fix: address PR #296 code review feedback

- Clamp ML predictions between physics floor (raw airtime) and ceiling
  (worst-case formula) so model can never produce unsafe timeouts
- Replace hourOfDay feature with secondsSinceLastRx for network activity
- Remove unused _ContactStats.stdDev and dead model persistence code
- Debounce observation writes (2s) instead of writing on every delivery
- Skip recording observations when pathLength is null to avoid corrupting
  training data
- Add comment explaining global (not per-contact) RX time tracking
- Remove notifyListeners from retrain to avoid unnecessary widget rebuilds
- Run dart format

lib/connector/meshcore_connector.dart

@@ -168,6 +168,8 @@ class MeshCoreConnector extends ChangeNotifier {
   bool _isLoadingChannels = false;
   bool _hasLoadedChannels = false;
   TimeoutPredictionService? _timeoutPredictionService;
+  // Intentionally global (not per-contact): tracks overall network activity.
+  // Frequent RX from any source indicates a busy network with more collisions.
   DateTime _lastRxTime = DateTime.now();
   bool _batteryRequested = false;
   bool _awaitingSelfInfo = false;
@@ -694,14 +696,19 @@ class MeshCoreConnector extends ChangeNotifier {
       updateMessageCallback: _updateMessage,
       clearContactPathCallback: clearContactPath,
       setContactPathCallback: setContactPath,
-      calculateTimeoutCallback: (pathLength, messageBytes, {String? contactKey}) =>
-          calculateTimeout(pathLength: pathLength, messageBytes: messageBytes, contactKey: contactKey),
+      calculateTimeoutCallback:
+          (pathLength, messageBytes, {String? contactKey}) => calculateTimeout(
+            pathLength: pathLength,
+            messageBytes: messageBytes,
+            contactKey: contactKey,
+          ),
       getSelfPublicKeyCallback: () => _selfPublicKey,
       prepareContactOutboundTextCallback: prepareContactOutboundText,
       appSettingsService: appSettingsService,
       debugLogService: _appDebugLogService,
       recordPathResultCallback: _recordPathResult,
-      onDeliveryObservedCallback: (contactKey, pathLength, messageBytes, tripTimeMs) {
+      onDeliveryObservedCallback:
+          (contactKey, pathLength, messageBytes, tripTimeMs) {
             final secSinceRx = DateTime.now().difference(_lastRxTime).inSeconds;
             _timeoutPredictionService?.recordObservation(
               contactKey: contactKey,
@@ -2890,14 +2897,54 @@ class MeshCoreConnector extends ChangeNotifier {
     }
   }
 
