page4_results.py

"""
page4_results.py — Simulation Results: 5 tables + charts for all analysis dimensions
"""

import streamlit as st
import plotly.graph_objects as go
import plotly.express as px
import pandas as pd
import numpy as np
from styles import (inject_global_css, page_header, metric_row,
                    WCO_GOLD, WCO_BLUE, WCO_GREEN, WCO_RED,
                    WCO_CARD_BG, WCO_BORDER, WCO_MUTED, RISK_COLORS)
from simulation_engine import (build_rule_hit_table, build_channel_table,
                                build_exploration_discovery_table,
                                build_offence_db_table, build_risk_score_table,
                                RISK_AREAS)

RISK_LEVEL_COLORS = {
    "FRAUD_DETECTED": "#C8102E",
    "DOC_QUERY":      "#F5A800",
    "CLEAN":          "#00843D",
    "NOT_INSPECTED":  "#6B85AA",
}


def offence_db_growth_chart(df):
    """Cumulative offence DB additions over bill sequence."""
    df_sorted = df.sort_index()
    df_sorted["cum_offence"] = df_sorted["added_to_offence_db"].cumsum()

    # Break down by risk area
    traces = []
    for area_name, area_cfg in RISK_AREAS.items():
        a_df = df_sorted[df_sorted["risk_area"] == area_name].copy()
        a_df["cum"] = a_df["added_to_offence_db"].cumsum()
        traces.append(go.Scatter(
            x=a_df.index, y=a_df["cum"],
            name=f"{area_cfg['icon']} {area_name}",
            line=dict(color=area_cfg["color"], width=2),
            mode="lines",
            hovertemplate=f"<b>{area_name}</b><br>Bill: %{{x}}<br>Cumulative: %{{y}}<extra></extra>",
        ))
    traces.append(go.Scatter(
        x=df_sorted.index, y=df_sorted["cum_offence"],
        name="📊 Total", line=dict(color=WCO_GOLD, width=3, dash="dot"),
        hovertemplate="<b>Total</b><br>Bill: %{x}<br>Cumulative: %{y}<extra></extra>",
    ))

    fig = go.Figure(traces)
    fig.update_layout(
        paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
        font=dict(family="IBM Plex Sans", color="#D0DCF0", size=11),
        height=380,
        title=dict(text="<b>Real-Time Offence Database Growth (Cumulative)</b>",
                   font=dict(color=WCO_GOLD, size=14, family="Playfair Display"), x=0.5),
        xaxis=dict(title="Bill Sequence", gridcolor="#1E3A6E"),
        yaxis=dict(title="Offences Added to DB", gridcolor="#1E3A6E"),
        legend=dict(bgcolor="#0F1C35", bordercolor=WCO_BORDER, font=dict(size=10)),
        margin=dict(l=55, r=20, t=55, b=45),
    )
    return fig


def exploration_vs_exploitation_chart(df):
    """Compare detection across exploit vs explore bills."""
    cats = ["Exploitation Bills", "Exploration Bills"]
    exploit_df = df[df["is_exploration"] == 0]
    explore_df  = df[df["is_exploration"] == 1]

    def detection_stats(sub):
        tot    = len(sub)
        fraud  = (sub["inspection_outcome"] == "FRAUD_DETECTED").sum()
        rev    = sub["detected_revenue"].sum()
        rate   = 100 * fraud / tot if tot else 0
        return tot, fraud, rev, rate

    e_tot, e_fraud, e_rev, e_rate = detection_stats(exploit_df)
    x_tot, x_fraud, x_rev, x_rate = detection_stats(explore_df)

    fig = go.Figure()
    fig.add_trace(go.Bar(name="🎯 Exploitations", x=["Bills","Frauds Detected","Revenue ($K)"],
                         y=[e_tot, e_fraud, e_rev/1000],
                         marker_color="#0066CC", opacity=0.85,
                         text=[e_tot, e_fraud, f"${e_rev/1000:.1f}K"], textposition="outside",
                         textfont=dict(color="#D0DCF0")))
    fig.add_trace(go.Bar(name="🔍 Explorations",  x=["Bills","Frauds Detected","Revenue ($K)"],
                         y=[x_tot, x_fraud, x_rev/1000],
                         marker_color="#CC77FF", opacity=0.85,
                         text=[x_tot, x_fraud, f"${x_rev/1000:.1f}K"], textposition="outside",
                         textfont=dict(color="#D0DCF0")))
    fig.update_layout(
        paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
        font=dict(family="IBM Plex Sans", color="#D0DCF0", size=12),
        height=340, barmode="group",
        title=dict(text="<b>Exploitation vs Exploration: Detection Comparison</b>",
                   font=dict(color=WCO_GOLD, size=14, family="Playfair Display"), x=0.5),
        xaxis=dict(gridcolor="#1E3A6E"),
        yaxis=dict(gridcolor="#1E3A6E", title="Count / Value"),
        legend=dict(bgcolor="#0F1C35", bordercolor=WCO_BORDER),
        margin=dict(l=50, r=20, t=55, b=40),
    )
    return fig


def weight_evolution_chart(df, orig_weights, updated_weights):
    """Visualise how rule weights increased after simulation."""
    rids, orig, updated, deltas = [], [], [], []
    for area in RISK_AREAS.values():
        for rule in area["rules"]:
            rid = rule["id"]
            o = orig_weights.get(rid, rule["weight"])
            u = updated_weights.get(rid, o)
            rids.append(rid)
            orig.append(o)
            updated.append(u)
            deltas.append(round(u - o, 4))

