diff --git a/README.md b/README.md
index d38d11b..01dcc1c 100644
--- a/README.md
+++ b/README.md
@@ -6,7 +6,7 @@
 
 [![Status](https://img.shields.io/badge/status-beta-yellow.svg)](#install)
 [![License](https://img.shields.io/badge/license-AGPL--3.0--or--later-blue.svg)](LICENSE)
-[![Version](https://img.shields.io/badge/version-2.0.1-success.svg)](https://github.com/ratspeak/rsDeck/releases)
+[![Version](https://img.shields.io/badge/version-2.0.2-success.svg)](https://github.com/ratspeak/rsDeck/releases)
 
 [Ratspeak](https://github.com/ratspeak/Ratspeak) |
 [Docs](https://ratspeak.org/docs.html) |
diff --git a/src/config/Config.h b/src/config/Config.h
index c40ca23..2b8f1ca 100644
--- a/src/config/Config.h
+++ b/src/config/Config.h
@@ -6,8 +6,8 @@
 
 #define RSDECK_VERSION_MAJOR  2
 #define RSDECK_VERSION_MINOR  0
-#define RSDECK_VERSION_PATCH  1
-#define RSDECK_VERSION_STRING "2.0.1"
+#define RSDECK_VERSION_PATCH  2
+#define RSDECK_VERSION_STRING "2.0.2"
 
 // --- Feature Flags ---
 #define HAS_DISPLAY     true
diff --git a/src/main.cpp b/src/main.cpp
index 55227b6..ea228d9 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -58,6 +58,7 @@
 #endif
 #include "config/UserConfig.h"
 #include "audio/AudioNotify.h"
+#include "util/PerfTrace.h"
 #include <ArduinoJson.h>
 #include <Preferences.h>
 #include <atomic>
@@ -1051,11 +1052,40 @@ static void bootRender() {
     // Legacy render kept as fallback
 }
 
+static unsigned long bootTraceStartMs = 0;
+static unsigned long bootTraceLastMs = 0;
+
+static void bootTraceBegin(unsigned long startMs) {
+#if RSDECK_PERF_TRACE
+    bootTraceStartMs = startMs;
+    bootTraceLastMs = startMs;
+#else
+    (void)startMs;
+#endif
+}
+
+static void bootTraceStage(const char* label) {
+#if RSDECK_PERF_TRACE
+    const unsigned long now = millis();
+    Serial.printf("[BOOT-PERF] %-22s +%lums total=%lums heap=%lu psram_free=%lu psram_largest=%lu\n",
+                  label ? label : "?",
+                  now - bootTraceLastMs,
+                  now - bootTraceStartMs,
+                  (unsigned long)ESP.getFreeHeap(),
+                  (unsigned long)heap_caps_get_free_size(MALLOC_CAP_SPIRAM),
+                  (unsigned long)heap_caps_get_largest_free_block(MALLOC_CAP_SPIRAM));
+    bootTraceLastMs = now;
+#else
+    (void)label;
+#endif
+}
+
 // =============================================================================
 // Setup — 26-step boot sequence
 // =============================================================================
 
 void setup() {
+    const unsigned long setupStartMs = millis();
     bool flashMounted = false;
 
     // Step 1: Power pin — CRITICAL: enables all T-Deck Plus peripherals
@@ -1086,17 +1116,21 @@ void setup() {
     Serial.printf("[BOOT] Reset: %s (%d)\n", reasonStr, (int)reason);
     Serial.printf("[BOOT] Heap: %lu  PSRAM: %lu\n",
                   (unsigned long)ESP.getFreeHeap(), (unsigned long)ESP.getPsramSize());
+    bootTraceBegin(setupStartMs);
+    bootTraceStage("serial-online");
 
     // Dual-boot layout: re-arm the launcher so the next reset shows the chooser.
     auto launcherBoot = rs_deck::returnToLauncherNextBoot();
     if (!launcherBoot.ok) {
         Serial.printf("[BOOT] Launcher return unavailable: %s\n", launcherBoot.message);
     }
+    bootTraceStage("launcher-return");
     if (!psramFound() || heap_caps_get_largest_free_block(MALLOC_CAP_SPIRAM) < 1024 * 1024) {
         Serial.printf("[BOOT] FATAL: PSRAM unavailable or too fragmented (largest=%lu)\n",
                       (unsigned long)heap_caps_get_largest_free_block(MALLOC_CAP_SPIRAM));
         while (true) delay(1000);
     }
+    bootTraceStage("psram-check");
 
     // Step 3: Initialize I2C bus (shared by keyboard + touchscreen)
     Wire.begin(I2C_SDA, I2C_SCL);
@@ -1108,6 +1142,7 @@ void setup() {
     // Deassert all slave CS pins to prevent bus contention
     pinMode(LORA_CS, OUTPUT); digitalWrite(LORA_CS, HIGH);
     pinMode(SD_CS, OUTPUT);   digitalWrite(SD_CS, HIGH);
+    bootTraceStage("i2c-spi");
 
     // Mount flash before radio bring-up so persisted RF settings are used from
     // the first SX1262 init, instead of always booting at the US default first.
@@ -1120,6 +1155,7 @@ void setup() {
     }
     // Select palette before any LVGL styles are built
     Theme::setScheme(userConfig.settings().themeLight ? Theme::Scheme::LIGHT : Theme::Scheme::DARK);
+    bootTraceStage("early-flash-config");
 
     // Step 4: Radio + SD init BEFORE display
     // Radio and SD must init while SPIClass exclusively owns SPI2_HOST.
@@ -1134,6 +1170,7 @@ void setup() {
     } else {
         Serial.println("[RADIO] SX1262 not detected!");
     }
+    bootTraceStage("radio-init");
 
     // SD card init (shared SPI, right after radio)
     digitalWrite(LORA_CS, HIGH);
@@ -1145,6 +1182,7 @@ void setup() {
     } else {
         Serial.println("[SD] Not detected");
     }
+    bootTraceStage("sd-probe");
 
     // Verify radio SPI still works after SD init
     if (radioOnline) {
@@ -1161,6 +1199,7 @@ void setup() {
     // SPIClass get valid device handles on the same SPI2_HOST bus.
     display.begin();
     Serial.println("[BOOT] Display initialized (LovyanGFX direct)");
+    bootTraceStage("display-init");
 
     // Step 5.5: Initialize LVGL display driver
     if (!display.beginLVGL()) {
@@ -1171,6 +1210,7 @@ void setup() {
         while (true) delay(1000);
     }
     Serial.println("[BOOT] LVGL initialized");
+    bootTraceStage("lvgl-init");
 
     // Verify radio SPI survives display init
     if (radioOnline) {
@@ -1191,21 +1231,25 @@ void setup() {
     // framebuffer; the setProgress() above has now flushed the boot screen.
     // powerMgr at step 24 overrides with the user's configured value.
     display.setBrightness(128);
+    bootTraceStage("boot-screen-painted");
 
     // Step 7: Touch HAL — GT911 I2C
     touch.begin();
     lvBootScreen.setProgress(0.50f, "Touch ready");
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("touch-init");
 
     // Step 8: Keyboard HAL — ESP32-C3 I2C
     keyboard.begin();
     lvBootScreen.setProgress(0.52f, "Keyboard ready");
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("keyboard-init");
 
     // Step 9: Trackball HAL — GPIO interrupts
     trackball.begin();
     lvBootScreen.setProgress(0.54f, "Trackball ready");
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("trackball-init");
 
     // Step 10: Input manager
     inputManager.begin(&keyboard, &trackball, &touch);
@@ -1216,6 +1260,7 @@ void setup() {
 
     lvBootScreen.setProgress(0.55f, "Input ready");
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("input-init");
 
     // Step 11: Register hotkeys
     hotkeys.registerHotkey('h', "Help", onHotkeyHelp);
@@ -1234,6 +1279,7 @@ void setup() {
     });
     lvBootScreen.setProgress(0.58f, "Hotkeys registered");
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("hotkeys");
 
     // Step 12: Mount LittleFS
     lvBootScreen.setProgress(0.60f, "Mounting flash...");
@@ -1247,6 +1293,7 @@ void setup() {
         flashMounted = true;
         Serial.println("[BOOT] LittleFS mounted OK");
     }
+    bootTraceStage("flash-mounted");
 
     // Step 13: Boot loop detection (NVS)
     {
@@ -1261,6 +1308,7 @@ void setup() {
             }
         }
     }
+    bootTraceStage("bootloop-nvs");
 
     lvBootScreen.setProgress(0.64f, "Loading config...");
     userConfig.load(sdStore, flash);
@@ -1271,6 +1319,7 @@ void setup() {
     }
     inputManager.setTrackballSpeed(userConfig.settings().trackballSpeed);
     applyRadioSettingsToHardware(userConfig.settings(), "BOOT PRE-RNS");
+    bootTraceStage("config-load");
 
     lvBootScreen.setProgress(0.65f, "Starting Reticulum...");
     // (LVGL boot renders via lv_timer_handler in setProgress)
@@ -1283,14 +1332,17 @@ void setup() {
         lvBootScreen.setProgress(0.72f, "RNS: FAILED");
     }
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("reticulum-begin");
 
     // Step 15.5: Identity manager
     identityMgr.begin(&flash, &sdStore);
+    bootTraceStage("identity-manager");
 
     // Step 16: Message store
     lvBootScreen.setProgress(0.72f, "Starting messaging...");
     // (LVGL boot renders via lv_timer_handler in setProgress)
     messageStore.begin(&flash, &sdStore, userConfig.settings().sdStorageEnabled);
+    bootTraceStage("message-store");
 
     // Step 17: LXMF init
     lxmf.begin(&rns, &messageStore);
@@ -1303,6 +1355,7 @@ void setup() {
     lxmf.unreadCount();
     lvBootScreen.setProgress(0.75f, "LXMF ready");
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("lxmf-begin");
 
     // Step 18: Announce manager
     lvBootScreen.setProgress(0.78f, "Loading contacts...");
@@ -1317,6 +1370,7 @@ void setup() {
     announceManager->loadNameCache();
     announceHandler = RNS::HAnnounceHandler(announceManager);
     RNS::Transport::register_announce_handler(announceHandler);
+    bootTraceStage("contacts-cache");
 
     // No default TCP hub.  Users opt in via Settings → TCP Server →
     // "Ratspeak Hub" (seeds rns.ratspeak.org) or "Custom" (host/port).
@@ -1342,6 +1396,7 @@ void setup() {
             }
         }
     }
+    bootTraceStage("identity-name-sync");
 
     // Step 20: Boot loop recovery
     if (bootLoopRecovery) {
@@ -1350,6 +1405,7 @@ void setup() {
     }
     lvBootScreen.setProgress(0.83f, "Config loaded");
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("bootloop-recovery");
 
