Update app.py

app.py

@@ -1,3 +1,5 @@
+# app.py — Market Breadth & Momentum (sticky results, no nested expanders)
+
 import os
 import io
 import time
@@ -41,6 +43,7 @@ if "last_params" not in st.session_state:
 with st.sidebar:
     st.header("Parameters")
 
+    # Each expander is independent (no nesting).
     with st.expander("Data Window", expanded=False):
         default_start = datetime(2015, 1, 1).date()
         default_end = (datetime.today().date() + timedelta(days=1))
@@ -150,7 +153,6 @@ with st.sidebar:
             heat_last_days=int(heat_last_days),
             mom_look=int(mom_look),
         )
-        # mark that results should be shown (and re-shown on reruns)
         st.session_state.run_id = f"{time.time():.0f}"
 
     if clear_clicked:
@@ -353,15 +355,13 @@ idx_volume = idx_df["Volume"].reindex(clean_close.index).ffill()
 
 # ===================== SECTION 1 — Breadth Dashboard =====================
 st.header("Breadth Dashboard")
-# (… the rest of your original analysis/plotting code is unchanged …)
-# NOTE: everything below stays exactly the same as your original file.
 
-# ---------------------- [KEEP YOUR ORIGINAL CODE BELOW THIS LINE] ----------------------
-# 1) Methodology expander
+# Methodology (standalone expander)
 with st.expander("Methodology", expanded=False):
     # Overview
     st.write("This panel tracks trend, participation, and momentum for a broad equity universe.")
     st.write("Use it to judge trend quality, spot divergences, and gauge risk bias.")
+
     # 1) Price trend (MAs, VWAP)
     st.write("**Price trend**")
     st.latex(r"\mathrm{SMA}_{n}(t)=\frac{1}{n}\sum_{k=0}^{n-1}P_{t-k}")
@@ -369,33 +369,39 @@ with st.expander("Methodology", expanded=False):
     st.latex(r"\mathrm{VWAP}_{200w}(t)=\frac{\sum_{k=0}^{N-1}P_{t-k}V_{t-k}}{\sum_{k=0}^{N-1}V_{t-k}},\quad N\approx200\times5")
     st.write("Price above both MAs and fast>slow = strong trend.")
     st.write("Price below both MAs and fast<slow = weak trend.")
+
     # 2) Participation breadth (% above MAs)
     st.write("**Participation breadth**")
     st.write("Share above n-day MA:")
     st.latex(r"\%\,\text{Above}_n(t)=100\cdot\frac{\#\{i:\ P_{i,t}>\mathrm{SMA}_{n,i}(t)\}}{N}")
     st.write("Zones: 0–20 weak, 20–50 neutral, 50–80 strong.")
     st.write("Higher shares mean broad support for the trend.")
-    # 3) A/D line
+
+    # 3) Advance–Decline line
     st.write("**Advance–Decline (A/D) line**")
     st.latex(r"A_t=\#\{i:\ P_{i,t}>P_{i,t-1}\},\quad D_t=\#\{i:\ P_{i,t}<P_{i,t-1}\}")
     st.latex(r"\mathrm{ADLine}_t=\sum_{u\le t}(A_u-D_u)")
     st.write("Rising A/D confirms uptrends. Falling A/D warns of narrow leadership.")
+
     # 4) Net new 52-week highs
     st.write("**Net new 52-week highs**")
     st.latex(r"H_{i,t}^{52}=\max_{u\in[t-251,t]}P_{i,u},\quad L_{i,t}^{52}=\min_{u\in[t-251,t]}P_{i,u}")
     st.latex(r"\text{NewHighs}_t=\sum_i \mathbf{1}\{P_{i,t}=H_{i,t}^{52}\},\quad \text{NewLows}_t=\sum_i \mathbf{1}\{P_{i,t}=L_{i,t}^{52}\}")
     st.latex(r"\text{NetHighs}_t=\text{NewHighs}_t-\text{NewLows}_t")
     st.write("Positive and persistent net highs support trend durability.")
+
     # 5) Smoothed advancing vs declining counts
     st.write("**Advancing vs declining (smoothed)**")
     st.latex(r"\overline{A}_t=\frac{1}{w}\sum_{k=0}^{w-1}A_{t-k},\quad \overline{D}_t=\frac{1}{w}\sum_{k=0}^{w-1}D_{t-k}")
     st.write("Advancers > decliners over the window = constructive breadth.")
+
     # 6) McClellan Oscillator
     st.write("**McClellan Oscillator (MO)**")
     st.latex(r"E^{(n)}_t=\text{EMA}_n(A_t-D_t)")
     st.latex(r"\mathrm{MO}_t=E^{(19)}_t-E^{(39)}_t")
     st.write("Zero-line up-cross = improving momentum. Down-cross = fading momentum.")
     st.write("A 9-day EMA of MO can act as a signal line.")
+
     # Practical reads
     st.write("**Practical use**")
     st.write("- Broad strength: % above 200-day ≥ 50% supports trends.")
@@ -546,75 +552,94 @@ fig.update_layout(
 )
 st.plotly_chart(fig, use_container_width=True)
 
-# === Dynamic Interpretation (unchanged) ===
+# --- Dynamic interpretation (standalone expander) ---
 with st.expander("Dynamic Interpretation", expanded=False):
     buf = io.StringIO()
+
     def _last_val(s):
         s = s.dropna()
         return s.iloc[-1] if len(s) else np.nan
+
     def _last_date(s):
         s = s.dropna()
         return s.index[-1] if len(s) else None
+
     def _pct(a, b):
         if not np.isfinite(a) or not np.isfinite(b) or b == 0:
             return np.nan
         return (a - b) / b * 100.0
+
     def _fmt_pct(x):
         return "n/a" if not np.isfinite(x) else f"{x:.1f}%"
+
     def _fmt_num(x):
         return "n/a" if not np.isfinite(x) else f"{x:,.2f}"
 
     as_of = _last_date(idx)
+
     px      = _last_val(idx)
     ma50    = _last_val(sma_fast_idx)
     ma200   = _last_val(sma_slow_idx)
     vwap200 = _last_val(vwap_idx)
+
     p50  = float(_last_val(pct_above_fast))
     p200 = float(_last_val(pct_above_slow))
+
     ad_now  = _last_val(ad_line)
     nh_now  = int(_last_val(new_highs)) if np.isfinite(_last_val(new_highs)) else 0
     nh_sma  = float(_last_val(sma10_net_hi))
+
     avg_adv_last  = float(_last_val(avg_adv))
     avg_decl_last = float(_last_val(avg_decl))
+
     _ema19     = net_adv.ewm(span=int(mo_span_fast), adjust=False).mean()
     _ema39     = net_adv.ewm(span=int(mo_span_slow), adjust=False).mean()
     mc_osc2    = (_ema19 - _ema39).rename("MO")
     mc_signal  = mc_osc2.ewm(span=int(mo_signal_span), adjust=False).mean().rename("Signal")
+
     mo_last     = float(_last_val(mc_osc2))
     mo_prev     = float(_last_val(mc_osc2.shift(1)))
     mo_5ago     = float(_last_val(mc_osc2.shift(5)))
     mo_slope5   = mo_last - mo_5ago
     mo_sig_last = float(_last_val(mc_signal))
     mo_sig_prev = float(_last_val(mc_signal.shift(1)))
+
     mo_roll   = mc_osc2.rolling(252, min_periods=126)
     mo_mean   = mo_roll.mean()
     mo_std    = mo_roll.std()
     mo_z      = (mc_osc2 - mo_mean) / mo_std
     mo_z_last = float(_last_val(mo_z))
+
     mo_abs = np.abs(mc_osc2.dropna())
     if len(mo_abs) >= 20:
         mo_ext = float(np.nanpercentile(mo_abs.tail(252), 90))
     else:
         mo_ext = np.nan
+
     look_fast = 10
     look_mid  = 20
     look_div  = 63
+
     ma50_slope  = _last_val(sma_fast_idx.diff(look_fast))
     ma200_slope = _last_val(sma_slow_idx.diff(look_mid))
     p50_chg     = p50  - float(_last_val(pct_above_fast.shift(look_fast)))
     p200_chg    = p200 - float(_last_val(pct_above_slow.shift(look_fast)))
     ad_mom      = ad_now - float(_last_val(ad_line.shift(look_mid)))
+
     d50   = _pct(px, ma50)
     d200  = _pct(px, ma200)
     dvw   = _pct(px, vwap200)
     h63   = float(_last_val(idx.rolling(look_div).max()))
     dd63  = _pct(px, h63) if np.isfinite(h63) else np.nan
+
     ad_63h = float(_last_val(ad_line.rolling(look_div).max()))
     mo_63h = float(_last_val(mc_osc2.rolling(look_div).max()))
     near_high_px = np.isfinite(h63) and np.isfinite(px) and px >= 0.995 * h63
     near_high_ad = np.isfinite(ad_63h) and np.isfinite(ad_now) and ad_now >= 0.995 * ad_63h
     near_high_mo = np.isfinite(mo_63h) and np.isfinite(mo_last) and mo_last >= 0.95 * mo_63h
+
     breadth_thrust = (p50 >= 55) and (p50_chg >= 20)
+
     score = 0
     score += 1 if px > ma50 else 0
     score += 1 if px > ma200 else 0
@@ -626,6 +651,7 @@ with st.expander("Dynamic Interpretation", expanded=False):
     score += 1 if nh_now > 0 and nh_sma >= 0 else 0
     score += 1 if avg_adv_last > avg_decl_last else 0
     score += 1 if (mo_last > 0 and mo_slope5 > 0) else 0
+
     if score >= 8:
         regime = "Risk-on bias"
     elif score >= 5:
@@ -634,6 +660,7 @@ with st.expander("Dynamic Interpretation", expanded=False):
         regime = "Risk-off bias"
 
     print(f"=== Market breadth narrative — {as_of.date() if as_of is not None else 'N/A'} ===", file=buf)
+
     # [Trend]
     print("\n[Trend]", file=buf)
     if np.isfinite(px) and np.isfinite(ma50) and np.isfinite(ma200):
@@ -802,557 +829,560 @@ with st.expander("Dynamic Interpretation", expanded=False):
 
     st.text(buf.getvalue())
 
