).size, 1, "automatic round URLs are not new visits");
     assert.ok(!JSON.stringify(received).includes("secret"));
     await page.evaluate(async () => (await import("https://aesthetic.computer/aesthetic.computer/lib/visit-tracker.mjs")).startVisitTracker());
     await page.evaluate(() => history.pushState({}, "", "/mail"));
diff --git a/toolchain/analytics/VISITS.md b/toolchain/analytics/VISITS.md
index 7ac180fe0d..af882fffe4 100644
--- a/toolchain/analytics/VISITS.md
+++ b/toolchain/analytics/VISITS.md
@@ -10,7 +10,7 @@ replays. It does not export operational records to PostHog.
 
 | Measure | Definition |
 | --- | --- |
-| Visit | One visible top-level page load, or new SPA pathname, with a random in-memory ID |
+| Visit | One visible top-level page load with a random in-memory ID; returning from a private route starts a fresh visit |
 | Interacted visit | Visible-page trusted pointer/touch/keyboard input outside form fields, or a pressed gamepad button |
 | Engaged visit | An interacted visit with at least 10 seconds of accumulated visible time |
 | Action visit | At least one occurrence of a reviewed action during that visit |
@@ -22,6 +22,8 @@ An automation framework can generate trusted events. Untagged automation may
 remain. IPs are not counted as people. No cross-page retention, unique-user,
 cross-domain journey or conversion-attribution claims can be made from these
 ephemeral IDs. A returning person loading three pages produces three visits.
+In-page SPA navigation and Oskiewar's automatic round-URL changes preserve the
+visit ID; the surface dimension describes the broad landing category.
 
 Visible-time lower-bound buckets are 0, 10, 30, 60, 180 and 600 seconds.
 This is foreground display time, not proof of attention. Polling gaps are