Fix routing bugs and unify chart formatting

- Fix double-click bug: add intermediate yield with loading state
- Fix Original Cost mismatch: use df-based calculation (same as Total Cost by Token)
- Let's ROUTE!! button already disabled until routing model selected
- Unify all token charts to use millions (M) with consistent formatting
- Update hover templates to show values in M format
- Consistent legend positioning across all charts

app.py

@@ -803,27 +803,27 @@ def create_basic_histograms(df: pd.DataFrame, input_price: float, cache_read_pri
 
     token_data = pd.DataFrame({
         "Token Type": ["Uncached Input", "Cache Read", "Cache Creation", "Completion"],
-        "Total Tokens": [total_uncached_input, total_cache_read, total_cache_creation, total_completion],
+        "Tokens (M)": [total_uncached_input / 1e6, total_cache_read / 1e6, total_cache_creation / 1e6, total_completion / 1e6],
     })
 
     fig_tokens = px.bar(
         token_data,
         x="Token Type",
-        y="Total Tokens",
+        y="Tokens (M)",
         title="Total Tokens by Type",
         color="Token Type",
         color_discrete_sequence=["#EF553B", "#19D3F3", "#FFA15A", "#AB63FA"],
     )
     fig_tokens.update_layout(
         xaxis_title="Token Type",
-        yaxis_title="Total Tokens",
+        yaxis_title="Tokens (M)",
         showlegend=False,
         margin=dict(l=40, r=20, t=40, b=40),
     )
 
-    total_all = token_data["Total Tokens"].sum()
+    total_all = total_uncached_input + total_cache_read + total_cache_creation + total_completion
     fig_tokens.add_annotation(
-        text=f"Total: {total_all:,.0f}",
+        text=f"Total: {total_all/1e6:.2f}M",
         xref="paper", yref="paper",
         x=0.95, y=0.95, showarrow=False,
         font=dict(size=12),
@@ -844,40 +844,40 @@ def create_basic_histograms(df: pd.DataFrame, input_price: float, cache_read_pri
     fig_stacked.add_trace(go.Bar(
         name="Uncached Input",
         x=df_sorted["trajectory_idx"],
-        y=df_sorted["uncached_input_tokens"],
+        y=df_sorted["uncached_input_tokens"] / 1e6,
         marker_color="#EF553B",
-        hovertemplate="Trajectory: %{x}<br>Uncached Input: %{y:,.0f}<extra></extra>",
+        hovertemplate="Trajectory: %{x}<br>Uncached Input: %{y:.3f}M<extra></extra>",
     ))
 
     fig_stacked.add_trace(go.Bar(
         name="Cache Read",
         x=df_sorted["trajectory_idx"],
-        y=df_sorted["cache_read_tokens"],
+        y=df_sorted["cache_read_tokens"] / 1e6,
         marker_color="#19D3F3",
-        hovertemplate="Trajectory: %{x}<br>Cache Read: %{y:,.0f}<extra></extra>",
+        hovertemplate="Trajectory: %{x}<br>Cache Read: %{y:.3f}M<extra></extra>",
     ))
 
     fig_stacked.add_trace(go.Bar(
         name="Cache Creation",
         x=df_sorted["trajectory_idx"],
-        y=df_sorted["cache_creation_tokens"],
+        y=df_sorted["cache_creation_tokens"] / 1e6,
         marker_color="#FFA15A",
-        hovertemplate="Trajectory: %{x}<br>Cache Creation: %{y:,.0f}<extra></extra>",
+        hovertemplate="Trajectory: %{x}<br>Cache Creation: %{y:.3f}M<extra></extra>",
     ))
 
     fig_stacked.add_trace(go.Bar(
         name="Completion",
         x=df_sorted["trajectory_idx"],
-        y=df_sorted["completion_tokens"],
+        y=df_sorted["completion_tokens"] / 1e6,
         marker_color="#AB63FA",
-        hovertemplate="Trajectory: %{x}<br>Completion: %{y:,.0f}<extra></extra>",
+        hovertemplate="Trajectory: %{x}<br>Completion: %{y:.3f}M<extra></extra>",
     ))
 
     fig_stacked.update_layout(
         barmode="stack",
         title="Tokens per Trajectory (stacked)",
         xaxis_title="Trajectory (sorted by total tokens)",
-        yaxis_title="Tokens",
+        yaxis_title="Tokens (M)",
         legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="right", x=1),
         margin=dict(l=50, r=20, t=60, b=40),
     )
@@ -1528,6 +1528,24 @@ def build_app():
                 )
                 return
 
+            df_key = "meta" if source == "Metadata" else "calculated"
+            df = state_data.get(df_key)
+            if df is not None and not df.empty:
+                if source == "Calculated":
+                    df = apply_thinking_overhead(df.copy(), overhead)
+                    if not with_cache:
+                        df = apply_no_cache(df)
+                df_temp = df.copy()
+                df_temp["uncached_input"] = (df_temp["prompt_tokens"] - df_temp["cache_read_tokens"] - df_temp["cache_creation_tokens"]).clip(lower=0)
+                total_original_cost_from_df = (
+                    df_temp["uncached_input"].sum() * base_input / 1e6 +
+                    df["cache_read_tokens"].sum() * base_cache_read / 1e6 +
+                    df["cache_creation_tokens"].sum() * base_cache_creation / 1e6 +
+                    df["completion_tokens"].sum() * base_completion / 1e6
+                )
+            else:
+                total_original_cost_from_df = None
+
             base_prices = {
                 "input": base_input,
                 "cache_read": base_cache_read,
@@ -1621,7 +1639,11 @@ def build_app():
 
             total_base_cost = calc_cost(total_base_tokens, base_prices)
             total_routing_cost = calc_cost(total_routing_tokens, routing_prices)
-            total_original_cost = calc_cost(total_original_tokens, base_prices)
+
+            if total_original_cost_from_df is not None:
+                total_original_cost = total_original_cost_from_df
+            else:
+                total_original_cost = calc_cost(total_original_tokens, base_prices)
 
             total_routed_cost = total_base_cost + total_routing_cost
             savings = total_original_cost - total_routed_cost
@@ -1645,6 +1667,13 @@ def build_app():
             additional_token_models = [(routing_model_1_val, total_routing_tokens)]
             additional_cost_models = [(routing_model_1_val, routing_costs)]
 
+            yield (
+                gr.update(visible=True, value="⏳ Creating charts..."),
+                gr.update(visible=True),
+                None,
+                None,
+            )
+
             tokens_chart = create_routed_token_chart(total_base_tokens, additional_token_models)
             cost_chart = create_routed_cost_chart(base_costs, additional_cost_models)