-    # ===================== SECTION 2 — Rebased Comparison =====================
-    st.header("Rebased Comparison (Last N sessions)")
-
-    with st.expander("Methodology", expanded=False):
-        st.write("Compares stock paths on a common scale and highlights leadership vs laggards.")
-        st.write("Use it to judge breadth, concentration, and dispersion over the selected window.")
-
-        st.write("**Rebasing (start = B)**")
-        st.latex(r"R_{i,t}= \frac{P_{i,t}}{P_{i,t_0}}\times B")
-        st.write("Each line shows cumulative performance since the window start.")
-        st.write("The index is rebased the same way for reference.")
-
-        st.write("**Log scale**")
-        st.write("We plot the y-axis in log scale so equal percent moves look equal.")
-        st.write("Y-range uses robust bounds (1st–99th percentiles) with padding.")
-
-        st.write("**Leaders and laggards**")
-        st.latex(r"\text{Perf}_{i}=R_{i,T}")
-        st.write("Leaders are highest Perf at T. Laggards are lowest.")
-        st.write("MAG7 are highlighted if present.")
-
-        st.write("**Equal-weight summaries**")
-        st.latex(r"\text{EWAvg}_T=\frac{1}{M}\sum_{i=1}^{M}R_{i,T}")
-        st.latex(r"\text{Median}_T=\operatorname{median}\{R_{i,T}\}")
-        st.latex(r"\%\text{Up}_T=100\cdot \frac{1}{M}\sum_{i=1}^{M}\mathbf{1}[R_{i,T}>B]")
-        st.latex(r"\%\text{BeatIdx}_T=100\cdot \frac{1}{M}\sum_{i=1}^{M}\mathbf{1}[R_{i,T}>R_{\text{idx},T}]")
-        st.write("These give a breadth read relative to the index and to flat (B).")
-
-        st.write("**Dispersion (cross-section)**")
-        st.latex(r"\sigma_T=\operatorname{stdev}\{R_{i,T}\},\quad \text{IQR}_T=Q_{0.75}-Q_{0.25}")
-        st.write("High dispersion means large performance spread across names.")
-
-        st.write("**Concentration (top N share of gains)**")
-        st.latex(r"\text{TopNShare}_T=\frac{\sum_{i\in \text{Top}N}(R_{i,T}-B)}{\sum_{j=1}^{M}(R_{j,T}-B)}\times 100")
-        st.write("Large TopNShare implies leadership is concentrated.")
-
-        st.write("**Correlation to index (optional diagnostic)**")
-        st.latex(r"\rho_i=\operatorname{corr}\big(\Delta \ln P_{i,t},\, \Delta \ln P_{\text{idx},t}\big)")
-        st.write("Lower median correlation favors stock picking. High correlation means beta drives moves.")
-
-        st.write("**Practical reads**")
-        st.write("- Broad advance: many lines above the index and %BeatIdx high.")
-        st.write("- Concentration risk: TopNShare large while most lines trail the index.")
-        st.write("- Rotation/dispersion: high cross-section std and lower median correlation.")
-        st.write("- Leadership quality: leaders holding gains on a log scale with limited drawdowns.")
-
-    n_days = int(rebase_days)
-    base = float(rebase_base)
-
-    recent = clean_close.iloc[-n_days:].dropna(axis=1, how="any")
-    if recent.empty:
-        st.warning("Not enough overlapping history for the rebased comparison window.")
-    else:
-        first = recent.iloc[0]
-        mask = (first > 0) & np.isfinite(first)
-        rebased = (recent.loc[:, mask] / first[mask]) * base
-
-        perf = rebased.iloc[-1].dropna()
-        mag7_all = ["AAPL","MSFT","AMZN","META","GOOGL","NVDA","TSLA"]
-        mag7 = [t for t in mag7_all if t in rebased.columns]
-        non_mag = perf.drop(index=mag7, errors="ignore")
-        top5 = non_mag.nlargest(min(5, len(non_mag))).index.tolist()
-        worst5 = non_mag.nsmallest(min(5, len(non_mag))).index.tolist()
-
-        mag_colors = {
-            "AAPL":"#00bfff","MSFT":"#3cb44b","AMZN":"#ffe119",
-            "META":"#4363d8","GOOGL":"#f58231","NVDA":"#911eb4","TSLA":"#46f0f0"
-        }
-
-        spx = idx.reindex(rebased.index).dropna()
-        spx_rebased = spx / spx.iloc[0] * base
-
-        def hover_tmpl(name: str) -> str:
-            return "%{y:.2f}<br>%{x|%Y-%m-%d}<extra>" + name + "</extra>"
-
-        fig2 = go.Figure()
-        for t in rebased.columns:
-            fig2.add_trace(go.Scatter(
-                x=rebased.index, y=rebased[t], name=t, mode="lines",
-                line=dict(width=1, color="rgba(160,160,160,0.4)"),
-                hovertemplate=hover_tmpl(t), showlegend=False
-            ))
-        for t in mag7:
-            fig2.add_trace(go.Scatter(
-                x=rebased.index, y=rebased[t], name=t, mode="lines",
-                line=dict(width=2, color=mag_colors.get(t, "#ffffff")),
-                hovertemplate=hover_tmpl(t)
-            ))
-        for t in top5:
-            fig2.add_trace(go.Scatter(
-                x=rebased.index, y=rebased[t], name=f"Top {t}", mode="lines",
-                line=dict(width=2, color="lime"),
-                hovertemplate=hover_tmpl(t), showlegend=False
-            ))
-        for t in worst5:
-            fig2.add_trace(go.Scatter(
-                x=rebased.index, y=rebased[t], name=f"Worst {t}", mode="lines",
-                line=dict(width=2, color="red", dash="dash"),
-                hovertemplate=hover_tmpl(t), showlegend=False
-            ))
+# ===================== SECTION 2 — Rebased Comparison =====================
+st.header("Rebased Comparison (Last N sessions)")
+
+# Methodology (standalone expander)
+with st.expander("Methodology", expanded=False):
+    st.write("Compares stock paths on a common scale and highlights leadership vs laggards.")
+    st.write("Use it to judge breadth, concentration, and dispersion over the selected window.")
+
+    st.write("**Rebasing (start = B)**")
+    st.latex(r"R_{i,t}= \frac{P_{i,t}}{P_{i,t_0}}\times B")
+    st.write("Each line shows cumulative performance since the window start.")
+    st.write("The index is rebased the same way for reference.")
+
+    st.write("**Log scale**")
+    st.write("We plot the y-axis in log scale so equal percent moves look equal.")
+    st.write("Y-range uses robust bounds (1st–99th percentiles) with padding.")
+
+    st.write("**Leaders and laggards**")
+    st.latex(r"\text{Perf}_{i}=R_{i,T}")
+    st.write("Leaders are highest Perf at T. Laggards are lowest.")
+    st.write("MAG7 are highlighted if present.")
+
+    st.write("**Equal-weight summaries**")
+    st.latex(r"\text{EWAvg}_T=\frac{1}{M}\sum_{i=1}^{M}R_{i,T}")
+    st.latex(r"\text{Median}_T=\operatorname{median}\{R_{i,T}\}")
+    st.latex(r"\%\text{Up}_T=100\cdot \frac{1}{M}\sum_{i=1}^{M}\mathbf{1}[R_{i,T}>B]")
+    st.latex(r"\%\text{BeatIdx}_T=100\cdot \frac{1}{M}\sum_{i=1}^{M}\mathbf{1}[R_{i,T}>R_{\text{idx},T}]")
+    st.write("These give a breadth read relative to the index and to flat (B).")
+
+    st.write("**Dispersion (cross-section)**")
+    st.latex(r"\sigma_T=\operatorname{stdev}\{R_{i,T}\},\quad \text{IQR}_T=Q_{0.75}-Q_{0.25}")
+    st.write("High dispersion means large performance spread across names.")
+
+    st.write("**Concentration (top N share of gains)**")
+    st.latex(r"\text{TopNShare}_T=\frac{\sum_{i\in \text{Top}N}(R_{i,T}-B)}{\sum_{j=1}^{M}(R_{j,T}-B)}\times 100")
+    st.write("Large TopNShare implies leadership is concentrated.")
+
+    st.write("**Correlation to index (optional diagnostic)**")
+    st.latex(r"\rho_i=\operatorname{corr}\big(\Delta \ln P_{i,t},\, \Delta \ln P_{\text{idx},t}\big)")
+    st.write("Lower median correlation favors stock picking. High correlation means beta drives moves.")
+
+    st.write("**Practical reads**")
+    st.write("- Broad advance: many lines above the index and %BeatIdx high.")
+    st.write("- Concentration risk: TopNShare large while most lines trail the index.")
+    st.write("- Rotation/dispersion: high cross-section std and lower median correlation.")
+    st.write("- Leadership quality: leaders holding gains on a log scale with limited drawdowns.")
+
+n_days = int(rebase_days)
+base = float(rebase_base)
+
+recent = clean_close.iloc[-n_days:].dropna(axis=1, how="any")
+if recent.empty:
+    st.warning("Not enough overlapping history for the rebased comparison window.")
+else:
+    first = recent.iloc[0]
+    mask = (first > 0) & np.isfinite(first)
+    rebased = (recent.loc[:, mask] / first[mask]) * base
+
+    perf = rebased.iloc[-1].dropna()
+    mag7_all = ["AAPL","MSFT","AMZN","META","GOOGL","NVDA","TSLA"]
+    mag7 = [t for t in mag7_all if t in rebased.columns]
+    non_mag = perf.drop(index=mag7, errors="ignore")
+    top5 = non_mag.nlargest(min(5, len(non_mag))).index.tolist()
+    worst5 = non_mag.nsmallest(min(5, len(non_mag))).index.tolist()
+
+    mag_colors = {
+        "AAPL":"#00bfff","MSFT":"#3cb44b","AMZN":"#ffe119",
+        "META":"#4363d8","GOOGL":"#f58231","NVDA":"#911eb4","TSLA":"#46f0f0"
+    }
+
+    spx = idx.reindex(rebased.index).dropna()
+    spx_rebased = spx / spx.iloc[0] * base
+
+    def hover_tmpl(name: str) -> str:
+        return "%{y:.2f}<br>%{x|%Y-%m-%d}<extra>" + name + "</extra>"
+
+    fig2 = go.Figure()
+    for t in rebased.columns:
         fig2.add_trace(go.Scatter(
-            x=spx_rebased.index, y=spx_rebased.values, name="S&P 500 (rebased)", mode="lines",
-            line=dict(width=3, color="white"), hovertemplate=hover_tmpl("S&P 500")
+            x=rebased.index, y=rebased[t], name=t, mode="lines",
+            line=dict(width=1, color="rgba(160,160,160,0.4)"),
+            hovertemplate=hover_tmpl(t), showlegend=False
         ))
+    for t in mag7:
+        fig2.add_trace(go.Scatter(
+            x=rebased.index, y=rebased[t], name=t, mode="lines",
+            line=dict(width=2, color=mag_colors.get(t, "#ffffff")),
+            hovertemplate=hover_tmpl(t)
+        ))
+    for t in top5:
+        fig2.add_trace(go.Scatter(
+            x=rebased.index, y=rebased[t], name=f"Top {t}", mode="lines",
+            line=dict(width=2, color="lime"),
+            hovertemplate=hover_tmpl(t), showlegend=False
+        ))
+    for t in worst5:
+        fig2.add_trace(go.Scatter(
+            x=rebased.index, y=rebased[t], name=f"Worst {t}", mode="lines",
+            line=dict(width=2, color="red", dash="dash"),
+            hovertemplate=hover_tmpl(t), showlegend=False
+        ))
+    fig2.add_trace(go.Scatter(
+        x=spx_rebased.index, y=spx_rebased.values, name="S&P 500 (rebased)", mode="lines",
+        line=dict(width=3, color="white"), hovertemplate=hover_tmpl("S&P 500")
+    ))
+
+    vals = pd.concat([rebased.stack(), pd.Series(spx_rebased.values, index=spx_rebased.index)])
+    vals = vals.replace([np.inf, -np.inf], np.nan).dropna()
+    vals = vals[vals > 0]
+    y_range = None
+    if len(vals) > 10:
+        qlo, qhi = vals.quantile([0.01, 0.99])
+        y_min = max(1e-2, qlo / y_pad)
+        y_max = max(y_min * 1.1, qhi * y_pad)
+        y_range = [np.log10(y_min), np.log10(y_max)]
+
+    fig2.update_yaxes(type="log", range=y_range, title=f"Rebased Price (start = {int(base)})")
+    fig2.update_xaxes(title="Date")
+    fig2.update_layout(
+        template="plotly_dark",
+        height=700,
+        margin=dict(l=60, r=30, t=70, b=90),
+        title=f"Price Level Comparison (Rebased, Log Scale) — Last {n_days} Sessions",
+        legend=dict(orientation="h", y=-0.18, yanchor="top", x=0, xanchor="left"),
+        hovermode="closest",
+        font=dict(color="white")
+    )
+    st.plotly_chart(fig2, use_container_width=True)
 