     // Step 21: Apply radio config
     if (radioOnline && userConfig.settings().loraEnabled) {
@@ -1362,6 +1418,7 @@ void setup() {
     }
     lvBootScreen.setProgress(0.84f, "Radio configured");
     // (LVGL boot renders via lv_timer_handler in setProgress)
+    bootTraceStage("radio-config");
 
     // Step 22: WiFi start
     RatWiFiMode wifiMode = userConfig.settings().wifiMode;
@@ -1418,6 +1475,7 @@ void setup() {
         lvBootScreen.setProgress(0.87f, "WiFi disabled");
         // (LVGL boot renders via lv_timer_handler in setProgress)
     }
+    bootTraceStage("wifi-start");
 
     // Step 23: BLE stays disabled in default builds.
     lvBootScreen.setProgress(0.90f, "Links ready");
@@ -1449,6 +1507,7 @@ void setup() {
     ui.lvStatusBar().setBLEActive(false);
     Serial.println("[BLE] Disabled in default firmware build");
 #endif
+    bootTraceStage("links-ready");
 
     // Step 24: Power manager
     lvBootScreen.setProgress(0.92f, "Power manager...");
@@ -1460,6 +1519,7 @@ void setup() {
     powerMgr.setKbBrightness(userConfig.settings().keyboardBrightness);
     powerMgr.setKbAutoOn(userConfig.settings().keyboardAutoOn);
     powerMgr.setKbAutoOff(userConfig.settings().keyboardAutoOff);
+    bootTraceStage("power-manager");
 
     // Step 24.5: GPS init
 #if HAS_GPS
@@ -1469,6 +1529,7 @@ void setup() {
         gps.setLocationEnabled(userConfig.settings().gpsLocationEnabled);
         gps.begin();
         Serial.println("[BOOT] GPS UART started (MIA-M10Q)");
+        bootTraceStage("gps-start");
     }
 #endif
 
@@ -1478,6 +1539,7 @@ void setup() {
     audio.setEnabled(userConfig.settings().audioEnabled);
     audio.setVolume(userConfig.settings().audioVolume);
     audio.begin();
+    bootTraceStage("audio-init");
 
     // Boot complete — transition to Home screen
     // Yield to LVGL instead of blocking delay
@@ -1485,6 +1547,7 @@ void setup() {
     for (int i = 0; i < 6; i++) { lv_timer_handler(); delay(1); }
     lvBootScreen.setProgress(1.0f, "Ready");
     audio.playBoot();
+    bootTraceStage("boot-ready-screen");
 
     bootComplete = true;
 
@@ -1535,14 +1598,14 @@ void setup() {
     lvHomeScreen.setTCPToggleCallback([]() {
         auto& s = userConfig.settings();
         bool enabled = false;
-        bool hasSavedRelay = false;
+        bool hasSavedTcpServer = false;
         for (const auto& ep : s.tcpConnections) {
-            if (!ep.host.isEmpty()) hasSavedRelay = true;
+            if (!ep.host.isEmpty()) hasSavedTcpServer = true;
             if (!ep.host.isEmpty() && ep.autoConnect) { enabled = true; break; }
         }
         if (enabled) {
             for (auto& ep : s.tcpConnections) ep.autoConnect = false;
-        } else if (hasSavedRelay) {
+        } else if (hasSavedTcpServer) {
             for (auto& ep : s.tcpConnections) {
                 if (!ep.host.isEmpty()) ep.autoConnect = true;
             }
@@ -1556,9 +1619,9 @@ void setup() {
         }
         bool ok = userConfig.save(sdStore, flash);
         ui.lvStatusBar().showToast(
-            ok ? "TCP relay saved; reboot to apply" : "Save failed",
+            ok ? "TCP server saved; reboot to apply" : "Save failed",
             ok ? 3000 : 2000);
-        Serial.printf("[TCP] Saved relay %s (save %s, reboot required)\n",
+        Serial.printf("[TCP] Saved server %s (save %s, reboot required)\n",
                       enabled ? "OFF" : "ON",
                       ok ? "OK" : "FAILED");
     });
@@ -1720,6 +1783,7 @@ void setup() {
     ui.lvTabBar().setTabCallback([](int tab) {
         if (lvTabScreens[tab]) ui.setScreen(lvTabScreens[tab]);
     });
+    bootTraceStage("screen-wiring");
 
     // Data clean screen (first boot only — when SD has old data)
     lvDataCleanScreen.setDoneCallback([](bool wipe) {
@@ -1839,6 +1903,7 @@ void setup() {
         // Everything configured — go straight to home
         goHome();
     }
+    bootTraceStage("boot-routing");
 
     // Clear boot loop counter — we survived!
     {
@@ -1848,17 +1913,20 @@ void setup() {
             prefs.end();
         }
     }
+    bootTraceStage("bootcounter-clear");
 
     if (userConfig.settings().keyboardAutoOn) {
         // We are in ACTIVE power state here, switch keyboard backlight ON
         keyboard.backlightOn();
     }
+    bootTraceStage("keyboard-auto");
 
     Serial.println("[BOOT] rsDeck ready");
     Serial.printf("[BOOT] Summary: radio=%s flash=%s sd=%s\n",
                   radioOnline ? "ONLINE" : "OFFLINE",
                   flash.isReady() ? "OK" : "FAIL",
                   sdStore.isReady() ? "OK" : "FAIL");
+    bootTraceStage("setup-complete");
 }
 
 // =============================================================================
diff --git a/src/reticulum/AnnounceManager.cpp b/src/reticulum/AnnounceManager.cpp
index e6bb272..c2cb935 100644
--- a/src/reticulum/AnnounceManager.cpp
+++ b/src/reticulum/AnnounceManager.cpp
@@ -4,6 +4,7 @@
 #include "storage/SDStore.h"
 #include "storage/FlashStore.h"
 #include "transport/LoRaInterface.h"
+#include "util/PerfTrace.h"
 #include <ArduinoJson.h>
 #include <LittleFS.h>
 
@@ -279,9 +280,11 @@ void AnnounceManager::persistKnownDestinationsAfterAnnounce(const char* reason,
     }
 
     _lastKnownDestinationsPersist = now;
+    unsigned long startMs = millis();
     RNS::Identity::persist_data();
-    Serial.printf("[ANNOUNCE] Known destinations persisted after %s\n",
-                  reason ? reason : "announce");
+    unsigned long elapsed = millis() - startMs;
+    Serial.printf("[ANNOUNCE] Known destinations persisted after %s (force=%s in %lums)\n",
+                  reason ? reason : "announce", force ? "yes" : "no", elapsed);
 }
 
 int AnnounceManager::nodesOnlineSince(unsigned long maxAgeMs) const {
@@ -470,19 +473,36 @@ std::string AnnounceManager::lookupName(const std::string& hexHash) const {
 }
 
 void AnnounceManager::saveNameCache() {
+    unsigned long startMs = millis();
     JsonDocument doc;
     for (auto& kv : _nameCache) {
         doc[kv.first] = kv.second;
     }
     String json;
+    unsigned long serializeStartMs = millis();
     serializeJson(doc, json);
+    unsigned long serializeMs = millis() - serializeStartMs;
+    size_t bytes = json.length();
+    bool sdOk = false;
+    bool flashOk = false;
+    unsigned long sdMs = 0;
+    unsigned long flashMs = 0;
     if (_sd && _sd->isReady()) {
-        _sd->writeString("/ratdeck/config/names.json", json);
+        unsigned long writeStartMs = millis();
+        sdOk = _sd->writeString("/ratdeck/config/names.json", json);
+        sdMs = millis() - writeStartMs;
     }
     if (_flash) {
-        _flash->writeString("/config/names.json", json);
+        unsigned long writeStartMs = millis();
+        flashOk = _flash->writeString("/config/names.json", json);
+        flashMs = millis() - writeStartMs;
     }
-    Serial.printf("[ANNOUNCE] Name cache saved (%d entries)\n", (int)_nameCache.size());
+    unsigned long elapsed = millis() - startMs;
+    Serial.printf("[ANNOUNCE] Name cache saved (%d entries, bytes=%u, serialize=%lums sd=%s/%lums flash=%s/%lums total=%lums)\n",
+                  (int)_nameCache.size(), (unsigned)bytes, serializeMs,
+                  (_sd && _sd->isReady()) ? (sdOk ? "ok" : "fail") : "skip", sdMs,
+                  _flash ? (flashOk ? "ok" : "fail") : "skip", flashMs,
+                  elapsed);
 }
 
 void AnnounceManager::loadNameCache() {
diff --git a/src/reticulum/LXMFManager.cpp b/src/reticulum/LXMFManager.cpp
index 2965c9d..b9292a9 100644
--- a/src/reticulum/LXMFManager.cpp
+++ b/src/reticulum/LXMFManager.cpp
@@ -22,11 +22,11 @@ bool LXMFManager::begin(ReticulumManager* rns, MessageStore* store) {
     dest.set_packet_callback(onPacketReceived);
     dest.set_link_established_callback(onLinkEstablished);
     if (_store) {
-        for (const auto& id : _store->loadRecentMessageIds(MAX_SEEN_IDS)) {
+        for (const auto& id : _store->startupRecentMessageIds(MAX_SEEN_IDS)) {
             rememberMessageId(id);
         }
-        std::vector<LXMFMessage> pending = _store->loadPendingOutgoing();
-        for (auto& msg : pending) {
+        const std::vector<LXMFMessage>& pending = _store->startupPendingOutgoing();
+        for (auto msg : pending) {
             if ((int)_outQueue.size() >= RSDECK_MAX_OUTQUEUE) break;
             msg.lastRetryMs = 0;
             _outQueue.push_back(msg);
@@ -508,6 +508,11 @@ std::vector<LXMFMessage> LXMFManager::getMessages(const std::string& peerHex) co
     return {};
 }
 
+std::vector<LXMFMessage> LXMFManager::getRecentMessages(const std::string& peerHex, size_t maxMessages) const {
+    if (_store) return _store->loadConversationTail(peerHex, maxMessages);
+    return {};
+}
+
 int LXMFManager::unreadCount(const std::string& peerHex) const {
     if (!_store) return 0;
     if (peerHex.empty()) return _store->totalUnreadCount();
diff --git a/src/reticulum/LXMFManager.h b/src/reticulum/LXMFManager.h
index d4c6978..ac307f7 100644
--- a/src/reticulum/LXMFManager.h
+++ b/src/reticulum/LXMFManager.h
@@ -35,6 +35,7 @@ public:
     uint32_t storeRevision() const { return _store ? _store->revision() : 0; }
     const std::vector<std::string>& conversations() const;
     std::vector<LXMFMessage> getMessages(const std::string& peerHex) const;
+    std::vector<LXMFMessage> getRecentMessages(const std::string& peerHex, size_t maxMessages) const;
     int unreadCount(const std::string& peerHex = "") const;
     void markRead(const std::string& peerHex);
     bool deleteConversation(const std::string& peerHex);
diff --git a/src/reticulum/ReticulumManager.cpp b/src/reticulum/ReticulumManager.cpp
index 12f59b2..37be62c 100644
--- a/src/reticulum/ReticulumManager.cpp
+++ b/src/reticulum/ReticulumManager.cpp
@@ -1,8 +1,10 @@
 // Direct port from Ratputer — microReticulum integration
 #include "ReticulumManager.h"
 #include "config/Config.h"
+#include "util/PerfTrace.h"
 #include <LittleFS.h>
 #include <Preferences.h>
+#include <cstring>
 #include <unordered_map>
 #include <string>
 
