fix streak signal, reweight predictions, and reorder UI

- Fix streak signal: was giving 100% to streak chair after normalization
  (non-streak chairs were 0), now properly distributes probability with
  streak chair getting less as streak grows (actual mean reversion)
- Change recent window from 20 to 50 games
- Reweight signals based on backtest: base_rate 0.15→0.20 (best performer),
  recent 0.10→0.15, streak 0.10→0.05, balance 0.15→0.10
- Move Live Market Sentiment above Signal Breakdown

app/db.py

@@ -1031,12 +1031,12 @@ def _bayesian_prediction(winners, markov1, markov2):
     key2 = f"{winners[-2]}{winners[-1]}"
     m2 = markov2.get(key2, {c: 1 / 3 for c in CHAIR_LABELS})
 
-    # Signal 4: Recent 20-game frequency — 15%
-    recent = winners[-20:] if len(winners) >= 20 else winners
+    # Signal 4: Recent 50-game frequency — 15%
+    recent = winners[-50:] if len(winners) >= 50 else winners
     recent_total = len(recent)
     rec = {c: recent.count(c) / recent_total for c in CHAIR_LABELS}
 
-    # Signal 5: Streak momentum/regression — 10%
+    # Signal 5: Streak regression — 5%
     streak_chair = winners[-1]
     streak_len = 0
     for w in reversed(winners):
@@ -1045,20 +1045,11 @@ def _bayesian_prediction(winners, markov1, markov2):
         else:
             break
     # Regression to mean: longer streaks → lower probability of continuation
-    streak = {}
-    for c in CHAIR_LABELS:
-        if c == streak_chair:
-            streak[c] = max(0.1, 1 / 3 - streak_len * 0.05)
-        else:
-            streak[c] = 0
-    # Normalize streak signal
-    s_total = sum(streak.values())
-    if s_total > 0:
-        streak = {c: streak[c] / s_total for c in CHAIR_LABELS}
-    else:
-        streak = {c: 1 / 3 for c in CHAIR_LABELS}
+    cont_prob = max(0.1, 1 / 3 - streak_len * 0.05)
+    other_prob = (1 - cont_prob) / 2
+    streak = {c: (cont_prob if c == streak_chair else other_prob) for c in CHAIR_LABELS}
 
-    # Signal 6: Balance / Mean Reversion — 15%
+    # Signal 6: Balance / Mean Reversion — 10%
     # Look at last 50 games, invert frequencies to favor under-represented chairs
     window = min(50, len(winners))
     recent_50 = winners[-window:]
@@ -1067,8 +1058,8 @@ def _bayesian_prediction(winners, markov1, markov2):
     bal_total = sum(balance.values())
     balance = {c: balance[c] / bal_total for c in CHAIR_LABELS}
 
-    weights = {"base_rate": 0.15, "markov_1": 0.25, "markov_2": 0.25, "recent_20": 0.10, "streak": 0.10, "balance": 0.15}
-    signals = {"base_rate": base, "markov_1": m1, "markov_2": m2, "recent_20": rec, "streak": streak, "balance": balance}
+    weights = {"base_rate": 0.20, "markov_1": 0.25, "markov_2": 0.25, "recent_50": 0.15, "streak": 0.05, "balance": 0.10}
+    signals = {"base_rate": base, "markov_1": m1, "markov_2": m2, "recent_50": rec, "streak": streak, "balance": balance}
 
     combined = {c: 0 for c in CHAIR_LABELS}
     for sig_name, weight in weights.items():
@@ -1189,7 +1180,7 @@ def _backtest_theories(winners):
     if len(winners) <= warmup:
         return {"error": "Not enough data for backtesting"}
 