-        vals = pd.concat([rebased.stack(), pd.Series(spx_rebased.values, index=spx_rebased.index)])
-        vals = vals.replace([np.inf, -np.inf], np.nan).dropna()
-        vals = vals[vals > 0]
-        y_range = None
-        if len(vals) > 10:
-            qlo, qhi = vals.quantile([0.01, 0.99])
-            y_min = max(1e-2, qlo / y_pad)
-            y_max = max(y_min * 1.1, qhi * y_pad)
-            y_range = [np.log10(y_min), np.log10(y_max)]
-
-        fig2.update_yaxes(type="log", range=y_range, title=f"Rebased Price (start = {int(base)})")
-        fig2.update_xaxes(title="Date")
-        fig2.update_layout(
-            template="plotly_dark",
-            height=700,
-            margin=dict(l=60, r=30, t=70, b=90),
-            title=f"Price Level Comparison (Rebased, Log Scale) — Last {n_days} Sessions",
-            legend=dict(orientation="h", y=-0.18, yanchor="top", x=0, xanchor="left"),
-            hovermode="closest",
-            font=dict(color="white")
-        )
-        st.plotly_chart(fig2, use_container_width=True)
-
-        with st.expander("Dynamic Interpretation", expanded=False):
-            buf2 = io.StringIO()
+    # Dynamic Interpretation (standalone expander)
+    with st.expander("Dynamic Interpretation", expanded=False):
+        buf2 = io.StringIO()
 
-            def _fmt_pct(x):
-                return "n/a" if pd.isna(x) else f"{x:.1f}%"
+        def _fmt_pct(x):
+            return "n/a" if pd.isna(x) else f"{x:.1f}%"
 
-            def _fmt_num(x):
-                return "n/a" if pd.isna(x) else f"{x:,.2f}"
+        def _fmt_num(x):
+            return "n/a" if pd.isna(x) else f"{x:,.2f}"
 