    fig = go.Figure()
    fig.add_trace(go.Bar(x=rids, y=orig, name="Original Weights",
                         marker_color="#1E3A6E", opacity=0.9))
    fig.add_trace(go.Bar(x=rids, y=updated, name="Updated Weights",
                         marker_color=WCO_GOLD, opacity=0.85))
    fig.add_trace(go.Scatter(x=rids, y=deltas, name="Δ Weight", mode="markers+lines",
                             marker=dict(color=WCO_RED, size=9, symbol="triangle-up"),
                             line=dict(color=WCO_RED, dash="dot", width=1.5),
                             yaxis="y2"))
    fig.update_layout(
        paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
        font=dict(family="IBM Plex Sans", color="#D0DCF0", size=10),
        height=360, barmode="group",
        title=dict(text="<b>Rule Weight Evolution After Self-Learning Feedback</b>",
                   font=dict(color=WCO_GOLD, size=14, family="Playfair Display"), x=0.5),
        xaxis=dict(gridcolor="#1E3A6E", tickangle=-45),
        yaxis=dict(title="Weight", gridcolor="#1E3A6E"),
        yaxis2=dict(title="Δ Weight", overlaying="y", side="right",
                    gridcolor="#1E3A6E", showgrid=False),
        legend=dict(bgcolor="#0F1C35", bordercolor=WCO_BORDER),
        margin=dict(l=55, r=55, t=55, b=90),
    )
    return fig


def styled_table(df: pd.DataFrame, title: str, highlight_col: str = None,
                  color_map: dict = None):
    st.markdown(f"""
    <div style="color:{WCO_GOLD};font-family:'Playfair Display',serif;
                font-size:15px;font-weight:700;margin:16px 0 10px;">
      {title}
    </div>""", unsafe_allow_html=True)
    if highlight_col and color_map:
        def _style(val):
            col = color_map.get(val, "")
            return f"background-color:{col}22;color:{col}" if col else ""
        styled = df.style.applymap(_style, subset=[highlight_col]) \
                         .set_properties(**{"background-color": WCO_CARD_BG,
                                            "color": "#D0DCF0", "font-size": "12px"})
        st.dataframe(styled, use_container_width=True, height=min(38 * len(df) + 40, 420))
    else:
        st.dataframe(
            df.style.set_properties(**{"background-color": WCO_CARD_BG,
                                        "color": "#D0DCF0", "font-size": "12px"}),
            use_container_width=True, height=min(38 * len(df) + 40, 420),
        )


def exploration_scatter_tab(df: pd.DataFrame):
    """
    Full scatter-plot breakdown of the bATE exploration logic applied to all bills.
    Shows every formula component:
      unc_i  = -1.8 * |fraud_score - 0.5| + 1
      S_i    = unc_i * log(pred_revenue + ε)
    Colour = channel assignment.  Shape = exploration pick or not.
    """
    import numpy as np

    # ── Formula explanation banner ────────────────────────────────
    st.markdown("""
    <div class="wco-card-gold">
    <h3>🧭 bATE Exploration Logic — All Bills Visualised</h3>
    <div style="display:flex;gap:30px;flex-wrap:wrap;font-size:13px;color:#D0DCF0;line-height:2;">
      <div>
        <b style="color:#CC77FF;">Step 1 — Uncertainty Score</b><br/>
        <code style="background:#111D30;padding:4px 10px;border-radius:5px;
                     color:#C8A951;font-size:13px;">
          unc_i = −1.8 × |ŷᶜˡˢ − 0.5| + 1
        </code><br/>
        <span style="color:#6B85AA;font-size:12px;">
          Max = 1.0 when fraud_score = 0.5 (model completely unsure)<br/>
          Min ≈ 0.1 when fraud_score → 0 or → 1 (model very confident)
        </span>
      </div>
      <div>
        <b style="color:#CC77FF;">Step 2 — Scale Factor (bATE)</b><br/>
        <code style="background:#111D30;padding:4px 10px;border-radius:5px;
                     color:#C8A951;font-size:13px;">
          S_i = unc_i × log(ŷʳᵉᵛ + ε)
        </code><br/>
        <span style="color:#6B85AA;font-size:12px;">
          High S_i = uncertain AND high revenue → top exploration candidates<br/>
          K-means++ diversity applied on top to avoid redundant picks
        </span>
      </div>
    </div>
    </div>""", unsafe_allow_html=True)

    st.markdown("<br/>", unsafe_allow_html=True)

    # ── Colour / marker helpers ───────────────────────────────────
    ch_color  = {"RED": "#C8102E", "YELLOW": "#F5A800", "GREEN": "#00843D"}
    ch_marker = {"RED": "circle", "YELLOW": "diamond", "GREEN": "square"}

    def base_layout(title, xlab, ylab, h=460):
        return dict(
            paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
            font=dict(family="IBM Plex Sans", color="#D0DCF0", size=11),
            height=h,
            title=dict(text=f"<b>{title}</b>",
                       font=dict(color=WCO_GOLD, size=14,
                                 family="Playfair Display"), x=0.5),
            xaxis=dict(title=xlab, gridcolor="#1E3A6E", zeroline=False),
            yaxis=dict(title=ylab, gridcolor="#1E3A6E", zeroline=False),
            legend=dict(bgcolor="#0F1C35", bordercolor=WCO_BORDER,
                        font=dict(size=10)),
            margin=dict(l=60, r=20, t=55, b=55),
        )

    # ─────────────────────────────────────────────────────────────
    # SCATTER 1 — fraud_score vs uncertainty_score (the unc_i curve)
    # ─────────────────────────────────────────────────────────────
    st.markdown('<div class="section-title">① Fraud Score vs Uncertainty Score — The unc_i Curve</div>',
                unsafe_allow_html=True)
    st.markdown("""<div class="alert-blue">
      The <b>uncertainty score unc_i</b> is a concave function maximised when
      <code>fraud_score ≈ 0.5</code>. Bills near 0.5 are the model's blind spots —
      exploration targets. Bills near 0 or 1 are confident predictions (exploitation territory).
      <b style="color:#CC77FF;">Purple stars</b> = actual gATE exploration picks.
    </div>""", unsafe_allow_html=True)