@@ -22,6 +24,7 @@ size_t LittleFSFileSystem::read_file(const char* p, RNS::Bytes& data) {
 }
 
 size_t LittleFSFileSystem::write_file(const char* p, const RNS::Bytes& data) {
+    unsigned long startMs = PerfTrace::nowMs();
     String path = String(p);
     int lastSlash = path.lastIndexOf('/');
     if (lastSlash > 0) {
@@ -29,9 +32,13 @@ size_t LittleFSFileSystem::write_file(const char* p, const RNS::Bytes& data) {
         if (!LittleFS.exists(dir.c_str())) { LittleFS.mkdir(dir.c_str()); }
     }
     File f = LittleFS.open(p, "w");
-    if (!f) return 0;
+    if (!f) {
+        PerfTrace::write("flash", "rns-write", p, data.size(), startMs, false);
+        return 0;
+    }
     size_t w = f.write(data.data(), data.size());
     f.close();
+    PerfTrace::write("flash", "rns-write", p, data.size(), startMs, w == data.size());
     return w;
 }
 
@@ -74,19 +81,29 @@ bool copyFlashToSD(SDStore* sd, const char* flashPath, const char* sdPath) {
     if (!sd || !sd->isReady()) return false;
     File in = LittleFS.open(flashPath, "r");
     if (!in || in.size() == 0) { if (in) in.close(); return false; }
+    unsigned long startMs = PerfTrace::nowMs();
+    size_t copiedBytes = 0;
 
     String path = String(sdPath);
     int lastSlash = path.lastIndexOf('/');
     if (lastSlash > 0) {
         String dir = path.substring(0, lastSlash);
-        if (!sd->ensureDir(dir.c_str())) { in.close(); return false; }
+        if (!sd->ensureDir(dir.c_str())) {
+            in.close();
+            PerfTrace::write("sd", "rns-copy", sdPath, copiedBytes, startMs, false);
+            return false;
+        }
     }
 
     String tmpPath = path + ".tmp";
     String bakPath = path + ".bak";
     SD.remove(tmpPath.c_str());
     File out = SD.open(tmpPath.c_str(), FILE_WRITE);
-    if (!out) { in.close(); return false; }
+    if (!out) {
+        in.close();
+        PerfTrace::write("sd", "rns-copy", sdPath, copiedBytes, startMs, false);
+        return false;
+    }
 
     uint8_t buf[RNS_COPY_CHUNK];
     bool ok = true;
@@ -94,12 +111,17 @@ bool copyFlashToSD(SDStore* sd, const char* flashPath, const char* sdPath) {
         size_t n = in.read(buf, sizeof(buf));
         if (n == 0) break;
         if (out.write(buf, n) != n) { ok = false; break; }
+        copiedBytes += n;
         RNS::Utilities::OS::reset_watchdog();
     }
     in.close();
     out.close();
 
-    if (!ok) { SD.remove(tmpPath.c_str()); return false; }
+    if (!ok) {
+        SD.remove(tmpPath.c_str());
+        PerfTrace::write("sd", "rns-copy", sdPath, copiedBytes, startMs, false);
+        return false;
+    }
     if (SD.exists(sdPath)) {
         SD.remove(bakPath.c_str());
         SD.rename(sdPath, bakPath.c_str());
@@ -108,24 +130,53 @@ bool copyFlashToSD(SDStore* sd, const char* flashPath, const char* sdPath) {
     if (!SD.rename(tmpPath.c_str(), sdPath)) {
         if (SD.exists(bakPath.c_str())) SD.rename(bakPath.c_str(), sdPath);
         SD.remove(tmpPath.c_str());
+        PerfTrace::write("sd", "rns-copy", sdPath, copiedBytes, startMs, false);
         return false;
     }
     SD.remove(bakPath.c_str());
+    PerfTrace::write("sd", "rns-copy", sdPath, copiedBytes, startMs, true);
     return true;
 }
 
 class LittleFSStreamImpl : public RNS::FileStreamImpl {
 public:
-    LittleFSStreamImpl(File&& f) : _f(std::move(f)) {}
-    ~LittleFSStreamImpl() override { if (_f) _f.close(); }
+    LittleFSStreamImpl(File&& f, const char* path, RNS::FileStream::MODE mode)
+        : _f(std::move(f)), _mode(mode), _startMs(PerfTrace::nowMs()) {
+        strncpy(_path, path ? path : "?", sizeof(_path) - 1);
+        _path[sizeof(_path) - 1] = '\0';
+    }
+    ~LittleFSStreamImpl() override { close(); }
 
 protected:
-    const char* name() override { return _f ? _f.name() : ""; }
+    const char* name() override { return _path; }
     size_t size() override { return _f ? _f.size() : 0; }
-    void close() override { if (_f) _f.close(); }
+    void close() override {
+        if (!_f) return;
+        if (!_logged && _mode != RNS::FileStream::MODE_READ) {
+            PerfTrace::write("flash",
+                             _mode == RNS::FileStream::MODE_APPEND ? "rns-stream-append" : "rns-stream",
+                             _path, _bytesWritten, _startMs, !_writeFailed);
+            _logged = true;
+        }
+        _f.close();
+    }
 
-    size_t write(uint8_t byte) override { return _f ? _f.write(byte) : 0; }
-    size_t write(const uint8_t* buffer, size_t len) override { return _f ? _f.write(buffer, len) : 0; }
+    size_t write(uint8_t byte) override {
+        size_t written = _f ? _f.write(byte) : 0;
+        if (_mode != RNS::FileStream::MODE_READ) {
+            _bytesWritten += written;
+            if (written != 1) _writeFailed = true;
+        }
+        return written;
+    }
+    size_t write(const uint8_t* buffer, size_t len) override {
+        size_t written = _f ? _f.write(buffer, len) : 0;
+        if (_mode != RNS::FileStream::MODE_READ) {
+            _bytesWritten += written;
+            if (written != len) _writeFailed = true;
+        }
+        return written;
+    }
 
     int available() override { return _f ? _f.available() : 0; }
     int read() override { return _f ? _f.read() : -1; }
@@ -134,6 +185,12 @@ protected:
 
 private:
     File _f;
+    RNS::FileStream::MODE _mode;
+    unsigned long _startMs = 0;
+    size_t _bytesWritten = 0;
+    bool _writeFailed = false;
+    bool _logged = false;
+    char _path[64] = {};
 };
 }  // namespace
 
@@ -150,7 +207,7 @@ RNS::FileStream LittleFSFileSystem::open_file(const char* path, RNS::FileStream:
     }
     File f = LittleFS.open(path, openMode);
     if (!f) return {RNS::Type::NONE};
-    return RNS::FileStream(new LittleFSStreamImpl(std::move(f)));
+    return RNS::FileStream(new LittleFSStreamImpl(std::move(f), path, mode));
 }
 bool LittleFSFileSystem::remove_file(const char* p) { return LittleFS.remove(p); }
 bool LittleFSFileSystem::rename_file(const char* f, const char* t) { return LittleFS.rename(f, t); }
@@ -359,20 +416,37 @@ void ReticulumManager::loop() {
 // node, so path tables are intentionally not persisted across boots.
 void ReticulumManager::persistData() {
     unsigned long start = millis();
+    unsigned long identityPersistMs = 0;
+    unsigned long sdMirrorMs = 0;
+    size_t sdMirrorBytes = 0;
+    bool sdMirrorAttempted = false;
+    bool sdMirrorOk = false;
+    const char* cycleName = "identity";
     switch (_persistCycle) {
         case 0:
+        {
+            unsigned long phaseMs = millis();
             RNS::Identity::persist_data();
+            identityPersistMs = millis() - phaseMs;
             break;
+        }
         case 1:
+            cycleName = "sd-mirror";
             if (_sd && _sd->isReady()) {
                 static const char* files[] = {"/known_destinations"};
                 for (const char* name : files) {
                     File f = LittleFS.open(name, "r");
                     if (f && f.size() > 0) {
+                        size_t len = f.size();
                         f.close();
                         char sdPath[64];
                         snprintf(sdPath, sizeof(sdPath), "/ratdeck/transport%s", name);
-                        copyFlashToSD(_sd, name, sdPath);
+                        unsigned long phaseMs = millis();
+                        bool ok = copyFlashToSD(_sd, name, sdPath);
+                        sdMirrorMs += millis() - phaseMs;
+                        sdMirrorBytes += ok ? len : 0;
+                        sdMirrorAttempted = true;
+                        sdMirrorOk = sdMirrorOk || ok;
                     } else {
                         if (f) f.close();
                     }
@@ -381,7 +455,10 @@ void ReticulumManager::persistData() {
             break;
     }
     unsigned long dur = millis() - start;
-    Serial.printf("[PERSIST] Cycle %d done (%lums)\n", _persistCycle, dur);
+    Serial.printf("[PERSIST] Cycle %d %s done total=%lums identity=%lums sd_copy=%lums sd_bytes=%u sd=%s\n",
+                  _persistCycle, cycleName, dur, identityPersistMs, sdMirrorMs,
+                  (unsigned)sdMirrorBytes,
+                  sdMirrorAttempted ? (sdMirrorOk ? "ok" : "fail") : "skip");
     if (dur > 500) {
         Serial.printf("[PERSIST] WARNING: Cycle %d blocked for %lums!\n", _persistCycle, dur);
     }
diff --git a/src/storage/FlashStore.cpp b/src/storage/FlashStore.cpp
index 61b70c3..aabc257 100644
--- a/src/storage/FlashStore.cpp
+++ b/src/storage/FlashStore.cpp
@@ -1,4 +1,5 @@
 #include "FlashStore.h"
+#include "util/PerfTrace.h"
 
 bool FlashStore::begin() {
     // Legacy standalone builds label this partition "littlefs"; bmorcelli/Launcher
@@ -71,17 +72,25 @@ bool FlashStore::remove(const char* path) {
 }
 
 bool FlashStore::writeAtomic(const char* path, const uint8_t* data, size_t len) {
-    if (!_ready) return false;
+    unsigned long startMs = PerfTrace::nowMs();
+    if (!_ready) {
+        PerfTrace::write("flash", "atomic", path, len, startMs, false);
+        return false;
+    }
 
     String tmpPath = String(path) + ".tmp";
     String bakPath = String(path) + ".bak";
 