-    theories = ["base_rate", "markov_1", "markov_2", "recent_20", "streak", "balance", "combined"]
+    theories = ["base_rate", "markov_1", "markov_2", "recent_50", "streak", "balance", "combined"]
     full_hits = {t: 0 for t in theories}
     semi_hits = {t: 0 for t in theories}
     total_tested = 0
@@ -1217,8 +1208,8 @@ def _backtest_theories(winners):
         m2_probs = m2.get(key2, {c: 1 / 3 for c in CHAIR_LABELS})
         m2_ranked = sorted(CHAIR_LABELS, key=lambda c: m2_probs.get(c, 0), reverse=True)
 
-        # Recent-20
-        recent = history[-20:] if len(history) >= 20 else history
+        # Recent-50
+        recent = history[-50:] if len(history) >= 50 else history
         rec = {c: recent.count(c) / len(recent) for c in CHAIR_LABELS}
         rec_ranked = sorted(CHAIR_LABELS, key=lambda c: rec[c], reverse=True)
 
@@ -1230,17 +1221,9 @@ def _backtest_theories(winners):
                 streak_len += 1
             else:
                 break
-        streak_probs = {}
-        for c in CHAIR_LABELS:
-            if c == streak_chair:
-                streak_probs[c] = max(0.1, 1 / 3 - streak_len * 0.05)
-            else:
-                streak_probs[c] = 0
-        s_total = sum(streak_probs.values())
-        if s_total > 0:
-            streak_probs = {c: streak_probs[c] / s_total for c in CHAIR_LABELS}
-        else:
-            streak_probs = {c: 1 / 3 for c in CHAIR_LABELS}
+        cont_prob = max(0.1, 1 / 3 - streak_len * 0.05)
+        other_prob = (1 - cont_prob) / 2
+        streak_probs = {c: (cont_prob if c == streak_chair else other_prob) for c in CHAIR_LABELS}
         streak_ranked = sorted(CHAIR_LABELS, key=lambda c: streak_probs[c], reverse=True)
 
         # Balance / Mean Reversion
@@ -1254,8 +1237,8 @@ def _backtest_theories(winners):
 
         # Combined Bayesian
         combined = {c: 0 for c in CHAIR_LABELS}
-        weights = {"base_rate": 0.15, "markov_1": 0.25, "markov_2": 0.25, "recent_20": 0.10, "streak": 0.10, "balance": 0.15}
-        signals = {"base_rate": base, "markov_1": m1_probs, "markov_2": m2_probs, "recent_20": rec, "streak": streak_probs, "balance": bal_probs}
+        weights = {"base_rate": 0.20, "markov_1": 0.25, "markov_2": 0.25, "recent_50": 0.15, "streak": 0.05, "balance": 0.10}
+        signals = {"base_rate": base, "markov_1": m1_probs, "markov_2": m2_probs, "recent_50": rec, "streak": streak_probs, "balance": bal_probs}
         for sig_name, weight in weights.items():
             for c in CHAIR_LABELS:
                 combined[c] += weight * signals[sig_name].get(c, 1 / 3)
@@ -1263,7 +1246,7 @@ def _backtest_theories(winners):
 
         ranked = {
             "base_rate": base_ranked, "markov_1": m1_ranked, "markov_2": m2_ranked,
-            "recent_20": rec_ranked, "streak": streak_ranked, "balance": bal_ranked,
+            "recent_50": rec_ranked, "streak": streak_ranked, "balance": bal_ranked,
             "combined": combined_ranked,
         }
         for t in theories:

static/predictions.html

@@ -253,19 +253,8 @@ body { font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-
         <div id="impact-content"></div>
       </div>
 