    # Theoretical curve overlay
    x_curve = np.linspace(0, 1, 300)
    y_curve  = -1.8 * np.abs(x_curve - 0.5) + 1

    fig1 = go.Figure()

    # Theoretical curve
    fig1.add_trace(go.Scatter(
        x=x_curve, y=y_curve,
        mode="lines",
        name="unc_i = −1.8×|ŷ−0.5|+1 (theoretical)",
        line=dict(color="#C8A951", width=2.5, dash="dot"),
        hoverinfo="skip",
    ))

    # All bills coloured by channel
    for ch in ["GREEN", "YELLOW", "RED"]:
        sub = df[df["channel"] == ch]
        non_exp = sub[sub["is_exploration"] == 0]
        fig1.add_trace(go.Scatter(
            x=non_exp["fraud_score"],
            y=non_exp["uncertainty_score"],
            mode="markers",
            name=f"{ch} (exploitation)",
            marker=dict(
                color=ch_color[ch], size=5, opacity=0.45,
                symbol=ch_marker[ch],
                line=dict(width=0),
            ),
            hovertemplate=(
                f"<b>{ch}</b><br>"
                "Fraud Score: %{x:.3f}<br>"
                "Uncertainty: %{y:.3f}<br>"
                "<extra></extra>"
            ),
        ))

    # Exploration picks on top
    exp_df = df[df["is_exploration"] == 1]
    fig1.add_trace(go.Scatter(
        x=exp_df["fraud_score"],
        y=exp_df["uncertainty_score"],
        mode="markers",
        name="🔍 gATE Exploration Picks",
        marker=dict(
            color="#CC77FF", size=11, opacity=0.95,
            symbol="star",
            line=dict(color="#ffffff", width=0.8),
        ),
        hovertemplate=(
            "<b>EXPLORATION PICK</b><br>"
            "Bill: %{customdata[0]}<br>"
            "Fraud Score: %{x:.3f}<br>"
            "Uncertainty: %{y:.3f}<br>"
            "Risk Area: %{customdata[1]}<br>"
            "<extra></extra>"
        ),
        customdata=exp_df[["bill_id", "risk_area"]].values,
    ))

    # Vertical line at 0.5
    fig1.add_vline(x=0.5, line=dict(color="#CC77FF", dash="dash", width=1.2),
                   annotation_text="Max uncertainty (0.5)",
                   annotation_font_color="#CC77FF",
                   annotation_position="top right")

    # Shaded exploration zone (0.3 – 0.7)
    fig1.add_vrect(x0=0.3, x1=0.7,
                   fillcolor="#CC77FF", opacity=0.05,
                   line_width=0,
                   annotation_text="Exploration zone",
                   annotation_font_color="#CC77FF",
                   annotation_position="top left")

    fig1.update_layout(**base_layout(
        "Fraud Score vs Uncertainty Score  |  unc_i = −1.8 × |ŷᶜˡˢ − 0.5| + 1",
        "Fraud Score ŷᶜˡˢ  (DATE exploitation output)",
        "Uncertainty Score unc_i",
        h=480,
    ))
    st.plotly_chart(fig1, use_container_width=True)

    # ─────────────────────────────────────────────────────────────
    # SCATTER 2 — uncertainty_score vs log(pred_revenue) = Scale Factor S_i
    # ─────────────────────────────────────────────────────────────
    st.markdown('<div class="section-title">② Uncertainty vs log(Revenue) — The S_i Scale Factor Space</div>',
                unsafe_allow_html=True)
    st.markdown("""<div class="alert-blue">
      <b>S_i = unc_i × log(ŷʳᵉᵛ + ε)</b> — bills in the
      <b style="color:#CC77FF;">top-right quadrant</b> score highest and are
      the primary exploration candidates: they are uncertain <i>and</i> financially significant.
      Bottom-left = low uncertainty low revenue = safe GREEN facilitation.
    </div>""", unsafe_allow_html=True)

    epsilon = 1e-6
    df_plot = df.copy()
    df_plot["log_rev"] = np.log(df_plot["pred_revenue"] + epsilon)

    fig2 = go.Figure()

    for ch in ["GREEN", "YELLOW", "RED"]:
        sub     = df_plot[(df_plot["channel"] == ch) & (df_plot["is_exploration"] == 0)]
        fig2.add_trace(go.Scatter(
            x=sub["log_rev"],
            y=sub["uncertainty_score"],
            mode="markers",
            name=f"{ch}",
            marker=dict(
                color=ch_color[ch], size=5, opacity=0.40,
                symbol=ch_marker[ch], line=dict(width=0),
            ),
            hovertemplate=(
                f"<b>{ch}</b><br>"
                "log(Revenue): %{x:.2f}<br>"
                "Uncertainty: %{y:.3f}<br>"
                "Scale Factor S_i: %{customdata:.3f}<br>"
                "<extra></extra>"
            ),
            customdata=sub["scale_factor"],
        ))

    # Exploration picks
    exp_plot = df_plot[df_plot["is_exploration"] == 1]
    fig2.add_trace(go.Scatter(
        x=exp_plot["log_rev"],
        y=exp_plot["uncertainty_score"],
        mode="markers",
        name="🔍 Exploration Picks",
        marker=dict(color="#CC77FF", size=12, opacity=0.95,
                    symbol="star", line=dict(color="#fff", width=0.8)),
        hovertemplate=(
            "<b>EXPLORATION PICK</b><br>"
            "Bill: %{customdata[0]}<br>"
            "log(Rev): %{x:.2f}<br>"
            "Uncertainty: %{y:.3f}<br>"
            "S_i: %{customdata[1]:.3f}<br>"
            "Risk Area: %{customdata[2]}<br>"
            "<extra></extra>"
        ),
        customdata=exp_plot[["bill_id", "scale_factor", "risk_area"]].values,
    ))

