Spread trade targets across distance bands

MKC showed 5 targets all far/Aggressive: target selection was top-5 by
quality (0.35*R:R + ...), and R:R grows with distance, so far levels
crowded out every nearby one.

generate_targets now selects for spread: always include the nearest
level, plus the best-quality representative from each distance band
(Conservative <=2.9 ATR, Moderate <=4.6 ATR, Aggressive beyond), then
fill remaining slots by quality. Restores a Conservative/Moderate/
Aggressive mix with the nearest target always present.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

app/services/recommendation_service.py

@@ -33,6 +33,12 @@ DEFAULT_RECOMMENDATION_CONFIG: dict[str, float] = {
 # metric we actually measure.
 _TARGET_HORIZON_DAYS = 30.0
 
+# Distance bands (in ATR) used to spread targets across Conservative / Moderate
+# / Aggressive. Aligned with where the touch-probability crosses 60% / 40% over
+# the horizon, so a target from each band tends to land in the matching label.
+_CONSERVATIVE_MAX_ATR = 2.9
+_MODERATE_MAX_ATR = 4.6
+
 
 def _clamp(value: float, low: float, high: float) -> float:
     return max(low, min(high, value))
@@ -255,12 +261,39 @@ class TargetGenerator:
                 }
             )
 
-        candidates.sort(key=lambda row: row["quality"], reverse=True)
-        selected = candidates[:5]
-        selected.sort(key=lambda row: row["distance_from_entry"])
+        if not candidates:
+            return []
 
-        # Classification is assigned later from the computed reach-probability
-        # (see _classify_by_probability), not from distance rank.
+        # Select up to 5 targets that SPAN the distance range, instead of the
+        # top-5 by quality (which biases toward far, high-R:R levels and buries
+        # every nearby target). Guarantees the nearest level plus a
+        # representative from each distance band when they exist, so the table
+        # offers a real mix of Conservative / Moderate / Aggressive.
+        conservative = [c for c in candidates if c["distance_atr_multiple"] <= _CONSERVATIVE_MAX_ATR]
+        moderate = [c for c in candidates if _CONSERVATIVE_MAX_ATR < c["distance_atr_multiple"] <= _MODERATE_MAX_ATR]
+        aggressive = [c for c in candidates if c["distance_atr_multiple"] > _MODERATE_MAX_ATR]
+
+        selected: list[dict[str, Any]] = []
+        selected_ids: set[int] = set()
+
+        def _add(candidate: dict[str, Any] | None) -> None:
+            if candidate is not None and candidate["sr_level_id"] not in selected_ids:
+                selected.append(candidate)
+                selected_ids.add(candidate["sr_level_id"])
+
+        # Nearest overall (the most likely / Conservative anchor)
+        _add(min(candidates, key=lambda c: c["distance_atr_multiple"]))
+        # Best-quality representative from each band → spread across labels
+        for bucket in (conservative, moderate, aggressive):
+            if bucket:
+                _add(max(bucket, key=lambda c: c["quality"]))
+        # Fill remaining slots with the next-best by quality
+        for candidate in sorted(candidates, key=lambda c: c["quality"], reverse=True):
+            if len(selected) >= 5:
+                break
+            _add(candidate)
+
+        selected.sort(key=lambda row: row["distance_from_entry"])
         for target in selected:
             target.pop("quality", None)