-      <div class="panel" style="margin-top:8px">
-        <div class="panel-title">Signal Breakdown</div>
-        <table class="signal-table" id="signal-table">
-          <thead><tr><th>Signal</th><th>Weight</th><th>A</th><th>B</th><th>C</th></tr></thead>
-          <tbody></tbody>
-        </table>
-      </div>
-    </div>
-
-    <!-- Live Trends: Whale + Public -->
-    <div class="section">
-      <div class="section-title">Live Market Sentiment</div>
-      <div class="trends-grid">
+      <!-- Live Trends: Whale + Public -->
+      <div class="trends-grid" style="margin-top:16px">
         <div class="trend-panel">
           <div class="panel-title">Whale Trend (Top 5 Bettors)</div>
           <div id="whale-trend"><div class="trend-empty">Waiting for bets...</div></div>
@@ -278,7 +267,15 @@ body { font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-
         </div>
       </div>
       <!-- Historical: did crowd pick the winner? -->
-      <div id="crowd-stats" class="crowd-stats"></div>
+      <div id="crowd-stats" class="crowd-stats" style="margin-top:12px"></div>
+
+      <div class="panel" style="margin-top:12px">
+        <div class="panel-title">Signal Breakdown</div>
+        <table class="signal-table" id="signal-table">
+          <thead><tr><th>Signal</th><th>Weight</th><th>A</th><th>B</th><th>C</th></tr></thead>
+          <tbody></tbody>
+        </table>
+      </div>
     </div>
 
     <!-- Last 50 Predictions vs Actual -->
@@ -480,7 +477,7 @@ function renderPrediction(data) {
 
   // Signal table
   const tbody = $('signal-table').querySelector('tbody');
-  const sigNames = {'base_rate':'Base Rate','markov_1':'Markov-1','markov_2':'Markov-2','recent_20':'Recent 20','streak':'Streak','balance':'Balance'};
+  const sigNames = {'base_rate':'Base Rate','markov_1':'Markov-1','markov_2':'Markov-2','recent_50':'Recent 50','streak':'Streak','balance':'Balance'};
   tbody.innerHTML = Object.entries(data.signals).map(([key, sig]) =>
     `<tr><td style="text-align:left">${sigNames[key]||key}</td><td>${(sig.weight*100).toFixed(0)}%</td>` +
     CHAIRS.map(c => `<td style="color:${CHAIR_COLORS[c]}">${pct(sig.probs[c])}</td>`).join('') + '</tr>'
@@ -1051,7 +1048,7 @@ function renderRunsTest(runs) {
 
 function renderBacktest(bt) {
   if (bt.error) { $('backtest-cards').innerHTML = `<div style="color:var(--text2)">${bt.error}</div>`; return; }
-  const names = {base_rate:'Base Rate',markov_1:'Markov-1',markov_2:'Markov-2',recent_20:'Recent 20',streak:'Streak',balance:'Balance',combined:'Combined'};
+  const names = {base_rate:'Base Rate',markov_1:'Markov-1',markov_2:'Markov-2',recent_50:'Recent 50',streak:'Streak',balance:'Balance',combined:'Combined'};
   $('backtest-cards').innerHTML = Object.entries(bt.accuracy).map(([key, acc]) => {
     const fh = bt.full_hits?.[key] ?? '?';
     const sh = bt.semi_hits?.[key] ?? '?';
@@ -1062,7 +1059,7 @@ function renderBacktest(bt) {
   }).join('');
   if (bt.rolling_accuracy) {
     const ctx = $('backtest-chart').getContext('2d');
-    const colors = {base_rate:'#8b8fa3',markov_1:'#3b82f6',markov_2:'#ec4899',recent_20:'#f59e0b',streak:'#10b981',balance:'#f472b6',combined:'#6c5ce7'};
+    const colors = {base_rate:'#8b8fa3',markov_1:'#3b82f6',markov_2:'#ec4899',recent_50:'#f59e0b',streak:'#10b981',balance:'#f472b6',combined:'#6c5ce7'};
     const datasets = Object.entries(bt.rolling_accuracy).map(([key, data]) => ({
       label: names[key]||key, data, borderColor: colors[key]||'#fff', backgroundColor: 'transparent',
       borderWidth: key === 'combined' ? 3 : 1.5, pointRadius: 0, tension: 0.3,