    # Fraud detected overlay
    fraud_plot = df_plot[df_plot["inspection_outcome"] == "FRAUD_DETECTED"]
    fig2.add_trace(go.Scatter(
        x=fraud_plot["log_rev"],
        y=fraud_plot["uncertainty_score"],
        mode="markers",
        name="🚨 Fraud Detected",
        marker=dict(color="#FF4466", size=8, opacity=0.85,
                    symbol="x", line=dict(color="#FF4466", width=1.5)),
        hovertemplate=(
            "<b>FRAUD DETECTED</b><br>"
            "Bill: %{customdata[0]}<br>"
            "Channel: %{customdata[1]}<br>"
            "log(Rev): %{x:.2f}<br>"
            "Uncertainty: %{y:.3f}<br>"
            "<extra></extra>"
        ),
        customdata=fraud_plot[["bill_id", "channel"]].values,
    ))

    # Quadrant annotations
    x_med = df_plot["log_rev"].median()
    y_med = df_plot["uncertainty_score"].median()
    fig2.add_hline(y=y_med, line=dict(color="#1E3A6E", dash="dash", width=1))
    fig2.add_vline(x=x_med, line=dict(color="#1E3A6E", dash="dash", width=1))

    for (xa, ya, txt, col) in [
        (0.98, 0.98, "HIGH S_i<br>Explore", "#CC77FF"),
        (0.02, 0.98, "Uncertain<br>Low Rev",  "#F5A800"),
        (0.98, 0.02, "High Rev<br>Confident", "#0066CC"),
        (0.02, 0.02, "LOW S_i<br>Facilitate", "#00843D"),
    ]:
        fig2.add_annotation(
            xref="paper", yref="paper", x=xa, y=ya,
            text=f"<b>{txt}</b>", showarrow=False,
            font=dict(color=col, size=10),
            bgcolor=col+"18", bordercolor=col,
            borderwidth=1, borderpad=4,
        )

    fig2.update_layout(**base_layout(
        "Uncertainty Score vs log(Revenue)  |  S_i = unc_i × log(ŷʳᵉᵛ + ε)",
        "log(Predicted Revenue ŷʳᵉᵛ)",
        "Uncertainty Score unc_i",
        h=500,
    ))
    st.plotly_chart(fig2, use_container_width=True)

    # ─────────────────────────────────────────────────────────────
    # SCATTER 3 — Scale Factor S_i vs fraud_score, sized by revenue
    # ─────────────────────────────────────────────────────────────
    st.markdown('<div class="section-title">③ Scale Factor S_i vs Fraud Score — Full bATE Decision Space</div>',
                unsafe_allow_html=True)
    st.markdown("""<div class="alert-blue">
      The <b>scale factor S_i</b> (y-axis) is what bATE actually ranks bills by.
      <b>Bubble size = predicted revenue.</b>
      Exploration picks (⭐) cluster at <b>mid fraud-score × high S_i</b> — uncertain AND valuable.
      Exploitation picks cluster at <b>high fraud-score</b> regardless of uncertainty.
    </div>""", unsafe_allow_html=True)

    # Normalise bubble size
    rev_norm = ((df["pred_revenue"] - df["pred_revenue"].min()) /
                (df["pred_revenue"].max() - df["pred_revenue"].min() + 1e-6))
    df_plot2 = df.copy()
    df_plot2["bubble"] = 5 + rev_norm * 22

    fig3 = go.Figure()

    # Non-selected bills (GREEN channel, not exploration)
    grn = df_plot2[(df_plot2["channel"] == "GREEN")]
    fig3.add_trace(go.Scatter(
        x=grn["fraud_score"], y=grn["scale_factor"],
        mode="markers",
        name="🟢 GREEN (Facilitated)",
        marker=dict(color="#00843D", size=grn["bubble"], opacity=0.25,
                    line=dict(width=0)),
        hovertemplate=(
            "<b>GREEN</b><br>Fraud Score: %{x:.3f}<br>"
            "S_i: %{y:.3f}<br>Revenue: $%{customdata:,.0f}<extra></extra>"
        ),
        customdata=grn["pred_revenue"],
    ))

    # Exploitation picks (RED + YELLOW, not exploration)
    expl_picks = df_plot2[(df_plot2["channel"].isin(["RED","YELLOW"])) &
                          (df_plot2["is_exploration"] == 0)]
    for ch in ["YELLOW", "RED"]:
        sub = expl_picks[expl_picks["channel"] == ch]
        fig3.add_trace(go.Scatter(
            x=sub["fraud_score"], y=sub["scale_factor"],
            mode="markers",
            name=f"{'🔴' if ch=='RED' else '🟡'} {ch} (Exploitation)",
            marker=dict(color=ch_color[ch], size=sub["bubble"],
                        opacity=0.65, symbol=ch_marker[ch],
                        line=dict(color=ch_color[ch], width=0.5)),
            hovertemplate=(
                f"<b>{ch} — EXPLOITATION</b><br>"
                "Fraud Score: %{x:.3f}<br>"
                "S_i: %{y:.3f}<br>"
                "Revenue: $%{customdata[0]:,.0f}<br>"
                "Bill: %{customdata[1]}<br>"
                "<extra></extra>"
            ),
            customdata=sub[["pred_revenue", "bill_id"]].values,
        ))