-            if rebased.empty or spx_rebased.empty:
-                print("No data for interpretation.", file=buf2)
+        if rebased.empty or spx_rebased.empty:
+            print("No data for interpretation.", file=buf2)
+        else:
+            as_of = rebased.index[-1].date()
+            perf_last = rebased.iloc[-1].dropna()
+            spx_last = float(spx_rebased.iloc[-1])
+            n_names = len(perf_last)
+            eq_avg = float(perf_last.mean())
+            eq_med = float(perf_last.median())
+            pct_pos = float((perf_last > base).mean() * 100)
+            pct_beat = float((perf_last > spx_last).mean() * 100)
+            disp_std = float(perf_last.std(ddof=0))
+            iqr_lo, iqr_hi = float(perf_last.quantile(0.25)), float(perf_last.quantile(0.75))
+            iqr_w = iqr_hi - iqr_lo
+
+            mag7_in = [t for t in mag7 if t in perf_last.index]
+            rest_idx = perf_last.index.difference(mag7_in)
+            mag7_mean = float(perf_last[mag7_in].mean()) if len(mag7_in) else np.nan
+            rest_mean = float(perf_last[rest_idx].mean()) if len(rest_idx) else np.nan
+            mag7_beat = float((perf_last[mag7_in] > spx_last).mean() * 100) if len(mag7_in) else np.nan
+
+            gains_all = float((perf_last - base).sum())
+            topN = 10
+            top_contrib = np.nan
+            if abs(gains_all) > 1e-9:
+                top_contrib = float((perf_last.sort_values(ascending=False).head(topN) - base).sum() / gains_all * 100)
+
+            rets = rebased.pct_change().replace([np.inf, -np.inf], np.nan).dropna(how="all")
+            spx_r = pd.Series(spx_rebased, index=spx_rebased.index).pct_change()
+            corr_to_spx = rets.corrwith(spx_r, axis=0).dropna()
+            corr_med = float(corr_to_spx.median()) if len(corr_to_spx) else np.nan
+            low_corr_share = float((corr_to_spx < 0.3).mean() * 100) if len(corr_to_spx) else np.nan
+
+            spx_chg = spx_last - base
+            k = min(5, n_names)
+            leaders = perf_last.sort_values(ascending=False).head(k)
+            laggards = perf_last.sort_values(ascending=True).head(k)
+
+            print(f"=== Rebased performance read — {as_of} (window: {n_days} sessions) ===\n", file=buf2)
+            print("[Market]", file=buf2)
+            print(f"S&P 500 is {_fmt_pct(spx_chg)} over the window.", file=buf2)
+            print(f"Equal-weight average is {_fmt_pct(eq_avg - base)}, median is {_fmt_pct(eq_med - base)}.", file=buf2)
+            if np.isfinite(eq_avg) and np.isfinite(spx_last):
+                gap = (eq_avg - spx_last)
+                side = "above" if gap >= 0 else "below"
+                print(f"Equal-weight sits {_fmt_pct(abs(gap))} {side} the index.", file=buf2)
+            print("", file=buf2)
+
+            print("[Breadth]", file=buf2)
+            print(f"{_fmt_pct(pct_pos)} of names are up. {_fmt_pct(pct_beat)} beat the index.", file=buf2)
+            print(f"Dispersion std is {_fmt_num(disp_std)} points on the rebased scale.", file=buf2)
+            print(f"IQR width is {_fmt_num(iqr_w)} points ({_fmt_num(iqr_lo)} to {_fmt_num(iqr_hi)}).", file=buf2)
+            if pct_pos >= 70 and pct_beat >= 55:
+                print("Rally is broad. Leadership is shared across many names.", file=buf2)
+            elif pct_pos <= 35 and pct_beat <= 45:
+                print("Rally is narrow or absent. Leadership is concentrated.", file=buf2)
             else:
-                as_of = rebased.index[-1].date()
-                perf_last = rebased.iloc[-1].dropna()
-                spx_last = float(spx_rebased.iloc[-1])
-                n_names = len(perf_last)
-                eq_avg = float(perf_last.mean())
-                eq_med = float(perf_last.median())
-                pct_pos = float((perf_last > base).mean() * 100)
-                pct_beat = float((perf_last > spx_last).mean() * 100)
-                disp_std = float(perf_last.std(ddof=0))
-                iqr_lo, iqr_hi = float(perf_last.quantile(0.25)), float(perf_last.quantile(0.75))
-                iqr_w = iqr_hi - iqr_lo
-
-                mag7_in = [t for t in mag7 if t in perf_last.index]
-                rest_idx = perf_last.index.difference(mag7_in)
-                mag7_mean = float(perf_last[mag7_in].mean()) if len(mag7_in) else np.nan
-                rest_mean = float(perf_last[rest_idx].mean()) if len(rest_idx) else np.nan
-                mag7_beat = float((perf_last[mag7_in] > spx_last).mean() * 100) if len(mag7_in) else np.nan
-
-                gains_all = float((perf_last - base).sum())
-                topN = 10
-                top_contrib = np.nan
-                if abs(gains_all) > 1e-9:
-                    top_contrib = float((perf_last.sort_values(ascending=False).head(topN) - base).sum() / gains_all * 100)
-
-                rets = rebased.pct_change().replace([np.inf, -np.inf], np.nan).dropna(how="all")
-                spx_r = pd.Series(spx_rebased, index=spx_rebased.index).pct_change()
-                corr_to_spx = rets.corrwith(spx_r, axis=0).dropna()
-                corr_med = float(corr_to_spx.median()) if len(corr_to_spx) else np.nan
-                low_corr_share = float((corr_to_spx < 0.3).mean() * 100) if len(corr_to_spx) else np.nan
-
-                spx_chg = spx_last - base
-                k = min(5, n_names)
-                leaders = perf_last.sort_values(ascending=False).head(k)
-                laggards = perf_last.sort_values(ascending=True).head(k)
-
-                print(f"=== Rebased performance read — {as_of} (window: {n_days} sessions) ===\n", file=buf2)
-                print("[Market]", file=buf2)
-                print(f"S&P 500 is {_fmt_pct(spx_chg)} over the window.", file=buf2)
-                print(f"Equal-weight average is {_fmt_pct(eq_avg - base)}, median is {_fmt_pct(eq_med - base)}.", file=buf2)
-                if np.isfinite(eq_avg) and np.isfinite(spx_last):
-                    gap = (eq_avg - spx_last)
-                    side = "above" if gap >= 0 else "below"
-                    print(f"Equal-weight sits {_fmt_pct(abs(gap))} {side} the index.", file=buf2)
-                print("", file=buf2)
-
-                print("[Breadth]", file=buf2)
-                print(f"{_fmt_pct(pct_pos)} of names are up. {_fmt_pct(pct_beat)} beat the index.", file=buf2)
-                print(f"Dispersion std is {_fmt_num(disp_std)} points on the rebased scale.", file=buf2)
-                print(f"IQR width is {_fmt_num(iqr_w)} points ({_fmt_num(iqr_lo)} to {_fmt_num(iqr_hi)}).", file=buf2)
-                if pct_pos >= 70 and pct_beat >= 55:
-                    print("Rally is broad. Leadership is shared across many names.", file=buf2)
-                elif pct_pos <= 35 and pct_beat <= 45:
-                    print("Rally is narrow or absent. Leadership is concentrated.", file=buf2)
-                else:
-                    print("Breadth is mixed. The tape can rotate quickly.", file=buf2)
-                print("", file=buf2)
-
-                print("[Concentration]", file=buf2)
-                if np.isfinite(top_contrib):
-                    print(f"Top {topN} names explain {_fmt_pct(top_contrib)} of equal-weight gains.", file=buf2)
-                if len(mag7_in):
-                    print(f"MAG7 equal-weight is {_fmt_pct(mag7_mean - base)}. Rest is {_fmt_pct(rest_mean - base)}.", file=buf2)
-                    if np.isfinite(mag7_beat):
-                        print(f"{_fmt_pct(mag7_beat)} of MAG7 beat the index.", file=buf2)
-                else:
-                    print("MAG7 tickers are not all present in this window.", file=buf2)
-                print("", file=buf2)
-
-                print("[Correlation]", file=buf2)
-                if len(corr_to_spx):
-                    print(f"Median correlation to the index is {_fmt_num(corr_med)}.", file=buf2)
-                    print(f"{_fmt_pct(low_corr_share)} of names show low correlation (<0.30).", file=buf2)
-                    if np.isfinite(corr_med) and corr_med < 0.5:
-                        print("Factor dispersion is high. Stock picking matters more.", file=buf2)
-                    elif np.isfinite(corr_med) and corr_med > 0.8:
-                        print("Common beta dominates. Moves are index-driven.", file=buf2)
-                    else:
-                        print("Correlation sits in a middle zone. Rotation can continue.", file=buf2)
+                print("Breadth is mixed. The tape can rotate quickly.", file=buf2)
+            print("", file=buf2)
+
+            print("[Concentration]", file=buf2)
+            if np.isfinite(top_contrib):
+                print(f"Top {topN} names explain {_fmt_pct(top_contrib)} of equal-weight gains.", file=buf2)
+            if len(mag7_in):
+                print(f"MAG7 equal-weight is {_fmt_pct(mag7_mean - base)}. Rest is {_fmt_pct(rest_mean - base)}.", file=buf2)
+                if np.isfinite(mag7_beat):
+                    print(f"{_fmt_pct(mag7_beat)} of MAG7 beat the index.", file=buf2)
+            else:
+                print("MAG7 tickers are not all present in this window.", file=buf2)
+            print("", file=buf2)
+
+            print("[Correlation]", file=buf2)
+            if len(corr_to_spx):
+                print(f"Median correlation to the index is {_fmt_num(corr_med)}.", file=buf2)
+                print(f"{_fmt_pct(low_corr_share)} of names show low correlation (<0.30).", file=buf2)
+                if np.isfinite(corr_med) and corr_med < 0.5:
+                    print("Factor dispersion is high. Stock picking matters more.", file=buf2)
+                elif np.isfinite(corr_med) and corr_med > 0.8:
+                    print("Common beta dominates. Moves are index-driven.", file=buf2)
                 else:
-                    print("Not enough data to compute correlations.", file=buf2)
-                print("", file=buf2)
-
-                print("[Leaders]", file=buf2)
-                for t, v in leaders.items():
-                    print(f"  {t}: {_fmt_pct(v - base)}", file=buf2)
-
-                print("\n[Laggards]", file=buf2)
-                for t, v in laggards.items():
-                    print(f"  {t}: {_fmt_pct(v - base)}", file=buf2)
-
-                print("\n[What to monitor]", file=buf2)
-                print("Watch the gap between equal-weight and index. A widening gap signals concentration risk.", file=buf2)
-                print("Track the share beating the index. Sustained readings above 55% support trend durability.", file=buf2)
-                print("Watch median correlation. Falling correlation favors dispersion and relative value setups.", file=buf2)
-
-            st.text(buf2.getvalue())
-
-    # ===================== SECTION 3 — Daily Return Heatmap =====================
-    st.header("Daily Return Heatmap")
-
-    with st.expander("Methodology", expanded=False):
-        st.write("Shows daily % returns for all names over the selected window. Highlights broad up/down days, dispersion, and leadership.")
-        st.write("Use it to spot synchronized moves, stress days, and rotation across the universe.")
-
-        st.write("**Daily return (per name)**")
-        st.latex(r"r_{i,t}=\frac{P_{i,t}}{P_{i,t-1}}-1")
-
-        st.write("**Heatmap values**")
-        st.write("Cells display r_{i,t}. Tickers are sorted by the most recent day’s return so leaders/laggards are obvious.")
-
-        st.write("**Robust color scale (cap extremes)**")
-        st.latex(r"c=\operatorname{P95}\left(\left|r_{i,t}\right|\right)\ \text{over the window}")
-        st.latex(r"\text{color range}=[-c,\,+c],\quad \text{midpoint}=0")
-        st.write("Capping avoids a few outliers overpowering the color scale.")
-
-        st.write("**Breadth and dispersion (how to read)**")
-        st.latex(r"\text{Up share}_t=100\cdot \frac{1}{N}\sum_{i=1}^{N}\mathbf{1}[r_{i,t}>0]")
-        st.latex(r"\sigma_{\text{cs},t}=\operatorname{stdev}\{r_{i,t}\}_{i=1}^{N}")
-        st.write("- High up share with low dispersion = uniform risk-on.")
-        st.write("- Mixed colors with high dispersion = rotation and factor spread.")
-        st.write("- Clusters of red/green by industry often flag sector moves.")
-
-        st.write("**Large-move counts (quick context)**")
-        st.latex(r"\text{BigUp}_t=\sum_{i}\mathbf{1}[r_{i,t}\ge \tau],\quad \text{BigDn}_t=\sum_{i}\mathbf{1}[r_{i,t}\le -\tau]")
-        st.latex(r"\tau=2\% \ \text{(default)}")
-        st.write("A jump in BigUp/BigDn signals a thrust or a shock day.")
-
-        st.write("**Short-horizon follow-through**")
-        st.latex(r"\bar{r}_{i,t}^{(w)}=\frac{1}{w}\sum_{k=0}^{w-1} r_{i,t-k},\quad w=5")
-        st.write("A broad rise in 5-day averages supports continuation; a fade warns of stall.")
-
-        st.write("**Practical reads**")
-        st.write("- Many greens, low dispersion: beta tailwind; index setups work.")
-        st.write("- Greens + high dispersion: stock picking/sector tilts matter.")
-        st.write("- Reds concentrated in a few groups: rotate risk, not necessarily de-risk.")
-        st.write("- Extreme red breadth with spikes in dispersion: watch liquidity and reduce gross.")
-
-
-    # Daily returns last N days
-    ret_daily = clean_close.pct_change().iloc[1:]
-    ret_window = int(heat_last_days)
-    ret_last = ret_daily.iloc[-ret_window:]
-    if ret_last.empty:
-        st.warning("Not enough data for the daily return heatmap.")
-    else:
-        order = ret_last.iloc[-1].sort_values(ascending=True).index
-        ret_last = ret_last[order]
-
-        abs_max = np.nanpercentile(np.abs(ret_last.values), 95)
-        z = ret_last.T.values
-        x = ret_last.index
-        y = list(order)
-
-        n_dates = len(x)
-        step = max(1, n_dates // 10)
-        xtick_vals = x[::step]
-        xtick_texts = [ts.strftime("%Y-%m-%d") for ts in xtick_vals]
-
-        fig_hm = go.Figure(go.Heatmap(
-            z=z, x=x, y=y,
-            colorscale="RdYlGn",
-            zmin=-abs_max, zmax=abs_max, zmid=0,
-            colorbar=dict(title="Daily Return", tickformat=".0%"),
-            hovertemplate="%{y}<br>%{x|%Y-%m-%d}<br>%{z:.2%}<extra></extra>"
-        ))
+                    print("Correlation sits in a middle zone. Rotation can continue.", file=buf2)
+            else:
+                print("Not enough data to compute correlations.", file=buf2)
+            print("", file=buf2)
 