-  /// Calculate timeout for a message based on radio settings and path length
-  /// Returns timeout in milliseconds, considering number of hops
+  /// Estimate single-packet airtime in ms from radio settings, or a fallback.
+  int _estimateAirtimeMs(int messageBytes) {
+    if (_currentFreqHz != null &&
+        _currentBwHz != null &&
+        _currentSf != null &&
+        _currentCr != null) {
+      final cr = _currentCr! <= 4 ? _currentCr! : _currentCr! - 4;
+      return calculateLoRaAirtime(
+        payloadBytes: messageBytes,
+        spreadingFactor: _currentSf!,
+        bandwidthHz: _currentBwHz!,
+        codingRate: cr,
+        lowDataRateOptimize: _currentSf! >= 11,
+      );
+    }
+    return 50; // fallback: ~SF7/BW125 for 100 bytes
+  }
+
+  /// Physics-based worst-case timeout (ceiling).
+  int _physicsMaxTimeout(int pathLength, int airtime) {
+    if (pathLength < 0) {
+      return 500 + (16 * airtime);
+    } else {
+      return 500 + ((airtime * 6 + 250) * (pathLength + 1));
+    }
+  }
+
+  /// Physics-based minimum timeout (floor): raw traversal time.
+  int _physicsMinTimeout(int pathLength, int airtime) {
+    if (pathLength < 0) {
+      return airtime;
+    } else {
+      return airtime * (pathLength + 1);
+    }
+  }
+
+  /// Calculate timeout for a message based on radio settings and path length.
+  /// Returns timeout in milliseconds, considering number of hops.
   int calculateTimeout({
     required int pathLength,
     int messageBytes = 100,
     String? contactKey,
   }) {
-    // Try ML-based prediction first
+    final airtime = _estimateAirtimeMs(messageBytes);
+    final physicsMin = _physicsMinTimeout(pathLength, airtime);
+    final physicsMax = _physicsMaxTimeout(pathLength, airtime);
+
+    // Try ML-based prediction, clamped between physics bounds
     final secSinceRx = DateTime.now().difference(_lastRxTime).inSeconds;
     final mlTimeout = _timeoutPredictionService?.predictTimeout(
       contactKey: contactKey,
@@ -2905,35 +2952,11 @@ class MeshCoreConnector extends ChangeNotifier {
       messageBytes: messageBytes,
       secondsSinceLastRx: secSinceRx,
     );
-    if (mlTimeout != null) return mlTimeout;
-
-    // If we have radio settings, use them for accurate calculation
-    if (_currentFreqHz != null &&
-        _currentBwHz != null &&
-        _currentSf != null &&
-        _currentCr != null) {
-      final cr = _currentCr! <= 4 ? _currentCr! : _currentCr! - 4;
-      return calculateMessageTimeout(
-        freqHz: _currentFreqHz!,
-        bwHz: _currentBwHz!,
-        sf: _currentSf!,
-        cr: cr,
-        pathLength: pathLength,
-        messageBytes: messageBytes,
-      );
+    if (mlTimeout != null) {
+      return mlTimeout.clamp(physicsMin, physicsMax);
     }
 
-    // Fallback: Conservative estimates based on typical settings
-    // Assume SF7, BW125, which gives ~50ms airtime for 100 bytes
-    const estimatedAirtime = 50;
-
-    if (pathLength < 0) {
-      // Flood mode: Base delay + 16× airtime
-      return 500 + (16 * estimatedAirtime);
-    } else {
-      // Direct path: Base delay + ((airtime×6 + 250ms)×(hops+1))
-      return 500 + ((estimatedAirtime * 6 + 250) * (pathLength + 1));
-    }
+    return physicsMax;
   }
 
   void _handleContact(Uint8List frame, {bool isContact = true}) {

lib/services/message_retry_service.dart

@@ -74,13 +74,19 @@ class MessageRetryService extends ChangeNotifier {
     required Function(Message) updateMessageCallback,
     Function(Contact)? clearContactPathCallback,
     Function(Contact, Uint8List, int)? setContactPathCallback,
-    Function(int pathLength, int messageBytes, {String? contactKey})? calculateTimeoutCallback,
+    Function(int pathLength, int messageBytes, {String? contactKey})?
+    calculateTimeoutCallback,
     Uint8List? Function()? getSelfPublicKeyCallback,
     String Function(Contact, String)? prepareContactOutboundTextCallback,
     AppSettingsService? appSettingsService,
     AppDebugLogService? debugLogService,
     Function(String, PathSelection, bool, int?)? recordPathResultCallback,
-    Function(String contactKey, int pathLength, int messageBytes, int tripTimeMs)?
+    Function(
+      String contactKey,
+      int pathLength,
+      int messageBytes,
+      int tripTimeMs,
+    )?
     onDeliveryObservedCallback,
   }) {
     _sendMessageCallback = sendMessageCallback;
@@ -750,10 +756,12 @@ class MessageRetryService extends ChangeNotifier {
           true,
           tripTimeMs,
         );
-        if (_onDeliveryObservedCallback != null && tripTimeMs > 0) {
+        if (_onDeliveryObservedCallback != null &&
+            tripTimeMs > 0 &&
+            message.pathLength != null) {
           _onDeliveryObservedCallback!(
             contact.publicKeyHex,
-            message.pathLength ?? 0,
+            message.pathLength!,
             message.text.length,
             tripTimeMs,
           );

lib/services/storage_service.dart

@@ -8,7 +8,6 @@ class StorageService {
   static const String _pendingMessagesKey = 'pending_messages';
   static const String _repeaterPasswordsKey = 'repeater_passwords';
   static const String _deliveryObservationsKey = 'delivery_observations';
-  static const String _timeoutModelKey = 'timeout_ml_model';
 