     File f = LittleFS.open(tmpPath.c_str(), "w");
-    if (!f) return false;
+    if (!f) {
+        PerfTrace::write("flash", "atomic", path, len, startMs, false);
+        return false;
+    }
     size_t written = f.write(data, len);
     f.close();
     if (written != len) {
         LittleFS.remove(tmpPath.c_str());
+        PerfTrace::write("flash", "atomic", path, len, startMs, false);
         return false;
     }
 
@@ -89,6 +98,7 @@ bool FlashStore::writeAtomic(const char* path, const uint8_t* data, size_t len)
     if (!verify || verify.size() != len) {
         if (verify) verify.close();
         LittleFS.remove(tmpPath.c_str());
+        PerfTrace::write("flash", "atomic", path, len, startMs, false);
         return false;
     }
     verify.close();
@@ -102,12 +112,14 @@ bool FlashStore::writeAtomic(const char* path, const uint8_t* data, size_t len)
         if (LittleFS.exists(bakPath.c_str())) {
             LittleFS.rename(bakPath.c_str(), path);
         }
+        PerfTrace::write("flash", "atomic", path, len, startMs, false);
         return false;
     }
 
     // Clean up backup file after successful write
     LittleFS.remove(bakPath.c_str());
 
+    PerfTrace::write("flash", "atomic", path, len, startMs, true);
     return true;
 }
 
diff --git a/src/storage/MessageStore.cpp b/src/storage/MessageStore.cpp
index 41c2645..ca518ed 100644
--- a/src/storage/MessageStore.cpp
+++ b/src/storage/MessageStore.cpp
@@ -1,5 +1,6 @@
 #include "MessageStore.h"
 #include "config/Config.h"
+#include "util/PerfTrace.h"
 #include <LittleFS.h>
 #include <ArduinoJson.h>
 #include <Preferences.h>
@@ -54,10 +55,13 @@ bool MessageStore::begin(FlashStore* flash, SDStore* sd, bool externalStorageEna
     buildSummaries();
     unsigned long summaryMs = millis() - phaseMs;
     unsigned long totalMs = millis() - beginMs;
-    if (totalMs > 100 || summaryMs > 50) {
-        Serial.printf("[PERF] MSG begin: total=%lums migrate=%lums trunc=%lums counter=%lums refresh=%lums summaries=%lums\n",
-                      totalMs, migrateMs, truncMs, counterMs, refreshMs, summaryMs);
-    }
+#if RSDECK_PERF_TRACE
+    Serial.printf("[PERF] MSG begin: total=%lums migrate=%lums trunc=%lums counter=%lums refresh=%lums summaries=%lums convs=%d summary_count=%d ext=%s sd=%s\n",
+                  totalMs, migrateMs, truncMs, counterMs, refreshMs, summaryMs,
+                  (int)_conversations.size(), (int)_summaries.size(),
+                  _externalStorageEnabled ? "on" : "off",
+                  (_sd && _sd->isReady()) ? "ready" : "no");
+#endif
     Serial.printf("[MSGSTORE] %d conversations found, receive counter=%lu\n",
                   (int)_conversations.size(), (unsigned long)_nextReceiveCounter);
     return true;
@@ -286,6 +290,7 @@ bool MessageStore::saveMessage(LXMFMessage& msg) {
     std::string peerHex = msg.incoming ?
         msg.sourceHash.toHex() : msg.destHash.toHex();
 
+    unsigned long serializeStartMs = millis();
     JsonDocument doc;
     doc["src"] = msg.sourceHash.toHex();
     doc["dst"] = msg.destHash.toHex();
@@ -301,6 +306,8 @@ bool MessageStore::saveMessage(LXMFMessage& msg) {
 
     String json;
     serializeJson(doc, json);
+    unsigned long serializeMs = millis() - serializeStartMs;
+    size_t jsonBytes = json.length();
 
     // Counter-based filename: unique, monotonic, sorts correctly
     uint32_t counter = _nextReceiveCounter++;
@@ -310,27 +317,37 @@ bool MessageStore::saveMessage(LXMFMessage& msg) {
              (unsigned long)counter, msg.incoming ? 'i' : 'o');
 
     // Persist counter to NVS
+    unsigned long nvsStartMs = millis();
     {
         Preferences p;
         p.begin("ratdeck_msg", false);
         p.putUInt("msgctr", _nextReceiveCounter);
         p.end();
     }
+    unsigned long nvsMs = millis() - nvsStartMs;
 
     bool sdOk = false;
     bool flashOk = false;
+    bool sdAttempted = false;
+    unsigned long sdMs = 0;
+    unsigned long flashMs = 0;
 
     if (_externalStorageEnabled && _sd && _sd->isReady()) {
+        sdAttempted = true;
+        unsigned long sdStartMs = millis();
         String sdDir = sdConversationDir(peerHex);
         _sd->ensureDir(sdDir.c_str());
         String sdPath = sdDir + "/" + filename;
         sdOk = _sd->writeString(sdPath.c_str(), json);
+        sdMs = millis() - sdStartMs;
     }
 
+    unsigned long flashStartMs = millis();
     String flashDir = conversationDir(peerHex);
     _flash->ensureDir(flashDir.c_str());
     String flashPath = flashDir + "/" + filename;
     flashOk = _flash->writeString(flashPath.c_str(), json);
+    flashMs = millis() - flashStartMs;
     bool saved = sdOk || flashOk;
 
     bool found = false;
@@ -339,10 +356,13 @@ bool MessageStore::saveMessage(LXMFMessage& msg) {
     }
     if (!found) _conversations.push_back(peerHex);
 
+    unsigned long enforceStartMs = millis();
     if (sdOk) enforceSDLimit(peerHex);
     if (flashOk) enforceFlashLimit(peerHex);
+    unsigned long enforceMs = millis() - enforceStartMs;
 
     // Update summary cache
+    unsigned long summaryStartMs = millis();
     {
         auto& s = _summaries[peerHex];
         s.lastTimestamp = msg.timestamp;
@@ -368,24 +388,46 @@ bool MessageStore::saveMessage(LXMFMessage& msg) {
             rebuildSummary(peerHex);
         }
     }
+    unsigned long summaryMs = millis() - summaryStartMs;
 
     if (saved) bumpRevision();
     unsigned long elapsed = millis() - startMs;
-    if (elapsed > 30) {
-        Serial.printf("[PERF] MSG save: %s dir=%c sd=%s flash=%s in %lums\n",
+    if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_MSG_TRACE_MS) ||
+        PerfTrace::shouldLog(sdMs, RSDECK_PERF_WRITE_TRACE_MS) ||
+        PerfTrace::shouldLog(flashMs, RSDECK_PERF_WRITE_TRACE_MS) ||
+        !saved) {
+        Serial.printf("[PERF] MSG save: peer=%s dir=%c bytes=%u counter=%lu sd=%s/%lums flash=%s/%lums serialize=%lums nvs=%lums enforce=%lums summary=%lums total=%lums\n",
                       peerHex.substr(0, 8).c_str(), msg.incoming ? 'i' : 'o',
-                      sdOk ? "ok" : "no", flashOk ? "ok" : "no",
-                      (unsigned long)elapsed);
+                      (unsigned)jsonBytes, (unsigned long)counter,
+                      sdAttempted ? (sdOk ? "ok" : "fail") : "skip", sdMs,
+                      flashOk ? "ok" : "fail", flashMs,
+                      serializeMs, nvsMs, enforceMs, summaryMs, elapsed);
     }
     return saved;
 }
 
 std::vector<LXMFMessage> MessageStore::loadConversation(const std::string& peerHex) const {
-    std::vector<LXMFMessage> messages;
+    return loadConversationTail(peerHex, 0);
+}
 
-    auto loadFromDir = [&](File& d, auto readFileFn) {
+std::vector<LXMFMessage> MessageStore::loadConversationTail(const std::string& peerHex, size_t maxMessages) const {
+    std::vector<LXMFMessage> messages;
+    unsigned long startMs = millis();
+    unsigned long collectMs = 0;
+    unsigned long sortMs = 0;
+    unsigned long readMs = 0;
+    unsigned long parseMs = 0;
+    int filesSeen = 0;
+    int parsedMessages = 0;
+    int parseFailures = 0;
+    size_t bytesRead = 0;
+    const char* backend = "none";
+
+    auto loadFromDir = [&](File& d, auto readFileFn, const char* source) {
+        backend = source;
         // Collect filenames first, then sort alphabetically (counter prefix = insertion order)
         std::vector<String> filenames;
+        unsigned long phaseMs = millis();
         File entry = d.openNextFile();
         while (entry) {
             if (!entry.isDirectory() && isJsonFile(entry.name())) {
@@ -393,13 +435,30 @@ std::vector<LXMFMessage> MessageStore::loadConversation(const std::string& peerH
             }
             entry = d.openNextFile();
         }
-        std::sort(filenames.begin(), filenames.end());
+        collectMs += millis() - phaseMs;
+        filesSeen += (int)filenames.size();
 
-        for (const auto& fname : filenames) {
+        phaseMs = millis();
+        std::sort(filenames.begin(), filenames.end());
+        sortMs += millis() - phaseMs;
+
+        size_t startIndex = 0;
+        if (maxMessages > 0 && filenames.size() > maxMessages) {
+            startIndex = filenames.size() - maxMessages;
+        }
+
+        for (size_t i = startIndex; i < filenames.size(); i++) {
+            const auto& fname = filenames[i];
+            unsigned long readStartMs = millis();
             String json = readFileFn(fname);
+            readMs += millis() - readStartMs;
             if (json.length() == 0) continue;
+            bytesRead += json.length();
             JsonDocument doc;
-            if (!deserializeJson(doc, json)) {
+            unsigned long parseStartMs = millis();
+            DeserializationError err = deserializeJson(doc, json);
+            parseMs += millis() - parseStartMs;
+            if (!err) {
                 LXMFMessage msg;
                 std::string srcHex = doc["src"] | "";
                 std::string dstHex = doc["dst"] | "";
@@ -424,6 +483,9 @@ std::vector<LXMFMessage> MessageStore::loadConversation(const std::string& peerH
                     msg.messageId.assignHex(msgIdHex.c_str());
                 }
                 messages.push_back(msg);
+                parsedMessages++;
+            } else {
+                parseFailures++;
             }
         }
     };
@@ -436,7 +498,7 @@ std::vector<LXMFMessage> MessageStore::loadConversation(const std::string& peerH
             loadFromDir(d, [&](const String& fname) {
                 String path = sdDir + "/" + fname;
                 return _sd->readString(path.c_str());
-            });
+            }, "sd");
             loadedFromSD = true;
         }
     }
@@ -453,10 +515,17 @@ std::vector<LXMFMessage> MessageStore::loadConversation(const std::string& peerH
                     if (size > 0 && size < 4096) return f.readString();
                 }
                 return String("");
-            });
+            }, "flash");
         }
     }
 