    # Exploration picks
    ep = df_plot2[df_plot2["is_exploration"] == 1]
    fig3.add_trace(go.Scatter(
        x=ep["fraud_score"], y=ep["scale_factor"],
        mode="markers",
        name="🔍 gATE Exploration Picks",
        marker=dict(color="#CC77FF", size=ep["bubble"] + 4,
                    opacity=1.0, symbol="star",
                    line=dict(color="#ffffff", width=1)),
        hovertemplate=(
            "<b>EXPLORATION PICK (gATE)</b><br>"
            "Bill: %{customdata[0]}<br>"
            "Risk Area: %{customdata[1]}<br>"
            "Fraud Score: %{x:.3f}<br>"
            "unc_i: %{customdata[2]:.3f}<br>"
            "S_i: %{y:.3f}<br>"
            "Pred Revenue: $%{customdata[3]:,.0f}<br>"
            "<extra></extra>"
        ),
        customdata=ep[["bill_id", "risk_area",
                        "uncertainty_score", "pred_revenue"]].values,
    ))

    # Fraud detected X marks
    fd = df_plot2[df_plot2["inspection_outcome"] == "FRAUD_DETECTED"]
    fig3.add_trace(go.Scatter(
        x=fd["fraud_score"], y=fd["scale_factor"],
        mode="markers",
        name="🚨 Fraud Detected",
        marker=dict(color="#FF4466", size=10, opacity=0.9,
                    symbol="x-thin", line=dict(color="#FF4466", width=2)),
        hovertemplate=(
            "<b>FRAUD CONFIRMED</b><br>"
            "Bill: %{customdata[0]}<br>"
            "Channel: %{customdata[1]}<br>"
            "Revenue Recovered: $%{customdata[2]:,.0f}<br>"
            "<extra></extra>"
        ),
        customdata=fd[["bill_id", "channel", "detected_revenue"]].values,
    ))

    # Threshold line — top n% by S_i (exploration selection boundary)
    n_explore = ep.shape[0]
    if n_explore > 0:
        si_threshold = df_plot2["scale_factor"].nlargest(n_explore * 3).min()
        fig3.add_hline(
            y=si_threshold,
            line=dict(color="#CC77FF", dash="dash", width=1.3),
            annotation_text=f"  gATE S_i candidate threshold",
            annotation_font_color="#CC77FF",
            annotation_position="right",
        )

    fig3.update_layout(**base_layout(
        "Scale Factor S_i vs Fraud Score  |  Bubble Size = Predicted Revenue",
        "Fraud Score ŷᶜˡˢ  (DATE output — Exploitation axis)",
        "Scale Factor S_i = unc_i × log(ŷʳᵉᵛ + ε)  (Exploration axis)",
        h=520,
    ))
    st.plotly_chart(fig3, use_container_width=True)

    # ─────────────────────────────────────────────────────────────
    # SCATTER 4 — Per risk area facet: fraud_score vs uncertainty_score
    # ─────────────────────────────────────────────────────────────
    st.markdown('<div class="section-title">④ Per Risk Area — Exploration Picks in Context</div>',
                unsafe_allow_html=True)
    st.markdown("""<div class="alert-blue">
      Same axes as Chart ① but broken down by risk area.
      Shows <b>which risk areas have the most uncertain bills</b>
      and whether exploration picks are concentrated in the right zone.
    </div>""", unsafe_allow_html=True)

    area_names = list(RISK_AREAS.keys())
    cols_fa = st.columns(5)
    for i, (area_name, area_cfg) in enumerate(RISK_AREAS.items()):
        a_df = df[df["risk_area"] == area_name]
        color = area_cfg["color"]
        exp_a = a_df[a_df["is_exploration"] == 1]

        fig_fa = go.Figure()
        # All bills in area
        fig_fa.add_trace(go.Scatter(
            x=a_df["fraud_score"], y=a_df["uncertainty_score"],
            mode="markers",
            name="All Bills",
            marker=dict(color=color, size=5, opacity=0.35,
                        line=dict(width=0)),
            showlegend=False,
        ))
        # Exploration picks
        if len(exp_a) > 0:
            fig_fa.add_trace(go.Scatter(
                x=exp_a["fraud_score"], y=exp_a["uncertainty_score"],
                mode="markers",
                name="Exploration",
                marker=dict(color="#CC77FF", size=10, opacity=0.95,
                            symbol="star",
                            line=dict(color="#fff", width=0.6)),
                showlegend=False,
            ))
        # Fraud
        fd_a = a_df[a_df["inspection_outcome"] == "FRAUD_DETECTED"]
        if len(fd_a) > 0:
            fig_fa.add_trace(go.Scatter(
                x=fd_a["fraud_score"], y=fd_a["uncertainty_score"],
                mode="markers",
                name="Fraud",
                marker=dict(color="#FF4466", size=7, opacity=0.85,
                            symbol="x-thin",
                            line=dict(color="#FF4466", width=1.5)),
                showlegend=False,
            ))

        fig_fa.add_vline(x=0.5, line=dict(color="#CC77FF33", width=1))
        fig_fa.update_layout(
            paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
            font=dict(color="#D0DCF0", size=9), height=240,
            title=dict(
                text=f"<b>{area_cfg['icon']} {area_name.split('/')[0]}</b>",
                font=dict(color=color, size=11), x=0.5,
            ),
            xaxis=dict(title="Fraud Score", gridcolor="#1E3A6E",
                       range=[0,1], tickfont=dict(size=8)),
            yaxis=dict(title="unc_i", gridcolor="#1E3A6E",
                       range=[0,1.1], tickfont=dict(size=8)),
            margin=dict(l=35, r=8, t=40, b=35),
            showlegend=False,
        )
        with cols_fa[i]:
            st.plotly_chart(fig_fa, use_container_width=True)
            n_exp = len(exp_a)
            n_tot = len(a_df)
            n_fd  = len(fd_a)
            st.markdown(f"""
            <div style="background:#0F1C35;border:1px solid {color};
                        border-radius:8px;padding:10px;font-size:11px;
                        text-align:center;margin-top:-10px;">
              <span style="color:{color};">Total:</span> {n_tot}<br/>
              <span style="color:#CC77FF;">⭐ Explore:</span> {n_exp}<br/>
              <span style="color:#FF4466;">🚨 Fraud:</span> {n_fd}<br/>
              <span style="color:#C8A951;">Avg S_i:</span>
              {a_df['scale_factor'].mean():.2f}
            </div>""", unsafe_allow_html=True)

