Clarify per-tech modulation probe and MoE routing description

app.py

@@ -890,7 +890,7 @@ with gr.Blocks(title="LWM-Spectro Lab") as demo:
                 """
             ### 🎯 Lightweight Modulation Head
             - Prototype how well the frozen LWM backbone separates modulation formats for each technology using spectrograms as input.
-            - The adaptive k-NN classifier approximates the behavior of the downstream residual 1D-CNN before heavy training.
+            - The adaptive k-NN classifier approximates the behavior of the downstream residual 1D-CNN before heavy training; each tech is evaluated separately to measure its expert’s modulation discrimination.
             - Sweep train/test splits and seeds to gauge robustness when only a portion of the dataset is labeled.
             """
             )
@@ -921,7 +921,7 @@ with gr.Blocks(title="LWM-Spectro Lab") as demo:
                 """
             ### 🌪️ Joint Channel Dynamics Benchmark
             - Evaluate the precomputed MoE embeddings on the 14-class joint SNR/Doppler recognition task.
-            - Mirrors the second stage of our reliability workflow where mobility-aware cues guide SNR-aware routing.
+            - Mirrors the second stage of our reliability workflow where, without an explicit technology label, the MoE router sends samples to the most relevant expert and mobility-aware cues guide SNR-aware routing.
             - Upload or reference Hub-hosted tensors to compare MoE vs. raw spectrogram baselines before fine-tuning heavier heads.
             """
             )