+    unsigned long elapsed = millis() - startMs;
+    if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_MSG_TRACE_MS) || filesSeen > 64) {
+        Serial.printf("[PERF] MSG loadConversation: peer=%s backend=%s files=%d msgs=%d cap=%u bytes=%u collect=%lums sort=%lums read=%lums parse=%lums parse_fail=%d total=%lums\n",
+                      peerHex.substr(0, 8).c_str(), backend, filesSeen,
+                      parsedMessages, (unsigned)maxMessages, (unsigned)bytesRead,
+                      collectMs, sortMs, readMs, parseMs, parseFailures, elapsed);
+    }
     return messages;
 }
 
@@ -569,11 +638,19 @@ void MessageStore::markConversationRead(const std::string& peerHex) {
     unsigned long startMs = millis();
     int scannedFiles = 0;
     int rewrittenFiles = 0;
-    auto markInDir = [&](auto openFn, auto writeFn, const String& dir) {
+    int sdWrites = 0;
+    int flashWrites = 0;
+    size_t rewrittenBytes = 0;
+    unsigned long collectMs = 0;
+    unsigned long readMs = 0;
+    unsigned long parseMs = 0;
+    unsigned long writeMs = 0;
+    auto markInDir = [&](auto openFn, auto writeFn, const String& dir, const char* backend) {
         // Collect only incoming (_i.json) filenames
         std::vector<String> incomingFiles;
         File d = openFn(dir.c_str());
         if (!d || !d.isDirectory()) return;
+        unsigned long phaseMs = millis();
         File entry = d.openNextFile();
         while (entry) {
             if (!entry.isDirectory() && isJsonFile(entry.name())) {
@@ -585,6 +662,7 @@ void MessageStore::markConversationRead(const std::string& peerHex) {
             }
             entry = d.openNextFile();
         }
+        collectMs += millis() - phaseMs;
 
         // Sort descending (newest first) to stop early at first already-read
         std::sort(incomingFiles.begin(), incomingFiles.end(),
@@ -594,24 +672,34 @@ void MessageStore::markConversationRead(const std::string& peerHex) {
             String path = dir + "/" + fname;
             // Read file via the appropriate storage
             String json;
+            unsigned long readStartMs = millis();
             File f = openFn(path.c_str());
             if (f && !f.isDirectory()) {
                 size_t size = f.size();
                 if (size > 0 && size < 4096) json = f.readString();
                 f.close();
             }
+            readMs += millis() - readStartMs;
             if (json.length() == 0) continue;
             scannedFiles++;
 
             JsonDocument doc;
-            if (deserializeJson(doc, json)) continue;
+            unsigned long parseStartMs = millis();
+            DeserializationError err = deserializeJson(doc, json);
+            parseMs += millis() - parseStartMs;
+            if (err) continue;
             bool isRead = doc["read"] | false;
             if (isRead) break; // all older must be read too
             doc["read"] = true;
             String updated;
             serializeJson(doc, updated);
+            unsigned long writeStartMs = millis();
             writeFn(path.c_str(), updated);
+            writeMs += millis() - writeStartMs;
+            rewrittenBytes += updated.length();
             rewrittenFiles++;
+            if (backend && backend[0] == 's') sdWrites++;
+            else flashWrites++;
         }
     };
 
@@ -619,14 +707,14 @@ void MessageStore::markConversationRead(const std::string& peerHex) {
         String sdDir = sdConversationDir(peerHex);
         markInDir([&](const char* p) { return _sd->openDir(p); },
                   [&](const char* p, const String& d) { _sd->writeString(p, d); return true; },
-                  sdDir);
+                  sdDir, "sd");
     }
 
     if (_flash) {
         String dir = conversationDir(peerHex);
         markInDir([](const char* p) { return LittleFS.open(p); },
                   [&](const char* p, const String& d) { _flash->writeString(p, d); return true; },
-                  dir);
+                  dir, "flash");
     }
 
     bool changed = false;
@@ -640,10 +728,11 @@ void MessageStore::markConversationRead(const std::string& peerHex) {
     }
     if (changed) bumpRevision();
     unsigned long elapsed = millis() - startMs;
-    if (elapsed > 25) {
-        Serial.printf("[PERF] MSG markRead: %s scanned=%d wrote=%d in %lums\n",
+    if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_MSG_TRACE_MS) || rewrittenFiles > 0) {
+        Serial.printf("[PERF] MSG markRead: peer=%s scanned=%d wrote=%d sd_writes=%d flash_writes=%d bytes=%u collect=%lums read=%lums parse=%lums write=%lums total=%lums\n",
                       peerHex.substr(0, 8).c_str(), scannedFiles, rewrittenFiles,
-                      (unsigned long)elapsed);
+                      sdWrites, flashWrites, (unsigned)rewrittenBytes,
+                      collectMs, readMs, parseMs, writeMs, elapsed);
     }
 }
 
@@ -773,7 +862,10 @@ bool MessageStore::updateMessageStatusByCounter(const std::string& peerHex, uint
     return updated;
 }
 
-ConversationSummary MessageStore::buildSummaryForPeer(const std::string& peerHex) const {
+ConversationSummary MessageStore::buildSummaryForPeer(
+    const std::string& peerHex,
+    std::vector<LXMFMessage>* pendingOut,
+    std::vector<std::pair<uint32_t, std::string>>* recentIds) const {
     unsigned long startMs = millis();
     ConversationSummary summary;
 
@@ -810,6 +902,7 @@ ConversationSummary MessageStore::buildSummaryForPeer(const std::string& peerHex
 
     std::sort(files.begin(), files.end());
     String basePath = loadedFromSD ? sdConversationDir(peerHex) : conversationDir(peerHex);
+    const bool collectStartup = pendingOut || recentIds;
 
     auto readJsonFile = [&](const String& path) -> String {
         if (loadedFromSD && _sd && _sd->isReady()) return _sd->readString(path.c_str());
@@ -817,20 +910,7 @@ ConversationSummary MessageStore::buildSummaryForPeer(const std::string& peerHex
         return String("");
     };
 
-    String lastPath = basePath + "/" + files.back();
-    String json = readJsonFile(lastPath);
-    if (json.length() > 0) {
-        JsonDocument doc;
-        if (!deserializeJson(doc, json)) {
-            summary.lastTimestamp = doc["ts"] | 0.0;
-            std::string content = doc["content"] | "";
-            summary.lastIncoming = doc["incoming"] | false;
-            std::string prefix = summary.lastIncoming ? "Them: " : "You: ";
-            if (content.size() > 15) content = content.substr(0, 15) + "...";
-            summary.lastPreview = prefix + content;
-        }
-    }
-
+    bool lastDone = false;
     bool unreadDone = false;
     for (int i = (int)files.size() - 1; i >= 0; i--) {
         const String& fname = files[i];
@@ -838,13 +918,32 @@ ConversationSummary MessageStore::buildSummaryForPeer(const std::string& peerHex
         const bool outgoingFile = hasDirectionSuffix(fname, 'o');
         const bool needIncoming = incomingFile && !unreadDone;
         const bool needOutgoing = outgoingFile;
-        if (!needIncoming && !needOutgoing) continue;
+        const bool needLast = !lastDone;
+        if (!collectStartup && !needLast && !needIncoming && !needOutgoing) continue;
 
         String fjson = readJsonFile(basePath + "/" + fname);
         if (fjson.length() == 0) continue;
 
         JsonDocument fdoc;
         if (deserializeJson(fdoc, fjson)) continue;
+        uint32_t counter = counterFromFilename(fname);
+
+        if (needLast) {
+            summary.lastTimestamp = fdoc["ts"] | 0.0;
+            std::string content = fdoc["content"] | "";
+            summary.lastIncoming = fdoc["incoming"] | false;
+            std::string prefix = summary.lastIncoming ? "Them: " : "You: ";
+            if (content.size() > 15) content = content.substr(0, 15) + "...";
+            summary.lastPreview = prefix + content;
+            lastDone = true;
+        }
+
+        if (recentIds) {
+            std::string msgIdHex = fdoc["msgid"] | "";
+            if (!msgIdHex.empty()) {
+                recentIds->push_back({counter, msgIdHex});
+            }
+        }
 
         if (needIncoming) {
             bool isRead = fdoc["read"] | false;
@@ -857,11 +956,37 @@ ConversationSummary MessageStore::buildSummaryForPeer(const std::string& peerHex
             if (!summary.hasOutgoing) {
                 summary.hasOutgoing = true;
                 summary.lastOutgoingStatus = status;
-                summary.lastOutgoingCounter = counterFromFilename(fname);
+                summary.lastOutgoingCounter = counter;
             }
             if (isPendingStatus(status)) {
                 summary.hasPending = true;
                 if (summary.pendingCount < UINT16_MAX) summary.pendingCount++;
+                if (pendingOut) {
+                    LXMFMessage msg;
+                    std::string srcHex = fdoc["src"] | "";
+                    std::string dstHex = fdoc["dst"] | "";
+                    if (!srcHex.empty()) {
+                        msg.sourceHash = RNS::Bytes();
+                        msg.sourceHash.assignHex(srcHex.c_str());
+                    }
+                    if (!dstHex.empty()) {
+                        msg.destHash = RNS::Bytes();
+                        msg.destHash.assignHex(dstHex.c_str());
+                    }
+                    msg.timestamp = fdoc["ts"] | 0.0;
+                    msg.content = fdoc["content"] | "";
+                    msg.title = fdoc["title"] | "";
+                    msg.incoming = fdoc["incoming"] | false;
+                    msg.status = status;
+                    msg.read = fdoc["read"] | false;
+                    msg.savedCounter = counter;
+                    std::string msgIdHex = fdoc["msgid"] | "";
+                    if (!msgIdHex.empty()) {
+                        msg.messageId = RNS::Bytes();
+                        msg.messageId.assignHex(msgIdHex.c_str());
+                    }
+                    pendingOut->push_back(msg);
+                }
             }
             if (status == LXMFStatus::FAILED) {
                 summary.hasFailed = true;
@@ -915,14 +1040,38 @@ void MessageStore::updateSummaryStatus(const std::string& peerHex, uint32_t coun
 
 void MessageStore::buildSummaries() {
     _summaries.clear();
+    _startupPendingOutgoing.clear();
+    _startupRecentMessageIds.clear();
+    std::vector<std::pair<uint32_t, std::string>> recentIds;
     for (const auto& peerHex : _conversations) {
-        _summaries[peerHex] = buildSummaryForPeer(peerHex);
+        _summaries[peerHex] = buildSummaryForPeer(peerHex, &_startupPendingOutgoing, &recentIds);
         yield();
     }
 
