Update app.py

app.py

@@ -245,8 +245,6 @@ def simulate_jitter(seed: int,
         jitter_rft_next = jitter + u_rft * 0.35
         duty_rft += int(abs(u_rft) > 0.01)
 
-        # Apply combined evolution (keep it fair by updating the same jitter state)
-        # We store both "what baseline would do" and "what RFT would do"
         act_baseline = u_base
         act_rft = u_rft
 
@@ -364,10 +362,6 @@ def simulate_landing(seed: int,
         conf = rft_confidence(uncertainty)
 
         # anomaly definition (reduced spam): only count if materially bad
-        # pitch/yaw/roll are not modelled here; we count "control-relevant" anomalies:
-        # - high wind
-        # - high lateral error near ground
-        # - high descent rate near ground
         anomaly_types = []
         if wind > (0.85 * wind_max):
             anomaly_types.append("High wind")
@@ -380,7 +374,7 @@ def simulate_landing(seed: int,
         if is_anomaly:
             anomalies += 1
 
-        # Baseline control: continuous proportional
+        # Baseline control: continuous proportional (computed for reference / logging)
         u_base_x = -kp_baseline * meas_x - 0.25 * meas_xv
         u_base_v = -kp_baseline * (meas_vv + 5.0)  # target ~ -5 m/s
 
@@ -399,12 +393,10 @@ def simulate_landing(seed: int,
             actions += 1
 
         # apply dynamics
-        # vertical: vv integrates thrust + gravity (simplified)
         g = -9.81
         vv = vv + (g + 0.18 * u_rft_v + 0.08 * thrust_dev) * dt
         alt = max(0.0, alt + vv * dt)
 
-        # lateral: wind pushes, control counters
         xv = xv + (0.35 * wind - 0.30 * u_rft_x) * dt
         x = x + xv * dt
 
@@ -516,7 +508,6 @@ def run_benchmarks(seed: int,
         tau_gain=tau_gain,
         show_debug=False
     )
-    # Baseline alerts equals df baseline count; derive from CSV
     neo_df = pd.read_csv(neo_csv)
     neo_base = int(neo_df["baseline_alert"].sum())
     neo_rft = int(neo_df["rft_alert"].sum())
@@ -541,14 +532,13 @@ def run_benchmarks(seed: int,
         dt=land_dt,
         wind_max=land_wind,
         thrust_noise=land_thrust_noise,
-        kp_baseline=0.06,   # baseline weaker to show "continuous but less decisive"
-        kp_rft=0.10,        # RFT stronger but gated and phase-weighted
+        kp_baseline=0.06,
+        kp_rft=0.10,
         gate_threshold=0.55,
         tau_gain=tau_gain,
         goal_m=10.0
     )
 
-    # Scorecard table
     score = pd.DataFrame([
         {
             "Module": "NEO",
@@ -578,7 +568,6 @@ def run_benchmarks(seed: int,
 
     score_path = df_to_csv_file(score, f"bench_score_seed{seed}.csv")
 
-    # Summary text
     txt = (
         f"Benchmarks (seed={seed})\n"
         f"- NEO: baseline alerts={neo_base}, RFT candidates={neo_candidates}, RFT filtered={neo_rft}\n"
@@ -586,7 +575,7 @@ def run_benchmarks(seed: int,
         f"- Landing: final offset={l_sum['final_landing_offset_m']:.2f} m (goal 10 m), anomalies={l_sum['total_anomalies_detected']}, actions={l_sum['total_control_actions']}\n"
     )
 
-    all_imgs = neo_imgs + jit_imgs + land_imgs
+    all_imgs = neo_imgs + jit_imgs + land_imgs  # 3 + 3 + 4 = 10 images
     return txt, score, score_path, all_imgs, [neo_csv, jit_csv, land_csv]
 
 # -----------------------------
@@ -781,7 +770,10 @@ def ui_run_landing(seed, steps, dt, wind_max, thrust_noise, kp_base, kp_rft, gat
     summary_txt = json.dumps(summary, indent=2)
     return summary_txt, imgs[0], imgs[1], imgs[2], imgs[3], csv_path
 
-def ui_run_bench(seed, neo_steps, neo_dt, neo_alert_km, neo_noise_km, jit_steps, jit_dt, jit_noise, land_steps, land_dt, land_wind, land_thrust_noise, tau_gain):
+def ui_run_bench(seed, neo_steps, neo_dt, neo_alert_km, neo_noise_km,
+                 jit_steps, jit_dt, jit_noise,
+                 land_steps, land_dt, land_wind, land_thrust_noise,
+                 tau_gain):
     txt, score_df, score_csv, imgs, logs = run_benchmarks(
         seed=int(seed),
         neo_steps=int(neo_steps), neo_dt=float(neo_dt), neo_alert_km=float(neo_alert_km), neo_noise_km=float(neo_noise_km),
@@ -789,7 +781,23 @@ def ui_run_bench(seed, neo_steps, neo_dt, neo_alert_km, neo_noise_km, jit_steps,
         land_steps=int(land_steps), land_dt=float(land_dt), land_wind=float(land_wind), land_thrust_noise=float(land_thrust_noise),
         tau_gain=float(tau_gain)
     )
-    return txt, score_df, score_csv, imgs[0], imgs[1], imgs[2], imgs[3], imgs[4], imgs[5], imgs[6], imgs[7], imgs[8], logs[0], logs[1], logs[2]
+
+    # IMPORTANT:
+    # The Live Console Benchmarks tab has 16 outputs:
+    #  - 1 textbox, 1 dataframe, 1 file
+    #  - 10 images
+    #  - 3 log files
+    # run_benchmarks returns 10 images (3 NEO + 3 jitter + 4 landing)
+    # so we MUST return imgs[0]..imgs[9] here.
+    while len(imgs) < 10:
+        imgs.append(None)
+
+    return (
+        txt, score_df, score_csv,
+        imgs[0], imgs[1], imgs[2], imgs[3], imgs[4],
+        imgs[5], imgs[6], imgs[7], imgs[8], imgs[9],
+        logs[0], logs[1], logs[2]
+    )
 
 with gr.Blocks(title="RFT — Agent Console (NEO / Jitter / Landing)") as demo:
     gr.Markdown(HOME_MD)
@@ -944,7 +952,7 @@ with gr.Blocks(title="RFT — Agent Console (NEO / Jitter / Landing)") as demo:
 
         # ----------------------------------------------------------
         with gr.Tab("Benchmarks"):
-            gr.Markdown("# Benchmarks\nIf someone wants to argue, this is the tab.\nBaseline vs RFT runs, same seed, exported logs.\nUse Live Console to run full benchmark packs.\n")
+            gr.Markdown("# Benchmarks\nBaseline vs RFT runs, same seed, exported logs.\nUse Live Console to run full benchmark packs.\n")
 
         # ----------------------------------------------------------
         with gr.Tab("Theory → Practice"):