diff --git a/server/biz/service/training.go b/server/biz/service/training.go
index b8b0d66..9a62568 100644
--- a/server/biz/service/training.go
+++ b/server/biz/service/training.go
@@ -84,15 +84,7 @@ func (s *trainingService) pollOne(ctx context.Context, runner common.TrainingRun
 		return
 	}
 	if result != "" {
-		// result.json 带 error 字段 = 脚本异常退出（如 tflite 导出失败），置失败并带出原因
-		var resErr struct {
-			Error string `json:"error"`
-		}
-		if json.Unmarshal([]byte(result), &resErr) == nil && resErr.Error != "" {
-			_ = s.finishFailed(ctx, t, "训练失败: %s", resErr.Error)
-			return
-		}
-		s.finishSuccess(ctx, runner, job, t, result, tail)
+		s.handleResult(ctx, runner, job, t, result, tail)
 		return
 	}
 	alive := true
@@ -117,6 +109,13 @@ func (s *trainingService) pollOne(ctx context.Context, runner common.TrainingRun
 		return
 	}
 	if !alive {
+		// 竞态防护：脚本原子写结果文件后进程随即退出，轮询可能命中「结果未就绪 + 进程已死」窗口；
+		// 判死前延迟重试一次，确认文件确实缺席（结果文件迟到属正常时序，非训练失败）
+		time.Sleep(2 * time.Second)
+		if r2, e2 := runner.FetchResult(ctx, job); e2 == nil && r2 != "" {
+			s.handleResult(ctx, runner, job, t, r2, tail)
+			return
+		}
 		_ = s.finishFailed(ctx, t, "训练进程已退出（无结果文件）")
 		return
 	}
@@ -127,6 +126,18 @@ func (s *trainingService) pollOne(ctx context.Context, runner common.TrainingRun
 	}
 }
 
+// handleResult 处理已就绪的 result.json：error 字段 = 脚本异常退出 → 置失败带出原因；否则按成功收尾
+func (s *trainingService) handleResult(ctx context.Context, runner common.TrainingRunner, job *common.TrainingJob, t *entity.ModelTraining, result, tail string) {
+	var resErr struct {
+		Error string `json:"error"`
+	}
+	if json.Unmarshal([]byte(result), &resErr) == nil && resErr.Error != "" {
+		_ = s.finishFailed(ctx, t, "训练失败: %s", resErr.Error)
+		return
+	}
+	s.finishSuccess(ctx, runner, job, t, result, tail)
+}
+
 // finishSuccess 训练成功：解析 result.json（最终指标 + 类别名）→ 拉取 tflite → 更新任务。
 // tflite 直写 trainings/<数据集名>.tflite（当前生效模型唯一位，无 per-task 存档、无 zip）。
 func (s *trainingService) finishSuccess(ctx context.Context, runner common.TrainingRunner, job *common.TrainingJob, t *entity.ModelTraining, result, tail string) {
diff --git a/server/common/training_runner.go b/server/common/training_runner.go
index df6aa76..1c0c24e 100644
--- a/server/common/training_runner.go
+++ b/server/common/training_runner.go
@@ -309,7 +309,7 @@ func (r *sshRunner) IsAlive(ctx context.Context, job *TrainingJob) (bool, error)
 	}
 	out, err := r.runCmd(ctx, job, fmt.Sprintf("kill -0 %d 2>/dev/null && echo alive || echo dead", pid))
 	if err != nil {
-		return false, nil // 连接故障按"状态未知"处理，不误判任务结束
+		return false, err // 连接故障 = 状态未知，交上层处理，不吞错
 	}
 	return strings.TrimSpace(out) == "alive", nil
 }
@@ -330,9 +330,15 @@ func (r *sshRunner) FetchLogTail(ctx context.Context, job *TrainingJob) (string,
 }
 
 func (r *sshRunner) FetchResult(ctx context.Context, job *TrainingJob) (string, error) {
-	out, err := r.runCmd(ctx, job, fmt.Sprintf("cat %s/results/%d.json 2>/dev/null", job.Workdir, job.TaskId))
-	if err != nil || out == "" {
-		return "", nil // 文件不存在 = 仍在跑
+	// if -f 区分「结果文件未就绪」（("", nil)）与 SSH 连接/命令真实失败（("", err)）：
+	// 原 cat 2>/dev/null 两者混为一谈，连接故障会被当成"文件不存在=仍在跑"而吞掉错误
+	path := filepath.Join(job.Workdir, "results", fmt.Sprintf("%d.json", job.TaskId))
+	out, err := r.runCmd(ctx, job, fmt.Sprintf("if [ -f %s ]; then cat %s; else echo __NO_RESULT__; fi", path, path))
+	if err != nil {
+		return "", err
+	}
+	if strings.TrimSpace(out) == "__NO_RESULT__" {
+		return "", nil
 	}
 	return out, nil
 }
diff --git a/server/data/observer.db b/server/data/observer.db
index 3e1c35f..2069632 100644
Binary files a/server/data/observer.db and b/server/data/observer.db differ
diff --git a/server/training/train_server.py b/server/training/train_server.py
index 84c89ec..d44c028 100644
--- a/server/training/train_server.py
+++ b/server/training/train_server.py
@@ -55,11 +55,11 @@ def on_fit_epoch_end(trainer):
     }, ensure_ascii=False) + "\n")
 
 
-def register_callback():
-    # ultralytics ≥8.4 中 callbacks 字典改由 get_default_callbacks() 取得
-    from ultralytics.utils.callbacks import get_default_callbacks
-    callbacks = get_default_callbacks()
-    callbacks["on_fit_epoch_end"].append(on_fit_epoch_end)
+def register_callback(model):
+    # ultralytics ≥8.4 中 trainer 使用 model 实例的 callbacks 字典（model.train 前以
+    # _callbacks=self.callbacks 传入），模块级 get_default_callbacks() 的 append 不生效；
+    # add_callback 追加到 model 字典即随训练触发，进度 jsonl 才有输出
+    model.add_callback("on_fit_epoch_end", on_fit_epoch_end)
 
 
 def write_result(result_file, payload):
@@ -245,8 +245,8 @@ def main():
 
     try:
         from ultralytics import YOLO
-        register_callback()
         model = YOLO("yolov8n.pt")
+        register_callback(model)
         model.train(
             data=data, imgsz=imgsz, epochs=epochs,
             patience=PATIENCE, batch=batch, device=device, workers=4,
diff --git a/server/workspace/trainings/野鸡.tflite b/server/workspace/trainings/野鸡.tflite
new file mode 100644
index 0000000..771fba8
Binary files /dev/null and b/server/workspace/trainings/野鸡.tflite differ