+    std::sort(_startupPendingOutgoing.begin(), _startupPendingOutgoing.end(),
+        [](const LXMFMessage& a, const LXMFMessage& b) {
+            return a.savedCounter < b.savedCounter;
+        });
+
+    std::sort(recentIds.begin(), recentIds.end(),
+        [](const auto& a, const auto& b) { return a.first < b.first; });
+    std::set<std::string> seen;
+    for (const auto& item : recentIds) {
+        if (seen.insert(item.second).second) _startupRecentMessageIds.push_back(item.second);
+    }
+
     Serial.printf("[MSGSTORE] Built summaries for %d conversations\n", (int)_summaries.size());
 }
 
+std::vector<std::string> MessageStore::startupRecentMessageIds(size_t maxIds) const {
+    if (maxIds == 0 || _startupRecentMessageIds.size() <= maxIds) {
+        return _startupRecentMessageIds;
+    }
+    return std::vector<std::string>(
+        _startupRecentMessageIds.end() - maxIds,
+        _startupRecentMessageIds.end());
+}
+
 const ConversationSummary* MessageStore::getSummary(const std::string& peerHex) const {
     auto it = _summaries.find(peerHex);
     return (it != _summaries.end()) ? &it->second : nullptr;
diff --git a/src/storage/MessageStore.h b/src/storage/MessageStore.h
index 676bb21..5097cee 100644
--- a/src/storage/MessageStore.h
+++ b/src/storage/MessageStore.h
@@ -8,6 +8,7 @@
 #include <string>
 #include <map>
 #include <set>
+#include <utility>
 
 struct ConversationSummary {
     double lastTimestamp = 0;
@@ -32,12 +33,15 @@ public:
 
     bool saveMessage(LXMFMessage& msg);
     std::vector<LXMFMessage> loadConversation(const std::string& peerHex) const;
+    std::vector<LXMFMessage> loadConversationTail(const std::string& peerHex, size_t maxMessages) const;
     const std::vector<std::string>& conversations() const { return _conversations; }
     void refreshConversations();
     int messageCount(const std::string& peerHex) const;
     bool deleteConversation(const std::string& peerHex);
     std::vector<LXMFMessage> loadPendingOutgoing() const;
     std::vector<std::string> loadRecentMessageIds(size_t maxIds) const;
+    const std::vector<LXMFMessage>& startupPendingOutgoing() const { return _startupPendingOutgoing; }
+    std::vector<std::string> startupRecentMessageIds(size_t maxIds) const;
     void markConversationRead(const std::string& peerHex);
     bool updateMessageStatus(const std::string& peerHex, double timestamp, bool incoming, LXMFStatus newStatus);
     bool updateMessageStatusByCounter(const std::string& peerHex, uint32_t counter, bool incoming, LXMFStatus newStatus);
@@ -56,7 +60,10 @@ private:
     void initReceiveCounter();
     void buildSummaries();
     void rebuildSummary(const std::string& peerHex);
-    ConversationSummary buildSummaryForPeer(const std::string& peerHex) const;
+    ConversationSummary buildSummaryForPeer(
+        const std::string& peerHex,
+        std::vector<LXMFMessage>* pendingOut = nullptr,
+        std::vector<std::pair<uint32_t, std::string>>* recentIds = nullptr) const;
     void updateSummaryStatus(const std::string& peerHex, uint32_t counter, LXMFStatus oldStatus, LXMFStatus newStatus);
     void bumpRevision();
 
@@ -65,6 +72,8 @@ private:
     bool _externalStorageEnabled = false;
     std::vector<std::string> _conversations;
     std::map<std::string, ConversationSummary> _summaries;
+    std::vector<LXMFMessage> _startupPendingOutgoing;
+    std::vector<std::string> _startupRecentMessageIds;
     uint32_t _nextReceiveCounter = 0;
     uint32_t _revision = 0;
 };
diff --git a/src/storage/SDStore.cpp b/src/storage/SDStore.cpp
index 38af9a8..60ba20d 100644
--- a/src/storage/SDStore.cpp
+++ b/src/storage/SDStore.cpp
@@ -1,5 +1,6 @@
 #include "SDStore.h"
 #include "config/Config.h"
+#include "util/PerfTrace.h"
 
 bool SDStore::begin(SPIClass* spi, int csPin) {
     if (!spi) return false;
@@ -93,7 +94,11 @@ bool SDStore::readFile(const char* path, uint8_t* buffer, size_t maxLen, size_t&
 }
 
 bool SDStore::writeAtomic(const char* path, const uint8_t* data, size_t len) {
-    if (!_ready) return false;
+    unsigned long startMs = PerfTrace::nowMs();
+    if (!_ready) {
+        PerfTrace::write("sd", "atomic", path, len, startMs, false);
+        return false;
+    }
 
     String tmpPath = String(path) + ".tmp";
     String bakPath = String(path) + ".bak";
@@ -101,6 +106,7 @@ bool SDStore::writeAtomic(const char* path, const uint8_t* data, size_t len) {
     File f = SD.open(tmpPath.c_str(), FILE_WRITE);
     if (!f) {
         Serial.printf("[SD] writeAtomic: failed to open tmp %s\n", tmpPath.c_str());
+        PerfTrace::write("sd", "atomic", path, len, startMs, false);
         return false;
     }
     size_t written = f.write(data, len);
@@ -108,6 +114,7 @@ bool SDStore::writeAtomic(const char* path, const uint8_t* data, size_t len) {
     if (written != len) {
         Serial.printf("[SD] writeAtomic: write incomplete (%d/%d)\n", (int)written, (int)len);
         SD.remove(tmpPath.c_str());
+        PerfTrace::write("sd", "atomic", path, len, startMs, false);
         return false;
     }
 
@@ -116,6 +123,7 @@ bool SDStore::writeAtomic(const char* path, const uint8_t* data, size_t len) {
         Serial.println("[SD] writeAtomic: verify failed");
         if (verify) verify.close();
         SD.remove(tmpPath.c_str());
+        PerfTrace::write("sd", "atomic", path, len, startMs, false);
         return false;
     }
     verify.close();
@@ -132,26 +140,35 @@ bool SDStore::writeAtomic(const char* path, const uint8_t* data, size_t len) {
     if (!SD.rename(tmpPath.c_str(), path)) {
         Serial.printf("[SD] writeAtomic: rename failed %s -> %s\n", tmpPath.c_str(), path);
         if (SD.exists(bakPath.c_str())) { SD.rename(bakPath.c_str(), path); }
+        PerfTrace::write("sd", "atomic", path, len, startMs, false);
         return false;
     }
     SD.remove(bakPath.c_str());
+    PerfTrace::write("sd", "atomic", path, len, startMs, true);
     return true;
 }
 
 bool SDStore::writeSimple(const char* path, const uint8_t* data, size_t len) {
-    if (!_ready) return false;
+    unsigned long startMs = PerfTrace::nowMs();
+    if (!_ready) {
+        PerfTrace::write("sd", "simple", path, len, startMs, false);
+        return false;
+    }
 
     File f = SD.open(path, FILE_WRITE);
     if (!f) {
         Serial.printf("[SD] writeSimple: failed to open %s\n", path);
+        PerfTrace::write("sd", "simple", path, len, startMs, false);
         return false;
     }
     size_t written = f.write(data, len);
     f.close();
     if (written != len) {
         Serial.printf("[SD] writeSimple: write incomplete (%d/%d)\n", (int)written, (int)len);
+        PerfTrace::write("sd", "simple", path, len, startMs, false);
         return false;
     }
+    PerfTrace::write("sd", "simple", path, len, startMs, true);
     return true;
 }
 
diff --git a/src/ui/UIManager.cpp b/src/ui/UIManager.cpp
index 163aba3..978a707 100644
--- a/src/ui/UIManager.cpp
+++ b/src/ui/UIManager.cpp
@@ -2,6 +2,7 @@
 #include "Theme.h"
 #include "LvTheme.h"
 #include "LvInput.h"
+#include "util/PerfTrace.h"
 
 // --- LvScreen base ---
 
@@ -42,27 +43,55 @@ void UIManager::begin() {
 
 void UIManager::setScreen(LvScreen* screen) {
     if (_currentLvScreen == screen) return;
+    const char* fromTitle = _currentLvScreen ? _currentLvScreen->title() : "none";
+    const char* toTitle = screen ? screen->title() : "none";
+    unsigned long startMs = PerfTrace::nowMs();
+    unsigned long exitMs = 0;
+    unsigned long destroyMs = 0;
+    unsigned long showMs = 0;
+    unsigned long cleanMs = 0;
+    unsigned long createMs = 0;
+    unsigned long enterMs = 0;
 
     // Transition from previous LVGL screen
     if (_currentLvScreen) {
+        unsigned long phaseMs = millis();
         _currentLvScreen->onExit();
+        exitMs = millis() - phaseMs;
+        phaseMs = millis();
         _currentLvScreen->destroyUI();
+        destroyMs = millis() - phaseMs;
     }
 
     _currentLvScreen = screen;
 
     // Show LVGL layers
+    unsigned long phaseMs = millis();
     if (!_bootMode) {
         lv_obj_clear_flag(_lvStatusBar.obj(), LV_OBJ_FLAG_HIDDEN);
         lv_obj_clear_flag(_lvTabBar.obj(), LV_OBJ_FLAG_HIDDEN);
     }
     lv_obj_clear_flag(_lvContent, LV_OBJ_FLAG_HIDDEN);
+    showMs = millis() - phaseMs;
 
     if (_currentLvScreen) {
         // Clean content area
+        phaseMs = millis();
         lv_obj_clean(_lvContent);
+        cleanMs = millis() - phaseMs;
+        phaseMs = millis();
         _currentLvScreen->createUI(_lvContent);
+        createMs = millis() - phaseMs;
+        phaseMs = millis();
         _currentLvScreen->onEnter();
+        enterMs = millis() - phaseMs;
+    }
+
+    unsigned long elapsed = millis() - startMs;
+    if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_UI_TRACE_MS)) {
+        Serial.printf("[PERF] UI transition: %s -> %s total=%lums exit=%lums destroy=%lums show=%lums clean=%lums create=%lums enter=%lums boot=%s\n",
+                      fromTitle, toTitle, elapsed, exitMs, destroyMs, showMs,
+                      cleanMs, createMs, enterMs, _bootMode ? "yes" : "no");
     }
 }
 
diff --git a/src/ui/screens/LvMessageView.cpp b/src/ui/screens/LvMessageView.cpp
index 622d393..89becc4 100644
--- a/src/ui/screens/LvMessageView.cpp
+++ b/src/ui/screens/LvMessageView.cpp
@@ -4,6 +4,7 @@
 #include "ui/LvTabBar.h"
 #include "reticulum/LXMFManager.h"
 #include "reticulum/AnnounceManager.h"
+#include "util/PerfTrace.h"
 #include <Arduino.h>
 #include <time.h>
 #include <cmath>
@@ -109,6 +110,22 @@ void LvMessageView::updateHeader() {
     }
 }
 