   Future<void> savePathHistory(
     String contactPubKeyHex,
@@ -143,10 +142,7 @@ class StorageService {
     try {
       final list = jsonDecode(jsonStr) as List;
       return list
-          .map(
-            (e) =>
-                DeliveryObservation.fromJson(e as Map<String, dynamic>),
-          )
+          .map((e) => DeliveryObservation.fromJson(e as Map<String, dynamic>))
           .toList();
     } catch (e) {
       return [];
@@ -157,19 +153,4 @@ class StorageService {
     final prefs = PrefsManager.instance;
     await prefs.remove(_deliveryObservationsKey);
   }
-
-  Future<void> saveTimeoutModel(String modelJson) async {
-    final prefs = PrefsManager.instance;
-    await prefs.setString(_timeoutModelKey, modelJson);
-  }
-
-  Future<String?> loadTimeoutModel() async {
-    final prefs = PrefsManager.instance;
-    return prefs.getString(_timeoutModelKey);
-  }
-
-  Future<void> clearTimeoutModel() async {
-    final prefs = PrefsManager.instance;
-    await prefs.remove(_timeoutModelKey);
-  }
 }

lib/services/timeout_prediction_service.dart

@@ -1,5 +1,4 @@
-import 'dart:convert';
-import 'dart:math';
+import 'dart:async';
 import 'package:flutter/foundation.dart';
 import 'package:ml_algo/ml_algo.dart';
 import 'package:ml_dataframe/ml_dataframe.dart';
@@ -9,16 +8,13 @@ import 'storage_service.dart';
 class _ContactStats {
   int count = 0;
   double _sum = 0;
-  double _sumSq = 0;
 
   void add(double ms) {
     count++;
     _sum += ms;
-    _sumSq += ms * ms;
   }
 
   double get mean => _sum / count;
-  double get stdDev => sqrt((_sumSq / count) - (mean * mean));
 }
 
 class TimeoutPredictionService extends ChangeNotifier {
@@ -27,9 +23,10 @@ class TimeoutPredictionService extends ChangeNotifier {
   static const int minObservations = 10;
   static const int maxObservations = 100;
   static const int _retrainInterval = 5;
+  // 1.5x multiplier on raw prediction to account for variance in delivery
+  // times — tight enough to improve on worst-case physics, loose enough
+  // to avoid premature timeouts from model noise.
   static const double _safetyMargin = 1.5;
-  static const int _minTimeoutMs = 2000;
-  static const int _maxTimeoutMs = 120000;
   static const int _minContactObservations = 10;
 
   List<DeliveryObservation> _observations = [];
@@ -37,6 +34,7 @@ class TimeoutPredictionService extends ChangeNotifier {
   List<String> _activeFeatures = [];
   int _observationsSinceLastTrain = 0;
   final Map<String, _ContactStats> _contactStats = {};
+  Timer? _persistTimer;
 
   TimeoutPredictionService(StorageService storage) : _storage = storage;
   TimeoutPredictionService.noStorage() : _storage = null;
@@ -89,7 +87,10 @@ class TimeoutPredictionService extends ChangeNotifier {
       _trainModel();
     }
 
+    _persistTimer?.cancel();
+    _persistTimer = Timer(const Duration(seconds: 2), () {
       _storage?.saveDeliveryObservations(_observations);
+    });
     debugPrint(
       'TimeoutPrediction: recorded ${tripTimeMs}ms for $pathLength hops '
       '(${_observations.length} total)',
@@ -123,7 +124,9 @@ class TimeoutPredictionService extends ChangeNotifier {
 
       final prediction = _model!.predict(features);
       final rawValue = prediction.rows.first.first;
-      var predictedMs = (rawValue is double) ? rawValue : (rawValue as num).toDouble();
+      var predictedMs = (rawValue is double)
+          ? rawValue
+          : (rawValue as num).toDouble();
 
       debugPrint(
         'TimeoutPrediction: raw prediction=$predictedMs for '
@@ -142,8 +145,8 @@ class TimeoutPredictionService extends ChangeNotifier {
         }
       }
 