    # ─────────────────────────────────────────────────────────────
    # Summary insight box
    # ─────────────────────────────────────────────────────────────
    st.markdown("<br/>", unsafe_allow_html=True)
    exp_df_s = df[df["is_exploration"] == 1]
    avg_unc  = exp_df_s["uncertainty_score"].mean()
    avg_si   = exp_df_s["scale_factor"].mean()
    avg_fs   = exp_df_s["fraud_score"].mean()
    all_avg_unc = df["uncertainty_score"].mean()
    all_avg_si  = df["scale_factor"].mean()
    all_avg_fs  = df["fraud_score"].mean()

    st.markdown(f"""
    <div class="wco-card-gold">
    <h3>📐 Exploration Selection Summary — Formula Validation</h3>
    <div style="display:grid;grid-template-columns:repeat(3,1fr);gap:16px;margin-top:8px;">

      <div style="background:#111D30;border-radius:8px;padding:14px;text-align:center;">
        <div style="color:#6B85AA;font-size:11px;text-transform:uppercase;
                    letter-spacing:0.07em;margin-bottom:6px;">Avg Fraud Score ŷᶜˡˢ</div>
        <div style="color:#CC77FF;font-size:22px;font-weight:700;
                    font-family:'Georgia',serif;">{avg_fs:.3f}</div>
        <div style="color:#6B85AA;font-size:11px;">Exploration picks</div>
        <div style="color:#44CC88;font-size:18px;font-weight:700;margin-top:4px;">{all_avg_fs:.3f}</div>
        <div style="color:#6B85AA;font-size:11px;">All bills (baseline)</div>
        <div style="color:#C8A951;font-size:11px;margin-top:6px;">
          {'✅ Closer to 0.5 = correct' if abs(avg_fs-0.5) < abs(all_avg_fs-0.5) else '↔ Similar to baseline'}
        </div>
      </div>

      <div style="background:#111D30;border-radius:8px;padding:14px;text-align:center;">
        <div style="color:#6B85AA;font-size:11px;text-transform:uppercase;
                    letter-spacing:0.07em;margin-bottom:6px;">Avg Uncertainty unc_i</div>
        <div style="color:#CC77FF;font-size:22px;font-weight:700;
                    font-family:'Georgia',serif;">{avg_unc:.3f}</div>
        <div style="color:#6B85AA;font-size:11px;">Exploration picks</div>
        <div style="color:#44CC88;font-size:18px;font-weight:700;margin-top:4px;">{all_avg_unc:.3f}</div>
        <div style="color:#6B85AA;font-size:11px;">All bills (baseline)</div>
        <div style="color:#C8A951;font-size:11px;margin-top:6px;">
          {'✅ Higher unc = targeting blind spots' if avg_unc > all_avg_unc else '↔ Check exploration ratio'}
        </div>
      </div>

      <div style="background:#111D30;border-radius:8px;padding:14px;text-align:center;">
        <div style="color:#6B85AA;font-size:11px;text-transform:uppercase;
                    letter-spacing:0.07em;margin-bottom:6px;">Avg Scale Factor S_i</div>
        <div style="color:#CC77FF;font-size:22px;font-weight:700;
                    font-family:'Georgia',serif;">{avg_si:.3f}</div>
        <div style="color:#6B85AA;font-size:11px;">Exploration picks</div>
        <div style="color:#44CC88;font-size:18px;font-weight:700;margin-top:4px;">{all_avg_si:.3f}</div>
        <div style="color:#6B85AA;font-size:11px;">All bills (baseline)</div>
        <div style="color:#C8A951;font-size:11px;margin-top:6px;">
          {'✅ Higher S_i = uncertain + valuable' if avg_si > all_avg_si else '↔ Revenue weighting check'}
        </div>
      </div>

    </div>
    </div>""", unsafe_allow_html=True)


def show():
    inject_global_css()
    page_header("📊", "Simulation Results & Analysis",
                "RULE HIT TABLES · OFFENCE DATABASE GROWTH · EXPLORATION GAINS · WEIGHT EVOLUTION")

    if "sim_df" not in st.session_state:
        st.markdown("""
        <div style="background:#0F1C35;border:2px dashed #1E3A6E;border-radius:14px;
                    padding:60px;text-align:center;margin-top:30px;">
          <div style="font-size:40px;margin-bottom:12px;">⚠️</div>
          <div style="color:#F5A800;font-size:18px;font-family:'Playfair Display',serif;">
            No Simulation Data Found
          </div>
          <div style="color:#6B85AA;margin-top:10px;">
            Please run the simulation on <b>Page 3</b> first.
          </div>
        </div>""", unsafe_allow_html=True)
        return

    df = st.session_state.sim_df
    updated_weights = st.session_state.get("sim_weights", {})
    orig_weights    = st.session_state.get("rule_weights", {})
    efficiency      = st.session_state.get("sim_efficiency", {})

    # KPI strip
    fraud_total  = (df["inspection_outcome"] == "FRAUD_DETECTED").sum()
    offence_adds = df["added_to_offence_db"].sum()
    explore_finds = df[(df["is_exploration"]==1) & (df["added_to_offence_db"]==1)].shape[0]
    revenue      = df["detected_revenue"].sum()
    metric_row([
        (fraud_total,              "Frauds Detected",         WCO_RED),
        (offence_adds,             "Offence DB Additions",    WCO_GOLD),
        (explore_finds,            "Exploration Discoveries", "#CC77FF"),
        (f"${revenue:,.0f}",       "Revenue Recovered",       WCO_GREEN),
        (f"{efficiency.get('hybrid',{}).get('efficiency_index',0):.3f}",
                                   "Efficiency Index",        WCO_GOLD),
    ])

    tabs = st.tabs([
        "📋 T1: Rule Hits", "📡 T2: Channel Stats",
        "🔍 T3: Exploration Finds", "🗄️ T4: Offence DB",
        "⚖️ T5: Risk Scoring", "📈 Charts",
        "🧭 Exploration Logic",
    ])