+void LvMessageView::markVisibleConversationRead() {
+    if (!_markReadPending || !_lxmf) return;
+
+    unsigned long startMs = PerfTrace::nowMs();
+    _lxmf->markRead(_peerHex);
+    _markReadPending = false;
+    if (_ui) {
+        _ui->lvTabBar().setUnreadCount(LvTabBar::TAB_MSGS, _lxmf->unreadCount());
+    }
+    unsigned long elapsed = PerfTrace::elapsedMs(startMs);
+    if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_MSG_TRACE_MS)) {
+        Serial.printf("[PERF] Chat markRead: peer=%s total=%lums\n",
+                      _peerHex.substr(0, 8).c_str(), elapsed);
+    }
+}
+
 void LvMessageView::updateComposerState() {
     if (!_btnSend) return;
     bool hasText = !_inputText.empty();
@@ -298,12 +315,18 @@ void LvMessageView::destroyUI() {
 }
 
 void LvMessageView::onEnter() {
+    unsigned long startMs = PerfTrace::nowMs();
+    unsigned long markReadMs = 0;
+    unsigned long tabBadgeMs = 0;
+    unsigned long callbackMs = 0;
+    unsigned long resetMs = 0;
+    unsigned long headerMs = 0;
+    unsigned long rebuildMs = 0;
     if (_lxmf) {
-        _lxmf->markRead(_peerHex);
-        // Update unread badge on Messages tab
-        if (_ui) _ui->lvTabBar().setUnreadCount(LvTabBar::TAB_MSGS, _lxmf->unreadCount());
+        _markReadPending = true;
         // Register status callback - partial update without full rebuild
         std::string peer = _peerHex;
+        unsigned long phaseMs = millis();
         _lxmf->setStatusCallback([this, peer](const std::string& peerHex, double ts, uint32_t savedCounter, LXMFStatus newStatus) {
             if (peerHex != peer) return;
             for (int i = (int)_cachedMsgs.size() - 1; i >= 0; i--) {
@@ -317,8 +340,11 @@ void LvMessageView::onEnter() {
                 }
             }
         });
+        callbackMs = millis() - phaseMs;
     }
+    unsigned long phaseMs = millis();
     _lastMsgCount = -1;
+    _knownTotalCount = -1;
     _lastRefreshMs = 0;
     _inputText.clear();
     hideSendModeMenu();
@@ -326,14 +352,27 @@ void LvMessageView::onEnter() {
     if (_textarea) {
         updateComposerText();
     }
+    resetMs = millis() - phaseMs;
+    phaseMs = millis();
     updateHeader();
     updateComposerState();
+    headerMs = millis() - phaseMs;
     _cachedMsgs.clear();  // Force fresh load
+    phaseMs = millis();
     rebuildMessages();
+    rebuildMs = millis() - phaseMs;
+
+    unsigned long elapsed = millis() - startMs;
+    if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_UI_TRACE_MS)) {
+        Serial.printf("[PERF] Chat onEnter: peer=%s msgs=%d total=%lums markRead=%lums tab=%lums callback=%lums reset=%lums header=%lums rebuild=%lums\n",
+                      _peerHex.substr(0, 8).c_str(), (int)_cachedMsgs.size(), elapsed,
+                      markReadMs, tabBadgeMs, callbackMs, resetMs, headerMs, rebuildMs);
+    }
 }
 
 void LvMessageView::onExit() {
     if (_lxmf) _lxmf->setStatusCallback(nullptr);
+    _markReadPending = false;
     hideSendModeMenu();
     _inputText.clear();
     _cachedMsgs.clear();
@@ -345,17 +384,23 @@ void LvMessageView::onExit() {
 void LvMessageView::refreshUI() {
     if (!_lxmf) return;
     unsigned long now = millis();
+    markVisibleConversationRead();
     if (now - _lastRefreshMs < REFRESH_INTERVAL_MS) return;
     _lastRefreshMs = now;
     updateHeader();
 
     // Only reload from disk when message count changes (new messages arrive)
     auto* summary = _lxmf->getConversationSummary(_peerHex);
-    if (summary && summary->totalCount == (int)_cachedMsgs.size()) return;
+    int totalCount = summary ? summary->totalCount : -1;
+    if (summary && totalCount == _knownTotalCount) return;
 
-    auto newMsgs = _lxmf->getMessages(_peerHex);
-    if (newMsgs.size() != _cachedMsgs.size()) {
-        if (newMsgs.size() > _cachedMsgs.size()) {
+    auto newMsgs = _lxmf->getRecentMessages(_peerHex, CHAT_VIEW_MAX_MESSAGES);
+    int newKnownTotal = summary ? totalCount : (int)newMsgs.size();
+    if (newKnownTotal != _knownTotalCount || newMsgs.size() != _cachedMsgs.size()) {
+        bool canAppend = !_cachedMsgs.empty() &&
+            _cachedMsgs.size() < CHAT_VIEW_MAX_MESSAGES &&
+            newMsgs.size() > _cachedMsgs.size();
+        if (canAppend) {
             // Incremental append - only create widgets for new messages
             size_t oldCount = _cachedMsgs.size();
             _cachedMsgs = std::move(newMsgs);
@@ -370,14 +415,15 @@ void LvMessageView::refreshUI() {
             }
             lv_obj_scroll_to_y(_msgScroll, LV_COORD_MAX, LV_ANIM_OFF);
         } else {
-            // Count decreased (deletion?) - full rebuild
+            // Tail window shifted, count decreased, or cache was empty - full visible-window rebuild.
             _cachedMsgs = std::move(newMsgs);
             _lastMsgCount = (int)_cachedMsgs.size();
             rebuildMessages();
         }
+        _knownTotalCount = newKnownTotal;
         // Mark as read since user is actively viewing this conversation
-        _lxmf->markRead(_peerHex);
-        if (_ui) _ui->lvTabBar().setUnreadCount(LvTabBar::TAB_MSGS, _lxmf->unreadCount());
+        _markReadPending = true;
+        markVisibleConversationRead();
     }
 }
 
@@ -486,19 +532,35 @@ void LvMessageView::appendMessage(const LXMFMessage& msg) {
 void LvMessageView::rebuildMessages() {
     if (!_lxmf || !_msgScroll) return;
     unsigned long startMs = millis();
+    unsigned long loadMs = 0;
+    unsigned long cleanMs = 0;
+    unsigned long emptyMs = 0;
+    unsigned long appendMs = 0;
+    unsigned long scrollMs = 0;
+    bool loadedFromStore = false;
 
     // Only load from disk if _cachedMsgs is empty (first call or after send)
     if (_cachedMsgs.empty()) {
-        _cachedMsgs = _lxmf->getMessages(_peerHex);
+        unsigned long phaseMs = millis();
+        _cachedMsgs = _lxmf->getRecentMessages(_peerHex, CHAT_VIEW_MAX_MESSAGES);
+        loadMs = millis() - phaseMs;
+        loadedFromStore = true;
+    }
+    if (_lxmf) {
+        auto* summary = _lxmf->getConversationSummary(_peerHex);
+        _knownTotalCount = summary ? summary->totalCount : (int)_cachedMsgs.size();
     }
     _lastMsgCount = (int)_cachedMsgs.size();
     _lastRefreshMs = millis();
+    unsigned long phaseMs = millis();
     lv_obj_clean(_msgScroll);
+    cleanMs = millis() - phaseMs;
     _statusLabels.clear();
     _textLabels.clear();
     _bubbleBoxes.clear();
 
     if (_cachedMsgs.empty()) {
+        phaseMs = millis();
         lv_obj_set_layout(_msgScroll, 0);
 
         lv_obj_t* empty = lv_obj_create(_msgScroll);
@@ -529,23 +591,36 @@ void LvMessageView::rebuildMessages() {
         lv_obj_set_style_text_color(sub, lv_color_hex(Theme::TEXT_SECONDARY), 0);
         lv_label_set_text(sub, "Thread is quiet");
         lv_obj_align(sub, LV_ALIGN_TOP_MID, 0, 54);
+        emptyMs = millis() - phaseMs;
+        unsigned long elapsed = millis() - startMs;
+        if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_UI_TRACE_MS)) {
+            Serial.printf("[PERF] Chat rebuild: peer=%s msgs=0 loaded=%s load=%lums clean=%lums empty=%lums append=0ms scroll=0ms total=%lums\n",
+                          _peerHex.substr(0, 8).c_str(), loadedFromStore ? "yes" : "no",
+                          loadMs, cleanMs, emptyMs, elapsed);
+        }
         return;
     }
 
+    phaseMs = millis();
     lv_obj_set_layout(_msgScroll, LV_LAYOUT_FLEX);
     lv_obj_set_flex_flow(_msgScroll, LV_FLEX_FLOW_COLUMN);
 
     for (const auto& msg : _cachedMsgs) {
         appendMessage(msg);
     }
+    appendMs = millis() - phaseMs;
 
     // Auto-scroll to bottom
+    phaseMs = millis();
     lv_obj_scroll_to_y(_msgScroll, LV_COORD_MAX, LV_ANIM_OFF);
+    scrollMs = millis() - phaseMs;
 
     unsigned long elapsed = millis() - startMs;
-    if (elapsed > 40) {
-        Serial.printf("[PERF] Chat rebuild: %d msgs in %lums\n",
-                      (int)_cachedMsgs.size(), (unsigned long)elapsed);
+    if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_UI_TRACE_MS)) {
+        Serial.printf("[PERF] Chat rebuild: peer=%s msgs=%d loaded=%s load=%lums clean=%lums empty=0ms append=%lums scroll=%lums total=%lums\n",
+                      _peerHex.substr(0, 8).c_str(), (int)_cachedMsgs.size(),
+                      loadedFromStore ? "yes" : "no", loadMs, cleanMs,
+                      appendMs, scrollMs, elapsed);
     }
 }
 
@@ -608,27 +683,56 @@ void LvMessageView::applyStatusGlyph(lv_obj_t* lbl, LXMFStatus status) {
 }
 
 void LvMessageView::sendCurrentMessage(bool viaLink) {
+    unsigned long startMs = PerfTrace::nowMs();
     if (!_lxmf || _peerHex.empty() || _inputText.empty()) return;
+    size_t inputBytes = _inputText.size();
     if (_inputText.size() > MAX_COMPOSER_CHARS) {
+#if RSDECK_PERF_TRACE
+        Serial.printf("[PERF] Chat send: peer=%s bytes=%u link=%s queued=no reason=too_long total=%lums\n",
+                      _peerHex.substr(0, 8).c_str(), (unsigned)inputBytes,
+                      viaLink ? "yes" : "no", PerfTrace::elapsedMs(startMs));
+#endif
         if (_ui) _ui->lvStatusBar().showToast("Message too long", 1500);
         return;
     }
 
