fix: deduplicate recommendations and widen position sizing range

- Add dedup check in recommendation worker: skip generation when latest
  rec for same ticker+window has identical action/mode/confidence
- Widen position sizing range (1-10% portfolio, 0.3-2% max loss) and
  factor in trend strength + evidence count for differentiated sizing
- API returns only latest recommendation per ticker by default (DISTINCT ON)
  to eliminate duplicate rows in the frontend list view

services/recommendation/eligibility.py

@@ -66,17 +66,17 @@ class EligibilityConfig:
 
     # --- Position sizing rules (Requirement 7.3) ---
     # Base portfolio allocation percentage
-    base_portfolio_pct: float = 0.02
+    base_portfolio_pct: float = 0.01
     # Maximum portfolio allocation percentage
-    max_portfolio_pct: float = 0.05
+    max_portfolio_pct: float = 0.10
     # Base max loss percentage
-    base_max_loss_pct: float = 0.005
+    base_max_loss_pct: float = 0.003
     # Maximum max loss percentage
-    max_max_loss_pct: float = 0.01
+    max_max_loss_pct: float = 0.02
     # Confidence scaling: higher confidence → larger position (linear)
-    confidence_sizing_weight: float = 0.5
+    confidence_sizing_weight: float = 0.8
     # Contradiction penalty: higher contradiction → smaller position
-    contradiction_sizing_penalty: float = 0.3
+    contradiction_sizing_penalty: float = 0.5
 
 
 DEFAULT_ELIGIBILITY_CONFIG = EligibilityConfig()
@@ -216,30 +216,40 @@ def _compute_position_sizing(
 ) -> PositionSizing:
     """Compute position sizing guidance from portfolio rules and signal quality.
 
-    Higher confidence → larger allocation (up to max).
+    Higher confidence and trend strength → larger allocation (up to max).
     Higher contradiction → smaller allocation (penalty).
+    Low evidence count further reduces allocation.
     """
-    # Start from base allocation
-    confidence_scale = config.base_portfolio_pct + (
-        config.confidence_sizing_weight
-        * summary.confidence
-        * (config.max_portfolio_pct - config.base_portfolio_pct)
-    )
+    # Confidence-based scaling over the full range
+    confidence_factor = config.confidence_sizing_weight * summary.confidence
+    # Trend strength multiplier — stronger trends justify larger positions
+    strength_factor = 0.5 + 0.5 * summary.trend_strength  # range [0.5, 1.0]
+
+    portfolio_range = config.max_portfolio_pct - config.base_portfolio_pct
+    raw_portfolio = config.base_portfolio_pct + confidence_factor * strength_factor * portfolio_range
 
     # Apply contradiction penalty
     contradiction_penalty = config.contradiction_sizing_penalty * summary.contradiction_score
-    portfolio_pct = confidence_scale * (1.0 - contradiction_penalty)
+    portfolio_pct = raw_portfolio * (1.0 - contradiction_penalty)
+
+    # Evidence count penalty — fewer sources = less confidence in sizing
+    evidence_count = len(summary.top_supporting_evidence) + len(summary.top_opposing_evidence)
+    if evidence_count < 3:
+        portfolio_pct *= 0.5
+    elif evidence_count < 5:
+        portfolio_pct *= 0.75
 
     # Clamp to bounds
     portfolio_pct = max(config.base_portfolio_pct * 0.5, min(portfolio_pct, config.max_portfolio_pct))
 
     # Max loss scales similarly
-    loss_scale = config.base_max_loss_pct + (
-        config.confidence_sizing_weight
-        * summary.confidence
-        * (config.max_max_loss_pct - config.base_max_loss_pct)
-    )
-    max_loss_pct = loss_scale * (1.0 - contradiction_penalty)
+    loss_range = config.max_max_loss_pct - config.base_max_loss_pct
+    raw_loss = config.base_max_loss_pct + confidence_factor * strength_factor * loss_range
+    max_loss_pct = raw_loss * (1.0 - contradiction_penalty)
+    if evidence_count < 3:
+        max_loss_pct *= 0.5
+    elif evidence_count < 5:
+        max_loss_pct *= 0.75
     max_loss_pct = max(config.base_max_loss_pct * 0.5, min(max_loss_pct, config.max_max_loss_pct))
 
     return PositionSizing(