-        height = max(800, min(3200, 18 * len(y)))
-        fig_hm.update_layout(
-            template="plotly_dark",
-            title=f"Last {ret_window}-Day Daily Return Heatmap",
-            height=height,
-            margin=dict(l=100, r=40, t=60, b=60),
-            font=dict(color="white")
-        )
-        fig_hm.update_yaxes(title="Tickers (sorted by latest daily return)", tickfont=dict(size=8))
-        fig_hm.update_xaxes(title="Date", tickmode="array", tickvals=xtick_vals, ticktext=xtick_texts, tickangle=45)
-        st.plotly_chart(fig_hm, use_container_width=True)
+            print("[Leaders]", file=buf2)
+            for t, v in leaders.items():
+                print(f"  {t}: {_fmt_pct(v - base)}", file=buf2)
 
-        with st.expander("Dynamic Interpretation", expanded=False):
-            buf3 = io.StringIO()
+            print("\n[Laggards]", file=buf2)
+            for t, v in laggards.items():
+                print(f"  {t}: {_fmt_pct(v - base)}", file=buf2)
 
-            def _pct(x):
-                return "n/a" if pd.isna(x) else f"{x*100:.1f}%"
+            print("\n[What to monitor]", file=buf2)
+            print("Watch the gap between equal-weight and index. A widening gap signals concentration risk.", file=buf2)
+            print("Track the share beating the index. Sustained readings above 55% support trend durability.", file=buf2)
+            print("Watch median correlation. Falling correlation favors dispersion and relative value setups.", file=buf2)
 
-            def _pp(x):
-                return "n/a" if pd.isna(x) else f"{x*100:.2f}%"
+        st.text(buf2.getvalue())
 