+    unsigned long hashMs = millis();
     RNS::Bytes destHash;
     destHash.assignHex(_peerHex.c_str());
+    hashMs = millis() - hashMs;
+    unsigned long queueStartMs = millis();
     bool queued = viaLink
         ? _lxmf->sendMessageViaLink(destHash, _inputText.c_str())
         : _lxmf->sendMessage(destHash, _inputText.c_str());
+    unsigned long queueMs = millis() - queueStartMs;
     if (!queued) {
+#if RSDECK_PERF_TRACE
+        Serial.printf("[PERF] Chat send: peer=%s bytes=%u link=%s queued=no reason=queue_full hash=%lums queue=%lums total=%lums\n",
+                      _peerHex.substr(0, 8).c_str(), (unsigned)inputBytes,
+                      viaLink ? "yes" : "no", hashMs, queueMs,
+                      PerfTrace::elapsedMs(startMs));
+#endif
         if (_ui) _ui->lvStatusBar().showToast("Message queue full", 1500);
         return;
     }
     if (viaLink && _ui) _ui->lvStatusBar().showToast("Link send queued", 1200);
 
+    unsigned long composerStartMs = millis();
     _inputText.clear();
     updateComposerState();
     _cachedMsgs.clear();  // Force fresh load in rebuildMessages
+    _knownTotalCount = -1;
+    unsigned long composerMs = millis() - composerStartMs;
+    unsigned long rebuildStartMs = millis();
     rebuildMessages();
+    unsigned long rebuildMs = millis() - rebuildStartMs;
+    unsigned long elapsed = millis() - startMs;
+#if RSDECK_PERF_TRACE
+    Serial.printf("[PERF] Chat send: peer=%s bytes=%u link=%s queued=yes hash=%lums queue=%lums composer=%lums rebuild=%lums msgs=%d total=%lums\n",
+                  _peerHex.substr(0, 8).c_str(), (unsigned)inputBytes,
+                  viaLink ? "yes" : "no", hashMs, queueMs, composerMs,
+                  rebuildMs, (int)_cachedMsgs.size(), elapsed);
+#endif
 }
 
 bool LvMessageView::handleKey(const KeyEvent& event) {
@@ -697,8 +801,14 @@ bool LvMessageView::handleKey(const KeyEvent& event) {
             if (_ui) _ui->lvStatusBar().showToast("Message too long", 900);
             return true;
         }
+        unsigned long inputStartMs = millis();
         _inputText += (char)event.character;
         updateComposerState();
+        unsigned long elapsed = millis() - inputStartMs;
+        if (PerfTrace::shouldLog(elapsed, RSDECK_PERF_UI_TRACE_MS)) {
+            Serial.printf("[PERF] Chat input: peer=%s chars=%u total=%lums\n",
+                          _peerHex.substr(0, 8).c_str(), (unsigned)_inputText.size(), elapsed);
+        }
         return true;
     }
 
diff --git a/src/ui/screens/LvMessageView.h b/src/ui/screens/LvMessageView.h
index dcbc574..1abb029 100644
--- a/src/ui/screens/LvMessageView.h
+++ b/src/ui/screens/LvMessageView.h
@@ -35,6 +35,7 @@ private:
     void appendMessage(const LXMFMessage& msg);
     std::string getPeerName();
     void updateHeader();
+    void markVisibleConversationRead();
     void updateComposerState();
     void refreshComposerPlaceholder();
     void updateComposerText();
@@ -50,8 +51,10 @@ private:
     std::string _peerHex;
     std::string _inputText;
     int _lastMsgCount = -1;
+    int _knownTotalCount = -1;
     unsigned long _lastRefreshMs = 0;
     std::vector<LXMFMessage> _cachedMsgs;
+    bool _markReadPending = false;
 
     void updateMessageStatus(int msgIdx, LXMFStatus status);
     static void applyStatusGlyph(lv_obj_t* lbl, LXMFStatus status);
@@ -77,5 +80,6 @@ private:
     std::vector<lv_obj_t*> _bubbleBoxes;
 
     static constexpr unsigned long REFRESH_INTERVAL_MS = 2000;  // Check for new messages every 2s
+    static constexpr size_t CHAT_VIEW_MAX_MESSAGES = 40;
     static constexpr size_t MAX_COMPOSER_CHARS = 120;
 };
diff --git a/src/ui/screens/LvSettingsScreen.cpp b/src/ui/screens/LvSettingsScreen.cpp
index 7954998..17bbe1e 100644
--- a/src/ui/screens/LvSettingsScreen.cpp
+++ b/src/ui/screens/LvSettingsScreen.cpp
@@ -179,8 +179,8 @@ bool LvSettingsScreen::settingNeedsReboot(const SettingItem& item) const {
     if (labelEq(item.label, "WiFi Profile")) return s.wifiSTASelected != _rebootSnap.wifiSTASelected;
     if (isWiFiSSIDLabel(item.label) || isWiFiPasswordLabel(item.label)) return interfaceSettingsChanged();
     if (labelEq(item.label, "Scan Networks") || labelEq(item.label, "Forget Network")) return interfaceSettingsChanged();
-    if (labelEq(item.label, "TCP Relay") || labelEq(item.label, "Relay Host") ||
-        labelEq(item.label, "Relay Port")) return tcpSettingsChanged();
+    if (labelEq(item.label, "TCP Server") || labelEq(item.label, "Host") ||
+        labelEq(item.label, "Port")) return tcpSettingsChanged();
     if (labelEq(item.label, "LAN Discovery")) return s.autoIfaceEnabled != _rebootSnap.autoIfaceEnabled;
     if (labelEq(item.label, "SD Message Store")) return storageSettingsChanged();
     return false;
@@ -807,7 +807,7 @@ void LvSettingsScreen::buildItems() {
     }
     {
         SettingItem tcpPreset;
-        tcpPreset.label = "TCP Relay";
+        tcpPreset.label = "TCP Server";
         tcpPreset.type = SettingType::ENUM_CHOICE;
         tcpPreset.getter = [&s]() {
             for (auto& ep : s.tcpConnections) {
@@ -837,7 +837,7 @@ void LvSettingsScreen::buildItems() {
     }
     {
         SettingItem tcpHost;
-        tcpHost.label = "Relay Host";
+        tcpHost.label = "Host";
         tcpHost.type = SettingType::TEXT_INPUT;
         tcpHost.textGetter = [&s]() { return s.tcpConnections.empty() ? String("") : s.tcpConnections[0].host; };
         tcpHost.textSetter = [&s](const String& v) {
@@ -850,7 +850,7 @@ void LvSettingsScreen::buildItems() {
         _items.push_back(tcpHost);
         idx++;
     }
-    _items.push_back({"Relay Port", SettingType::INTEGER,
+    _items.push_back({"Port", SettingType::INTEGER,
         [&s]() { return s.tcpConnections.empty() ? TCP_DEFAULT_PORT : (int)s.tcpConnections[0].port; },
         [&s](int v) {
             if (s.tcpConnections.empty()) {
@@ -2138,7 +2138,7 @@ void LvSettingsScreen::applyAndSave() {
     else if (_sd && _flash) { saved = _cfg->save(*_sd, *_flash); }
     else if (_flash) { saved = _cfg->save(*_flash); }
 
-    // TCP relay changes are persisted only. Recreating clients live can race
+    // TCP server changes are persisted only. Recreating clients live can race
     // in-flight sockets/netif teardown on ESP32; reboot applies them cleanly.
 
     // Apply GPS toggle live (start/stop GPS UART)
@@ -2156,12 +2156,12 @@ void LvSettingsScreen::applyAndSave() {
             _ui->lvStatusBar().showToast("Save failed", 2000);
         } else if (_rebootNeeded && !wasRebootNeeded) {
             _ui->lvStatusBar().showToast(
-                tcpChanged ? "TCP relay saved; reboot to apply" : "Interface changes saved; reboot to apply",
+                tcpChanged ? "TCP server saved; reboot to apply" : "Interface changes saved; reboot to apply",
                 3000);
         } else if (!_rebootNeeded && wasRebootNeeded) {
             _ui->lvStatusBar().showToast("Pending reboot cleared", 1500);
         } else if (tcpChanged) {
-            _ui->lvStatusBar().showToast("TCP relay saved; reboot to apply", 3000);
+            _ui->lvStatusBar().showToast("TCP server saved; reboot to apply", 3000);
         } else {
             _ui->lvStatusBar().showToast("Saved", 800);
         }
diff --git a/src/util/PerfTrace.h b/src/util/PerfTrace.h
new file mode 100644
index 0000000..25aa2f9
--- /dev/null
+++ b/src/util/PerfTrace.h
@@ -0,0 +1,70 @@
+#pragma once
+
+#include <Arduino.h>
+
+#ifndef RSDECK_PERF_TRACE
+#define RSDECK_PERF_TRACE 1
+#endif
+
+#ifndef RSDECK_PERF_WRITE_TRACE_MS
+#define RSDECK_PERF_WRITE_TRACE_MS 20UL
+#endif
+
+#ifndef RSDECK_PERF_MSG_TRACE_MS
+#define RSDECK_PERF_MSG_TRACE_MS 25UL
+#endif
+
+#ifndef RSDECK_PERF_UI_TRACE_MS
+#define RSDECK_PERF_UI_TRACE_MS 16UL
+#endif
+
+#ifndef RSDECK_PERF_PERSIST_TRACE_MS
+#define RSDECK_PERF_PERSIST_TRACE_MS 25UL
+#endif
+
+namespace PerfTrace {
+
+inline unsigned long nowMs() {
+    return millis();
+}
+
+inline unsigned long elapsedMs(unsigned long startMs) {
+    return millis() - startMs;
+}
+
+inline bool shouldLog(unsigned long durationMs, unsigned long thresholdMs) {
+#if RSDECK_PERF_TRACE
+    return durationMs >= thresholdMs;
+#else
+    (void)durationMs;
+    (void)thresholdMs;
+    return false;
+#endif
+}
+
+inline void write(const char* backend, const char* op, const char* path,
+                  size_t bytes, unsigned long startMs, bool ok,
+                  unsigned long thresholdMs = RSDECK_PERF_WRITE_TRACE_MS) {
+#if RSDECK_PERF_TRACE
+    const unsigned long durationMs = elapsedMs(startMs);
+    if (!ok || durationMs >= thresholdMs) {
+        Serial.printf("[PERF] WRITE backend=%s op=%s path=%s bytes=%u ok=%s in %lums\n",
+                      backend ? backend : "?",
+                      op ? op : "?",
+                      path ? path : "?",
+                      (unsigned)bytes,
+                      ok ? "yes" : "no",
+                      durationMs);
+    }
+#else
+    (void)backend;
+    (void)op;
+    (void)path;
+    (void)bytes;
+    (void)startMs;
+    (void)ok;
+    (void)thresholdMs;
+#endif
+}
+
+}  // namespace PerfTrace