    # ── TABLE 1: Rule Hits ────────────────────────────────────────
    with tabs[0]:
        st.markdown("""<div class="alert-blue">
          How many bills were hit by each risk rule, and precision of each rule.
        </div>""", unsafe_allow_html=True)
        df_t1 = build_rule_hit_table(df)

        # Mini bar chart
        fig_t1 = go.Figure(go.Bar(
            x=df_t1["Rule ID"], y=df_t1["Total Bills Hit"],
            marker_color=df_t1["Risk Area"].map(
                {k: v["color"] for k, v in RISK_AREAS.items()}),
            opacity=0.85,
            text=df_t1["Precision (%)"].apply(lambda x: f"{x}%"), textposition="outside",
            textfont=dict(color="#D0DCF0", size=10),
            hovertext=df_t1["Rule Name"], hoverinfo="text+y",
        ))
        fig_t1.update_layout(
            paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
            font=dict(color="#D0DCF0", size=10), height=280,
            title=dict(text="<b>Bills Hit per Rule (label = detection precision)</b>",
                       font=dict(color=WCO_GOLD, size=13), x=0.5),
            xaxis=dict(gridcolor="#1E3A6E", tickangle=-45),
            yaxis=dict(gridcolor="#1E3A6E", title="Bills Hit"),
            margin=dict(l=45, r=15, t=45, b=75),
        )
        st.plotly_chart(fig_t1, use_container_width=True)
        styled_table(df_t1, "Table 1 — Risk Rule Hit Analysis")

    # ── TABLE 2: Channel Stats ────────────────────────────────────
    with tabs[1]:
        st.markdown("""<div class="alert-blue">
          Bills treated per channel, exploitation vs exploration split, fraud detection per channel.
        </div>""", unsafe_allow_html=True)
        df_t2 = build_channel_table(df)

        c1, c2 = st.columns(2)
        with c1:
            fig_ch = go.Figure(go.Bar(
                x=df_t2["Channel"],
                y=df_t2["Total Bills"],
                marker_color=["#C8102E","#F5A800","#00843D"], opacity=0.85,
                text=df_t2["Total Bills"], textposition="outside",
                textfont=dict(color="#D0DCF0"),
            ))
            fig_ch.update_layout(
                paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
                font=dict(color="#D0DCF0", size=12), height=300,
                title=dict(text="<b>Bills per Channel</b>",
                           font=dict(color=WCO_GOLD, size=13), x=0.5),
                xaxis=dict(gridcolor="#1E3A6E"),
                yaxis=dict(gridcolor="#1E3A6E"),
                margin=dict(l=45, r=15, t=45, b=40),
            )
            st.plotly_chart(fig_ch, use_container_width=True)
        with c2:
            fig_pie = go.Figure(go.Pie(
                labels=["Exploitation","Exploration"],
                values=[df_t2["Exploitation Bills"].sum(), df_t2["Exploration Bills"].sum()],
                marker=dict(colors=["#0066CC","#CC77FF"],
                            line=dict(color="#070E1C", width=2)),
                hole=0.55, textfont=dict(color="#D0DCF0"),
            ))
            fig_pie.update_layout(
                paper_bgcolor="#070E1C",
                font=dict(color="#D0DCF0"), height=300,
                title=dict(text="<b>Exploit vs Explore Split</b>",
                           font=dict(color=WCO_GOLD, size=13), x=0.5),
                legend=dict(bgcolor="#0F1C35"),
                margin=dict(l=20, r=20, t=45, b=20),
            )
            st.plotly_chart(fig_pie, use_container_width=True)

        styled_table(df_t2, "Table 2 — Channel Treatment Statistics")

    # ── TABLE 3: Exploration Discoveries ─────────────────────────
    with tabs[2]:
        st.markdown("""<div class="alert-blue">
          New bills discovered and added to offence database via exploration strategy (concept drift detection).
        </div>""", unsafe_allow_html=True)
        df_t3 = build_exploration_discovery_table(df)
        st.plotly_chart(exploration_vs_exploitation_chart(df), use_container_width=True)
        styled_table(df_t3, "Table 3 — Exploration Discovery Analysis per Risk Area")

    # ── TABLE 4: Offence DB ───────────────────────────────────────
    with tabs[3]:
        st.markdown("""<div class="alert-blue">
          Real-time offence database augmentation from Red + Yellow channel inspection feedback.
        </div>""", unsafe_allow_html=True)
        st.plotly_chart(offence_db_growth_chart(df), use_container_width=True)
        df_t4 = build_offence_db_table(df)
        styled_table(df_t4, "Table 4 — Offence Database Augmentation by Risk Area")

        # Radar chart of offence DB by area
        areas = df_t4["Risk Area"].tolist()
        vals  = df_t4["Total Added"].tolist()
        fig_radar = go.Figure(go.Scatterpolar(
            r=vals + [vals[0]], theta=areas + [areas[0]],
            fill="toself", fillcolor="rgba(200,169,81,0.15)",
            line=dict(color=WCO_GOLD, width=2),
            marker=dict(size=8, color=WCO_GOLD),
        ))
        fig_radar.update_layout(
            paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
            font=dict(color="#D0DCF0", size=11), height=360,
            title=dict(text="<b>Offence DB Coverage Radar</b>",
                       font=dict(color=WCO_GOLD, size=13), x=0.5),
            polar=dict(bgcolor="#0B1220",
                       angularaxis=dict(gridcolor="#1E3A6E", linecolor="#1E3A6E"),
                       radialaxis=dict(gridcolor="#1E3A6E", linecolor="#1E3A6E")),
            margin=dict(l=60, r=60, t=55, b=40),
        )
        st.plotly_chart(fig_radar, use_container_width=True)