-      final timeout =
-          (predictedMs * _safetyMargin).ceil().clamp(_minTimeoutMs, _maxTimeoutMs);
+      // Connector clamps this between physics min/max bounds
+      final timeout = (predictedMs * _safetyMargin).ceil();
       debugPrint(
         'TimeoutPrediction: ML timeout ${timeout}ms '
         '(raw: ${predictedMs.round()}ms, contact: $contactKey)',
@@ -174,7 +177,9 @@ class TimeoutPredictionService extends ChangeNotifier {
       }
 
       if (_activeFeatures.isEmpty) {
-        debugPrint('TimeoutPrediction: no features with variance, skipping training');
+        debugPrint(
+          'TimeoutPrediction: no features with variance, skipping training',
+        );
         return;
       }
 
@@ -190,25 +195,19 @@ class TimeoutPredictionService extends ChangeNotifier {
         return row;
       });
 
-      final data = DataFrame(
-        [header, ...rows],
-        headerExists: true,
-      );
+      final data = DataFrame([header, ...rows], headerExists: true);
 
       _model = LinearRegressor(data, 'deliveryMs');
       _observationsSinceLastTrain = 0;
 
       // Log training summary with sample predictions
-      final avgMs = _observations.map((o) => o.deliveryMs).reduce((a, b) => a + b) /
+      final avgMs =
+          _observations.map((o) => o.deliveryMs).reduce((a, b) => a + b) /
           _observations.length;
       debugPrint(
         'TimeoutPrediction: trained on ${_observations.length} observations '
         '(avg: ${avgMs.round()}ms, features: $_activeFeatures)',
       );
-
-      final modelJson = jsonEncode(_model!.toJson());
-      _storage?.saveTimeoutModel(modelJson);
-      notifyListeners();
     } catch (e) {
       debugPrint('TimeoutPrediction: training failed: $e');
     }

test/services/ml_algo_sanity_test.dart

@@ -6,7 +6,8 @@ import 'package:ml_dataframe/ml_dataframe.dart';
 void main() {
   test('LinearRegressor basic sanity check', () {
     // Simple: y = 2x + 100
-    final data = DataFrame([
+    final data = DataFrame(
+      [
         [1.0, 102.0],
         [2.0, 104.0],
         [3.0, 106.0],
@@ -17,7 +18,10 @@ void main() {
         [50.0, 200.0],
         [0.0, 100.0],
         [100.0, 300.0],
-    ], headerExists: false, header: ['x', 'y']);
+      ],
+      headerExists: false,
+      header: ['x', 'y'],
+    );
 
     debugPrint('Training data columns: ${data.header}');
     debugPrint('Training data rows: ${data.rows.length}');
@@ -25,7 +29,9 @@ void main() {
     final model = LinearRegressor(data, 'y');
 
     final testDf = DataFrame(
-      [[25.0]],
+      [
+        [25.0],
+      ],
       headerExists: false,
       header: ['x'],
     );
@@ -38,28 +44,41 @@ void main() {
 
   test('LinearRegressor multi-feature with constant column produces zeros', () {
     // isFlood=0 for all rows → zero-variance column → singular matrix
-    final data = DataFrame([
+    final data = DataFrame(
+      [
         [0.0, 50.0, 14.0, 0.0, 1900.0],
         [0.0, 80.0, 14.0, 0.0, 2200.0],
         [2.0, 50.0, 14.0, 0.0, 5000.0],
         [4.0, 50.0, 14.0, 0.0, 9500.0],
-    ], headerExists: false, header: [
-      'pathLength', 'messageBytes', 'hourOfDay', 'isFlood', 'deliveryMs',
-    ]);
+      ],
+      headerExists: false,
+      header: [
+        'pathLength',
+        'messageBytes',
+        'hourOfDay',
+        'isFlood',
+        'deliveryMs',
+      ],
+    );
 