-            if ret_last.empty:
-                print("No data for interpretation.", file=buf3)
-            else:
-                as_of = ret_last.index[-1].date()
-                last = ret_last.iloc[-1]
-                N = last.shape[0]
-                up = int((last > 0).sum())
-                dn = int((last < 0).sum())
-                flat = int(N - up - dn)
-                mean = float(last.mean()); med = float(last.median())
-                std = float(last.std(ddof=0))
-                q25 = float(last.quantile(0.25)); q75 = float(last.quantile(0.75))
-                iqr = q75 - q25
-                thr = 0.02
-                big_up = int((last >=  thr).sum())
-                big_dn = int((last <= -thr).sum())
-                w = min(5, len(ret_last))
-                avg_w = ret_last.tail(w).mean()
-                pct_pos_w = float((avg_w > 0).mean())
-                cs_std = ret_last.std(axis=1, ddof=0)
-                today_std = float(cs_std.iloc[-1])
-                disp_pct  = float((cs_std <= today_std).mean())
-                k = min(10, N)
-                leaders  = last.sort_values(ascending=False).head(k)
-                laggards = last.sort_values(ascending=True ).head(k)
-
-                def _streak(s, max_look=20):
-                    v = s.tail(max_look).to_numpy(dtype=float)
-                    sign = np.sign(v); sign[np.isnan(sign)] = 0
-                    if len(sign) == 0 or sign[-1] == 0:
-                        return 0
-                    tgt = sign[-1]; cnt = 0
-                    for x in sign[::-1]:
-                        if x == tgt: cnt += 1
-                        else: break
-                    return int(cnt if tgt > 0 else -cnt)
-
-                streaks = {t: _streak(ret_last[t]) for t in set(leaders.index).union(laggards.index)}
-
-                print(f"=== Daily return heatmap read — {as_of} (last {len(ret_last)} sessions) ===", file=buf3)
-                print("\n[Today]", file=buf3)
-                print(f"Up: {up}/{N} ({_pct(up/N)}). Down: {dn}/{N} ({_pct(dn/N)}). Flat: {flat}.", file=buf3)
-                print(f"Mean: {_pp(mean)}. Median: {_pp(med)}. Std: {_pp(std)}. IQR: {_pp(iqr)}.", file=buf3)
-                print(f"Moves ≥ {int(thr*100)}%: +{big_up}. Moves ≤ -{int(thr*100)}%: {big_dn}.", file=buf3)
-
-                print("\n[Recent breadth]", file=buf3)
-                print(f"{_pct(pct_pos_w)} of names have a positive average over the last {w} sessions.", file=buf3)
-
-                print("\n[Dispersion]", file=buf3)
-                print(f"Cross-section std today: {_pp(today_std)} (window percentile ~{disp_pct*100:.0f}th).", file=buf3)
-
-                print("\n[Leaders today]", file=buf3)
-                for t, v in leaders.items():
-                    stv = streaks.get(t, 0)
-                    lab = ("flat" if stv == 0 else (f"{stv}d up" if stv > 0 else f"{-stv}d down"))
-                    print(f"  {t}: {_pp(v)} ({lab})", file=buf3)
-
-                print("\n[Laggards today]", file=buf3)
-                for t, v in laggards.items():
-                    stv = streaks.get(t, 0)
-                    lab = ("flat" if stv == 0 else (f"{stv}d up" if stv > 0 else f"{-stv}d down"))
-                    print(f"  {t}: {_pp(v)} ({lab})", file=buf3)
-
-                print("\n[What to monitor]", file=buf3)
-                print("Watch big-move counts and the 5-day positive share for follow-through.", file=buf3)
-                print("Track dispersion; elevated dispersion favors relative moves over index moves.", file=buf3)
-
-            st.text(buf3.getvalue())
-
-    # ===================== SECTION 4 — Percentile Momentum Heatmap =====================
-    st.header("Percentile Momentum Heatmap")
-
-    with st.expander("Methodology", expanded=False):
-        st.write("Ranks each stock’s medium-horizon return against the cross-section each day.")
-        st.write("Use it to spot broad momentum, rotation, and persistence.")
-
-        st.write("**n-day return (per name)**")
-        st.latex(r"r^{(n)}_{i,t}=\frac{P_{i,t}}{P_{i,t-n}}-1")
-
-        st.write("**Cross-sectional percentile (per day)**")
-        st.latex(r"p_{i,t}=\frac{\operatorname{rank}\!\left(r^{(n)}_{i,t}\right)}{N}")
-        st.write("0 means worst in the universe that day. 1 means best.")
-        st.write("The heatmap shows p_{i,t}. Rows are sorted by the latest percentile.")
-
-        st.write("**Breadth buckets (how to read)**")
-        st.latex(r"\text{Top\,20\%}_t=\frac{1}{N}\sum_{i}\mathbf{1}[p_{i,t}\ge 0.80]")
-        st.latex(r"\text{Bottom\,20\%}_t=\frac{1}{N}\sum_{i}\mathbf{1}[p_{i,t}\le 0.20]")
-        st.write("High Top-20% share signals broad upside momentum. High Bottom-20% share signals broad weakness.")
-
-        st.write("**Momentum shift vs a short lookback**")
-        st.latex(r"\Delta p_i=p_{i,T}-p_{i,T-w}")
-        st.write("Improving names: Δp_i > 0. Weakening names: Δp_i < 0.")
-
-        st.write("**Persistence (top/bottom quintile)**")
-        st.latex(r"\text{TopQ}_{i}=\sum_{k=0}^{w-1}\mathbf{1}[p_{i,T-k}\ge 0.80]")
-        st.latex(r"\text{BotQ}_{i}=\sum_{k=0}^{w-1}\mathbf{1}[p_{i,T-k}\le 0.20]")
-        st.write("Names with TopQ = w held leadership. BotQ = w stayed weak.")
-        
-        st.write("**Practical reads**")
-        st.write("- Rising median percentile and high Top-20% share: trend has breadth.")
-        st.write("- Mixed median with both tails active: rotation/dispersion regime.")
-        st.write("- Persistent top-quintile list: candidates for follow-through.")
-        st.write("- Persistent bottom-quintile list: candidates for mean-reversion checks.")
-
-
-    look_days = int(mom_look)
-    ret_n = clean_close.pct_change(look_days)
-    ret_n = ret_n.iloc[look_days:]
-    if ret_n.empty:
-        st.warning("Not enough data for the momentum heatmap.")
-    else:
-        perc = ret_n.rank(axis=1, pct=True)
-        order2 = perc.iloc[-1].sort_values(ascending=True).index
-        perc = perc[order2]
-
-        z = perc.T.values
-        x = perc.index
-        y = list(order2)
-
-        n_dates = len(x)
-        step = max(1, n_dates // 10)
-        xtick_vals  = x[::step]
-        xtick_texts = [ts.strftime("%Y-%m-%d") for ts in xtick_vals]
-
-        fig_pm = go.Figure(go.Heatmap(
-            z=z, x=x, y=y,
-            colorscale="Viridis",
-            zmin=0, zmax=1,
-            colorbar=dict(title="Return Percentile"),
-            hovertemplate="%{y}<br>%{x|%Y-%m-%d}<br>%{z:.0%}<extra></extra>"
-        ))
+# ===================== SECTION 3 — Daily Return Heatmap =====================
+st.header("Daily Return Heatmap")
 