    # ── TABLE 5: Risk Scoring ─────────────────────────────────────
    with tabs[4]:
        st.markdown("""<div class="alert-blue">
          Per-transaction risk scores, rule hits, cumulative weights, and weight evolution after
          self-learning feedback. Shows how detection efficiency drives automatic weight uplift.
        </div>""", unsafe_allow_html=True)
        df_t5 = build_risk_score_table(df, updated_weights)

        # Weight evolution chart
        if orig_weights and updated_weights:
            st.plotly_chart(weight_evolution_chart(df, orig_weights, updated_weights),
                            use_container_width=True)

        styled_table(df_t5, "Table 5 — Transaction Risk Scoring & Weight Evolution",
                     highlight_col="Channel",
                     color_map={"RED": "#C8102E", "YELLOW": "#F5A800", "GREEN": "#00843D"})

        # Weight delta table
        st.markdown('<div class="section-title">📈 Rule Weight Update Summary</div>',
                    unsafe_allow_html=True)
        weight_rows = []
        for area_name, area_cfg in RISK_AREAS.items():
            for rule in area_cfg["rules"]:
                rid = rule["id"]
                o = orig_weights.get(rid, rule["weight"])
                u = updated_weights.get(rid, o)
                weight_rows.append({
                    "Risk Area":      area_name,
                    "Rule ID":        rid,
                    "Rule Name":      rule["name"],
                    "Original Weight":f"{o:.3f}",
                    "Updated Weight": f"{u:.3f}",
                    "Δ Weight":       f"+{u-o:.4f}" if u >= o else f"{u-o:.4f}",
                    "% Change":       f"{100*(u-o)/o:.1f}%" if o > 0 else "—",
                })
        df_wd = pd.DataFrame(weight_rows)
        styled_table(df_wd, "Rule Weight Delta After Self-Learning Cycle")

    # ── Charts summary tab ────────────────────────────────────────
    with tabs[5]:
        c1, c2 = st.columns(2)
        with c1:
            # Outcome distribution pie
            outcomes = df[df["channel"] != "GREEN"]["inspection_outcome"].value_counts()
            fig_out = go.Figure(go.Pie(
                labels=outcomes.index, values=outcomes.values,
                marker=dict(colors=[RISK_LEVEL_COLORS.get(k,"#6B85AA") for k in outcomes.index],
                            line=dict(color="#070E1C", width=2)),
                hole=0.5, textfont=dict(color="#D0DCF0"),
            ))
            fig_out.update_layout(
                paper_bgcolor="#070E1C", font=dict(color="#D0DCF0"), height=340,
                title=dict(text="<b>Inspection Outcome Distribution</b>",
                           font=dict(color=WCO_GOLD, size=13), x=0.5),
                legend=dict(bgcolor="#0F1C35"), margin=dict(l=20,r=20,t=45,b=20),
            )
            st.plotly_chart(fig_out, use_container_width=True)
        with c2:
            # Revenue by risk area
            rev_by_area = df.groupby("risk_area")["detected_revenue"].sum().reset_index()
            colors = [RISK_AREAS[a]["color"] for a in rev_by_area["risk_area"]]
            fig_rev = go.Figure(go.Bar(
                x=rev_by_area["risk_area"],
                y=rev_by_area["detected_revenue"],
                marker_color=colors, opacity=0.85,
                text=rev_by_area["detected_revenue"].apply(lambda x: f"${x:,.0f}"),
                textposition="outside", textfont=dict(color="#D0DCF0", size=10),
            ))
            fig_rev.update_layout(
                paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
                font=dict(color="#D0DCF0", size=10), height=340,
                title=dict(text="<b>Revenue Recovered by Risk Area</b>",
                           font=dict(color=WCO_GOLD, size=13), x=0.5),
                xaxis=dict(gridcolor="#1E3A6E", tickangle=-25),
                yaxis=dict(gridcolor="#1E3A6E", title="Revenue ($)"),
                margin=dict(l=50, r=15, t=45, b=80),
            )
            st.plotly_chart(fig_rev, use_container_width=True)

        # Fraud score scatter
        sample_df = df.sample(min(300, len(df)), random_state=42)
        fig_sc = px.scatter(
            sample_df, x="fraud_score", y="detected_revenue",
            color="channel", symbol="risk_area",
            color_discrete_map={"RED":"#C8102E","YELLOW":"#F5A800","GREEN":"#00843D"},
            opacity=0.7, height=400,
            labels={"fraud_score":"Fraud Score (DATE)","detected_revenue":"Revenue Detected ($)"},
            title="<b>Fraud Score vs Revenue Recovered (Sample 300 bills)</b>",
        )
        fig_sc.update_layout(
            paper_bgcolor="#070E1C", plot_bgcolor="#0B1220",
            font=dict(color="#D0DCF0", family="IBM Plex Sans"),
            title_font=dict(color=WCO_GOLD, family="Playfair Display"),
            xaxis=dict(gridcolor="#1E3A6E"), yaxis=dict(gridcolor="#1E3A6E"),
            legend=dict(bgcolor="#0F1C35", bordercolor=WCO_BORDER),
            margin=dict(l=55, r=20, t=55, b=45),
        )
        st.plotly_chart(fig_sc, use_container_width=True)

    # ── TAB 7: Exploration Logic Scatter ─────────────────────────
    with tabs[6]:
        exploration_scatter_tab(df)