     final model = LinearRegressor(data, 'deliveryMs');
     final testDf = DataFrame(
-      [[2.0, 50.0, 14.0, 0.0]],
+      [
+        [2.0, 50.0, 14.0, 0.0],
+      ],
       headerExists: false,
       header: ['pathLength', 'messageBytes', 'hourOfDay', 'isFlood'],
     );
     final pred = model.predict(testDf).rows.first.first;
-    debugPrint('With constant isFlood column: hops=2 → ${(pred as num).round()}ms (likely 0)');
+    debugPrint(
+      'With constant isFlood column: hops=2 → ${(pred as num).round()}ms (likely 0)',
+    );
   });
 
   test('LinearRegressor 2-feature works correctly', () {
     // Just pathLength + messageBytes → deliveryMs
-    final data = DataFrame([
+    final data = DataFrame(
+      [
         [0.0, 50.0, 1900.0],
         [0.0, 80.0, 2200.0],
         [2.0, 50.0, 5000.0],
@@ -70,13 +89,18 @@ void main() {
         [2.0, 30.0, 4800.0],
         [4.0, 30.0, 9000.0],
         [0.0, 60.0, 2000.0],
-    ], headerExists: false, header: ['pathLength', 'messageBytes', 'deliveryMs']);
+      ],
+      headerExists: false,
+      header: ['pathLength', 'messageBytes', 'deliveryMs'],
+    );
 
     final model = LinearRegressor(data, 'deliveryMs');
 
     for (final hops in [0.0, 2.0, 4.0]) {
       final testDf = DataFrame(
-        [[hops, 50.0]],
+        [
+          [hops, 50.0],
+        ],
         headerExists: false,
         header: ['pathLength', 'messageBytes'],
       );
@@ -87,7 +111,8 @@ void main() {
 
   test('LinearRegressor multi-feature with variance in all columns', () {
     // Mix flood and direct so isFlood has variance
-    final data = DataFrame([
+    final data = DataFrame(
+      [
         [0.0, 50.0, 14.0, 0.0, 1900.0],
         [0.0, 80.0, 10.0, 0.0, 2200.0],
         [2.0, 50.0, 16.0, 0.0, 5000.0],
@@ -98,9 +123,16 @@ void main() {
         [-1.0, 60.0, 18.0, 1.0, 6500.0],
         [-1.0, 30.0, 10.0, 1.0, 4000.0],
         [-1.0, 80.0, 22.0, 1.0, 7000.0],
-    ], headerExists: false, header: [
-      'pathLength', 'messageBytes', 'hourOfDay', 'isFlood', 'deliveryMs',
-    ]);
+      ],
+      headerExists: false,
+      header: [
+        'pathLength',
+        'messageBytes',
+        'hourOfDay',
+        'isFlood',
+        'deliveryMs',
+      ],
+    );
 
     final model = LinearRegressor(data, 'deliveryMs');
 
@@ -116,7 +148,9 @@ void main() {
         header: ['pathLength', 'messageBytes', 'hourOfDay', 'isFlood'],
       );
       final pred = model.predict(testDf).rows.first.first;
-      debugPrint('4-feature: hops=${tc[0]} flood=${tc[3]} → ${(pred as num).round()}ms');
+      debugPrint(
+        '4-feature: hops=${tc[0]} flood=${tc[3]} → ${(pred as num).round()}ms',
+      );
     }
   });
 }

test/services/timeout_prediction_service_test.dart

@@ -64,9 +64,9 @@ void main() {
     expect(direct4!, greaterThan(direct2!));
     expect(direct2, greaterThan(direct0!));
 
-    // All should be within the clamp range
-    expect(direct0, greaterThanOrEqualTo(2000));
-    expect(direct4, lessThanOrEqualTo(120000));
+    // All should be positive
+    expect(direct0, greaterThan(0));
+    expect(direct4, greaterThan(0));
 
     // Print predictions for visibility
     debugPrint('Predictions (with 1.5x safety margin):');