-        height = max(800, min(3200, 18 * len(y)))
-        fig_pm.update_layout(
-            template="plotly_dark",
-            title=f"{look_days}-Day Return Percentile Heatmap",
-            height=height,
-            margin=dict(l=110, r=40, t=60, b=60),
-            font=dict(color="white")
-        )
-        fig_pm.update_yaxes(title="Tickers (sorted by latest %ile)", tickfont=dict(size=8))
-        fig_pm.update_xaxes(title="Date", tickmode="array", tickvals=xtick_vals, ticktext=xtick_texts, tickangle=45)
-        st.plotly_chart(fig_pm, use_container_width=True)
-
-        with st.expander("Dynamic Interpretation", expanded=False):
-            buf4 = io.StringIO()
-            if perc.empty or ret_n.empty:
-                print("No data for interpretation.", file=buf4)
-            else:
-                as_of = perc.index[-1].date()
-                last_p = perc.iloc[-1].astype(float)
-                last_r = ret_n.iloc[-1].astype(float)
-
-                N = int(last_p.shape[0])
-                mean_p = float(last_p.mean()); med_p = float(last_p.median())
-                q25 = float(last_p.quantile(0.25)); q75 = float(last_p.quantile(0.75))
-                iqr_w = q75 - q25
-
-                top10 = float((last_p >= 0.90).mean() * 100)
-                top20 = float((last_p >= 0.80).mean() * 100)
-                mid40 = float(((last_p > 0.40) & (last_p < 0.60)).mean() * 100)
-                bot20 = float((last_p <= 0.20).mean() * 100)
-                bot10 = float((last_p <= 0.10).mean() * 100)
-
-                pct_up = float((last_r > 0).mean() * 100)
-
-                look = min(5, len(perc))
-                delta = (last_p - perc.iloc[-look].astype(float)).dropna()
-                improving = float((delta > 0).mean() * 100)
-                weakening = float((delta < 0).mean() * 100)
-                delta_med = float(delta.median())
-
-                k = min(10, N)
-                leaders  = last_p.sort_values(ascending=False).head(k)
-                laggards = last_p.sort_values(ascending=True ).head(k)
-
-                window_p = 5
-                top_quint = (perc.tail(window_p) >= 0.80).sum()
-                bot_quint = (perc.tail(window_p) <= 0.20).sum()
-                persistent_up = top_quint[top_quint == window_p].index.tolist()
-                persistent_dn = bot_quint[bot_quint == window_p].index.tolist()
-
-                print(f"=== {look_days}-day momentum read — {as_of} ===", file=buf4)
-                print("\n[Snapshot]", file=buf4)
-                print(f"Names: {N}. Up on window: {pct_up:.1f}%.", file=buf4)
-                print(f"Mean percentile: {mean_p:.2f}. Median: {med_p:.2f}.", file=buf4)
-                print(f"IQR: {q25:.2f}–{q75:.2f} (width {iqr_w:.2f}).", file=buf4)
-
-                print("\n[Breadth]", file=buf4)
-                print(f"Top 10%: {top10:.1f}%. Top 20%: {top20:.1f}%.", file=buf4)
-                print(f"Middle 40–60%: {mid40:.1f}%.", file=buf4)
-                print(f"Bottom 20%: {bot20:.1f}%. Bottom 10%: {bot10:.1f}%.", file=buf4)
-
-                print("\n[Shift]", file=buf4)
-                print(f"Improving vs {look} days ago: {improving:.1f}%. Weakening: {weakening:.1f}%.", file=buf4)
-                print(f"Median percentile change: {delta_med:+.2f}.", file=buf4)
-
-                print("\n[Leaders]", file=buf4)
-                for t, v in leaders.items():
-                    print(f"  {t}: {v:.2f}", file=buf4)
-
-                print("\n[Laggards]", file=buf4)
-                for t, v in laggards.items():
-                    print(f"  {t}: {v:.2f}", file=buf4)
-
-                print("\n[Persistence]", file=buf4)
-                if persistent_up:
-                    up_list = ", ".join(persistent_up[:15]) + ("…" if len(persistent_up) > 15 else "")
-                    print(f"Top-quintile {window_p} days: {up_list}", file=buf4)
-                else:
-                    print("No names stayed in the top quintile.", file=buf4)
-                if persistent_dn:
-                    dn_list = ", ".join(persistent_dn[:15]) + ("…" if len(persistent_dn) > 15 else "")
-                    print(f"Bottom-quintile {window_p} days: {dn_list}", file=buf4)
-                else:
-                    print("No names stayed in the bottom quintile.", file=buf4)
+# Methodology (standalone expander)
+with st.expander("Methodology", expanded=False):
+    st.write("Shows daily % returns for all names over the selected window. Highlights broad up/down days, dispersion, and leadership.")
+    st.write("Use it to spot synchronized moves, stress days, and rotation across the universe.")
+
+    st.write("**Daily return (per name)**")
+    st.latex(r"r_{i,t}=\frac{P_{i,t}}{P_{i,t-1}}-1")
+
+    st.write("**Heatmap values**")
+    st.write("Cells display r_{i,t}. Tickers are sorted by the most recent day’s return so leaders/laggards are obvious.")
+
+    st.write("**Robust color scale (cap extremes)**")
+    st.latex(r"c=\operatorname{P95}\left(\left|r_{i,t}\right|\right)\ \text{over the window}")
+    st.latex(r"\text{color range}=[-c,\,+c],\quad \text{midpoint}=0")
+    st.write("Capping avoids a few outliers overpowering the color scale.")
+
+    st.write("**Breadth and dispersion (how to read)**")
+    st.latex(r"\text{Up share}_t=100\cdot \frac{1}{N}\sum_{i=1}^{N}\mathbf{1}[r_{i,t}>0]")
+    st.latex(r"\sigma_{\text{cs},t}=\operatorname{stdev}\{r_{i,t}\}_{i=1}^{N}")
+    st.write("- High up share with low dispersion = uniform risk-on.")
+    st.write("- Mixed colors with high dispersion = rotation and factor spread.")
+    st.write("- Clusters of red/green by industry often flag sector moves.")
+
+    st.write("**Large-move counts (quick context)**")
+    st.latex(r"\text{BigUp}_t=\sum_{i}\mathbf{1}[r_{i,t}\ge \tau],\quad \text{BigDn}_t=\sum_{i}\mathbf{1}[r_{i,t}\le -\tau]")
+    st.latex(r"\tau=2\% \ \text{(default)}")
+    st.write("A jump in BigUp/BigDn signals a thrust or a shock day.")
+
+    st.write("**Short-horizon follow-through**")
+    st.latex(r"\bar{r}_{i,t}^{(w)}=\frac{1}{w}\sum_{k=0}^{w-1} r_{i,t-k},\quad w=5")
+    st.write("A broad rise in 5-day averages supports continuation; a fade warns of stall.")
+
+    st.write("**Practical reads**")
+    st.write("- Many greens, low dispersion: beta tailwind; index setups work.")
+    st.write("- Greens + high dispersion: stock picking/sector tilts matter.")
+    st.write("- Reds concentrated in a few groups: rotate risk, not necessarily de-risk.")
+    st.write("- Extreme red breadth with spikes in dispersion: watch liquidity and reduce gross.")
+
+# Daily returns last N days
+ret_daily = clean_close.pct_change().iloc[1:]
+ret_window = int(heat_last_days)
+ret_last = ret_daily.iloc[-ret_window:]
+if ret_last.empty:
+    st.warning("Not enough data for the daily return heatmap.")
+else:
+    order = ret_last.iloc[-1].sort_values(ascending=True).index
+    ret_last = ret_last[order]
+
+    abs_max = np.nanpercentile(np.abs(ret_last.values), 95)
+    z = ret_last.T.values
+    x = ret_last.index
+    y = list(order)
+
+    n_dates = len(x)
+    step = max(1, n_dates // 10)
+    xtick_vals = x[::step]
+    xtick_texts = [ts.strftime("%Y-%m-%d") for ts in xtick_vals]
+
+    fig_hm = go.Figure(go.Heatmap(
+        z=z, x=x, y=y,
+        colorscale="RdYlGn",
+        zmin=-abs_max, zmax=abs_max, zmid=0,
+        colorbar=dict(title="Daily Return", tickformat=".0%"),
+        hovertemplate="%{y}<br>%{x|%Y-%m-%d}<br>%{z:.2%}<extra></extra>"
+    ))
+
+    height = max(800, min(3200, 18 * len(y)))
+    fig_hm.update_layout(
+        template="plotly_dark",
+        title=f"Last {ret_window}-Day Daily Return Heatmap",
+        height=height,
+        margin=dict(l=100, r=40, t=60, b=60),
+        font=dict(color="white")
+    )
+    fig_hm.update_yaxes(title="Tickers (sorted by latest daily return)", tickfont=dict(size=8))
+    fig_hm.update_xaxes(title="Date", tickmode="array", tickvals=xtick_vals, ticktext=xtick_texts, tickangle=45)
+    st.plotly_chart(fig_hm, use_container_width=True)
+
+    # Dynamic Interpretation (standalone expander)
+    with st.expander("Dynamic Interpretation", expanded=False):
+        buf3 = io.StringIO()
+
+        def _pct(x):
+            return "n/a" if pd.isna(x) else f"{x*100:.1f}%"
 
-                print("\n[Focus]", file=buf4)
-                print("Watch the top-quintile share. Rising share supports continuation.", file=buf4)
-                print("Track the median percentile. Sustained readings above 0.60 show broad momentum.", file=buf4)
-                print("Use persistence lists for follow-through and mean-reversion checks.", file=buf4)
+        def _pp(x):
+            return "n/a" if pd.isna(x) else f"{x*100:.2f}%"
+
+        if ret_last.empty:
+            print("No data for interpretation.", file=buf3)
+        else:
+            as_of = ret_last.index[-1].date()
+            last = ret_last.iloc[-1]
+            N = last.shape[0]
+            up = int((last > 0).sum())
+            dn = int((last < 0).sum())
+            flat = int(N - up - dn)
+            mean = float(last.mean()); med = float(last.median())
+            std = float(last.std(ddof=0))
+            q25 = float(last.quantile(0.25)); q75 = float(last.quantile(0.75))
+            iqr = q75 - q25
+            thr = 0.02
+            big_up = int((last >=  thr).sum())
+            big_dn = int((last <= -thr).sum())
+            w = min(5, len(ret_last))
+            avg_w = ret_last.tail(w).mean()
+            pct_pos_w = float((avg_w > 0).mean())
+            cs_std = ret_last.std(axis=1, ddof=0)
+            today_std = float(cs_std.iloc[-1])
+            disp_pct  = float((cs_std <= today_std).mean())
+            k = min(10, N)
+            leaders  = last.sort_values(ascending=False).head(k)
+            laggards = last.sort_values(ascending=True ).head(k)
+
+            def _streak(s, max_look=20):
+                v = s.tail(max_look).to_numpy(dtype=float)
+                sign = np.sign(v); sign[np.isnan(sign)] = 0
+                if len(sign) == 0 or sign[-1] == 0:
+                    return 0
+                tgt = sign[-1]; cnt = 0
+                for x in sign[::-1]:
+                    if x == tgt: cnt += 1
+                    else: break
+                return int(cnt if tgt > 0 else -cnt)
+
+            streaks = {t: _streak(ret_last[t]) for t in set(leaders.index).union(laggards.index)}
+
+            print(f"=== Daily return heatmap read — {as_of} (last {len(ret_last)} sessions) ===", file=buf3)
+            print("\n[Today]", file=buf3)
+            print(f"Up: {up}/{N} ({_pct(up/N)}). Down: {dn}/{N} ({_pct(dn/N)}). Flat: {flat}.", file=buf3)
+            print(f"Mean: {_pp(mean)}. Median: {_pp(med)}. Std: {_pp(std)}. IQR: {_pp(iqr)}.", file=buf3)
+            print(f"Moves ≥ {int(thr*100)}%: +{big_up}. Moves ≤ -{int(thr*100)}%: {big_dn}.", file=buf3)
+
+            print("\n[Recent breadth]", file=buf3)
+            print(f"{_pct(pct_pos_w)} of names have a positive average over the last {w} sessions.", file=buf3)
+
+            print("\n[Dispersion]", file=buf3)
+            print(f"Cross-section std today: {_pp(today_std)} (window percentile ~{disp_pct*100:.0f}th).", file=buf3)
+
+            print("\n[Leaders today]", file=buf3)
+            for t, v in leaders.items():
+                stv = streaks.get(t, 0)
+                lab = ("flat" if stv == 0 else (f"{stv}d up" if stv > 0 else f"{-stv}d down"))
+                print(f"  {t}: {_pp(v)} ({lab})", file=buf3)
+
+            print("\n[Laggards today]", file=buf3)
+            for t, v in laggards.items():
+                stv = streaks.get(t, 0)
+                lab = ("flat" if stv == 0 else (f"{stv}d up" if stv > 0 else f"{-stv}d down"))
+                print(f"  {t}: {_pp(v)} ({lab})", file=buf3)
+
+            print("\n[What to monitor]", file=buf3)
+            print("Watch big-move counts and the 5-day positive share for follow-through.", file=buf3)
+            print("Track dispersion; elevated dispersion favors relative moves over index moves.", file=buf3)
+
+        st.text(buf3.getvalue())
+
+# ===================== SECTION 4 — Percentile Momentum Heatmap =====================
+st.header("Percentile Momentum Heatmap")
+
+# Methodology (standalone expander)
+with st.expander("Methodology", expanded=False):
+    st.write("Ranks each stock’s medium-horizon return against the cross-section each day.")
+    st.write("Use it to spot broad momentum, rotation, and persistence.")
+
+    st.write("**n-day return (per name)**")
+    st.latex(r"r^{(n)}_{i,t}=\frac{P_{i,t}}{P_{i,t-n}}-1")
+
+    st.write("**Cross-sectional percentile (per day)**")
+    st.latex(r"p_{i,t}=\frac{\operatorname{rank}\!\left(r^{(n)}_{i,t}\right)}{N}")
+    st.write("0 means worst in the universe that day. 1 means best.")
+    st.write("The heatmap shows p_{i,t}. Rows are sorted by the latest percentile.")
+
+    st.write("**Breadth buckets (how to read)**")
+    st.latex(r"\text{Top\,20\%}_t=\frac{1}{N}\sum_{i}\mathbf{1}[p_{i,t}\ge 0.80]")
+    st.latex(r"\text{Bottom\,20\%}_t=\frac{1}{N}\sum_{i}\mathbf{1}[p_{i,t}\le 0.20]")
+    st.write("High Top-20% share signals broad upside momentum. High Bottom-20% share signals broad weakness.")
+
+    st.write("**Momentum shift vs a short lookback**")
+    st.latex(r"\Delta p_i=p_{i,T}-p_{i,T-w}")
+    st.write("Improving names: Δp_i > 0. Weakening names: Δp_i < 0.")
+
+    st.write("**Persistence (top/bottom quintile)**")
+    st.latex(r"\text{TopQ}_{i}=\sum_{k=0}^{w-1}\mathbf{1}[p_{i,T-k}\ge 0.80]")
+    st.latex(r"\text{BotQ}_{i}=\sum_{k=0}^{w-1}\mathbf{1}[p_{i,T-k}\le 0.20]")
+    st.write("Names with TopQ = w held leadership. BotQ = w stayed weak.")
+    
+    st.write("**Practical reads**")
+    st.write("- Rising median percentile and high Top-20% share: trend has breadth.")
+    st.write("- Mixed median with both tails active: rotation/dispersion regime.")
+    st.write("- Persistent top-quintile list: candidates for follow-through.")
+    st.write("- Persistent bottom-quintile list: candidates for mean-reversion checks.")
+
+look_days = int(mom_look)
+ret_n = clean_close.pct_change(look_days)
+ret_n = ret_n.iloc[look_days:]
+if ret_n.empty:
+    st.warning("Not enough data for the momentum heatmap.")
+else:
+    perc = ret_n.rank(axis=1, pct=True)
+    order2 = perc.iloc[-1].sort_values(ascending=True).index
+    perc = perc[order2]
+
+    z = perc.T.values
+    x = perc.index
+    y = list(order2)
+
+    n_dates = len(x)
+    step = max(1, n_dates // 10)
+    xtick_vals  = x[::step]
+    xtick_texts = [ts.strftime("%Y-%m-%d") for ts in xtick_vals]
+
+    fig_pm = go.Figure(go.Heatmap(
+        z=z, x=x, y=y,
+        colorscale="Viridis",
+        zmin=0, zmax=1,
+        colorbar=dict(title="Return Percentile"),
+        hovertemplate="%{y}<br>%{x|%Y-%m-%d}<br>%{z:.0%}<extra></extra>"
+    ))
+
+    height = max(800, min(3200, 18 * len(y)))
+    fig_pm.update_layout(
+        template="plotly_dark",
+        title=f"{look_days}-Day Return Percentile Heatmap",
+        height=height,
+        margin=dict(l=110, r=40, t=60, b=60),
+        font=dict(color="white")
+    )
+    fig_pm.update_yaxes(title="Tickers (sorted by latest %ile)", tickfont=dict(size=8))
+    fig_pm.update_xaxes(title="Date", tickmode="array", tickvals=xtick_vals, ticktext=xtick_texts, tickangle=45)
+    st.plotly_chart(fig_pm, use_container_width=True)
+
+    # Dynamic Interpretation (standalone expander)
+    with st.expander("Dynamic Interpretation", expanded=False):
+        buf4 = io.StringIO()
+        if perc.empty or ret_n.empty:
+            print("No data for interpretation.", file=buf4)
+        else:
+            as_of = perc.index[-1].date()
+            last_p = perc.iloc[-1].astype(float)
+            last_r = ret_n.iloc[-1].astype(float)
+
+            N = int(last_p.shape[0])
+            mean_p = float(last_p.mean()); med_p = float(last_p.median())
+            q25 = float(last_p.quantile(0.25)); q75 = float(last_p.quantile(0.75))
+            iqr_w = q75 - q25
+
+            top10 = float((last_p >= 0.90).mean() * 100)
+            top20 = float((last_p >= 0.80).mean() * 100)
+            mid40 = float(((last_p > 0.40) & (last_p < 0.60)).mean() * 100)
+            bot20 = float((last_p <= 0.20).mean() * 100)
+            bot10 = float((last_p <= 0.10).mean() * 100)
+
+            pct_up = float((last_r > 0).mean() * 100)
+
+            look = min(5, len(perc))
+            delta = (last_p - perc.iloc[-look].astype(float)).dropna()
+            improving = float((delta > 0).mean() * 100)
+            weakening = float((delta < 0).mean() * 100)
+            delta_med = float(delta.median())
+
+            k = min(10, N)
+            leaders  = last_p.sort_values(ascending=False).head(k)
+            laggards = last_p.sort_values(ascending=True ).head(k)
+
+            window_p = 5
+            top_quint = (perc.tail(window_p) >= 0.80).sum()
+            bot_quint = (perc.tail(window_p) <= 0.20).sum()
+            persistent_up = top_quint[top_quint == window_p].index.tolist()
+            persistent_dn = bot_quint[bot_quint == window_p].index.tolist()
+
+            print(f"=== {look_days}-day momentum read — {as_of} ===", file=buf4)
+            print("\n[Snapshot]", file=buf4)
+            print(f"Names: {N}. Up on window: {pct_up:.1f}%.", file=buf4)
+            print(f"Mean percentile: {mean_p:.2f}. Median: {med_p:.2f}.", file=buf4)
+            print(f"IQR: {q25:.2f}–{q75:.2f} (width {iqr_w:.2f}).", file=buf4)
+
+            print("\n[Breadth]", file=buf4)
+            print(f"Top 10%: {top10:.1f}%. Top 20%: {top20:.1f}%.", file=buf4)
+            print(f"Middle 40–60%: {mid40:.1f}%.", file=buf4)
+            print(f"Bottom 20%: {bot20:.1f}%. Bottom 10%: {bot10:.1f}%.", file=buf4)
+
+            print("\n[Shift]", file=buf4)
+            print(f"Improving vs {look} days ago: {improving:.1f}%. Weakening: {weakening:.1f}%.", file=buf4)
+            print(f"Median percentile change: {delta_med:+.2f}.", file=buf4)
+
+            print("\n[Leaders]", file=buf4)
+            for t, v in leaders.items():
+                print(f"  {t}: {v:.2f}", file=buf4)
+
+            print("\n[Laggards]", file=buf4)
+            for t, v in laggards.items():
+                print(f"  {t}: {v:.2f}", file=buf4)
+
+            print("\n[Persistence]", file=buf4)
+            if persistent_up:
+                up_list = ", ".join(persistent_up[:15]) + ("…" if len(persistent_up) > 15 else "")
+                print(f"Top-quintile {window_p} days: {up_list}", file=buf4)
+            else:
+                print("No names stayed in the top quintile.", file=buf4)
+            if persistent_dn:
+                dn_list = ", ".join(persistent_dn[:15]) + ("…" if len(persistent_dn) > 15 else "")
+                print(f"Bottom-quintile {window_p} days: {dn_list}", file=buf4)
+            else:
+                print("No names stayed in the bottom quintile.", file=buf4)
 
-            st.text(buf4.getvalue())
+            print("\n[Focus]", file=buf4)
+            print("Watch the top-quintile share. Rising share supports continuation.", file=buf4)
+            print("Track the median percentile. Sustained readings above 0.60 show broad momentum.", file=buf4)
+            print("Use persistence lists for follow-through and mean-reversion checks.", file=buf4)
 
+        st.text(buf4.getvalue())
 
 # Hide default Streamlit style
 st.markdown(