Filter late priced market noise

tests/test_ops_market_opportunities.py

@@ -154,6 +154,91 @@ def test_build_market_opportunities_aggregates_fahrenheit_distribution():
     assert round(rows[0]["edge"], 2) == 0.46
 
 
+def test_build_market_opportunities_filters_late_priced_no_noise():
+    event = {
+        "slug": "highest-temperature-in-jeddah-on-july-3-2026",
+        "markets": [
+            {
+                "question": "Will the highest temperature in Jeddah be 36°C on July 3?",
+                "slug": "highest-temperature-in-jeddah-on-july-3-2026-36c",
+                "active": True,
+                "closed": False,
+                "enableOrderBook": True,
+                "liquidity": "154",
+                "volume": "6833",
+                "outcomes": '["Yes", "No"]',
+                "clobTokenIds": '["yes-36", "no-36"]',
+            }
+        ],
+    }
+    row = _row("jeddah")
+    row["local_time"] = "19:22"
+    row["current_max_so_far"] = None
+    row["deb_prediction"] = 35.3
+    row["model_cluster_sources"] = {"ECMWF": 34.5, "GFS": 44.0, "ICON": 35.0}
+    row["distribution_full"] = [
+        {"value": 36, "probability": 0.10},
+        {"value": 37, "probability": 0.90},
+    ]
+
+    rows = build_market_opportunity_rows(
+        [row],
+        {"jeddah": event},
+        {"yes-36": 0.99, "no-36": 0.01},
+        max_price=0.20,
+        side="both",
+        positive_edge_only=True,
+        min_edge=0.0,
+        limit=20,
+    )
+
+    assert rows == []
+
+
+def test_build_market_opportunities_keeps_late_no_when_bucket_conflicts_with_anchors():
+    event = {
+        "slug": "highest-temperature-in-jeddah-on-july-3-2026",
+        "markets": [
+            {
+                "question": "Will the highest temperature in Jeddah be 36°C on July 3?",
+                "slug": "highest-temperature-in-jeddah-on-july-3-2026-36c",
+                "active": True,
+                "closed": False,
+                "enableOrderBook": True,
+                "liquidity": "154",
+                "volume": "6833",
+                "outcomes": '["Yes", "No"]',
+                "clobTokenIds": '["yes-36", "no-36"]',
+            }
+        ],
+    }
+    row = _row("jeddah")
+    row["local_time"] = "19:22"
+    row["current_max_so_far"] = None
+    row["deb_prediction"] = 31.0
+    row["model_cluster_sources"] = {"ECMWF": 30.5, "GFS": 31.0, "ICON": 31.5}
+    row["distribution_full"] = [
+        {"value": 36, "probability": 0.10},
+        {"value": 31, "probability": 0.90},
+    ]
+
+    rows = build_market_opportunity_rows(
+        [row],
+        {"jeddah": event},
+        {"yes-36": 0.99, "no-36": 0.01},
+        max_price=0.20,
+        side="both",
+        positive_edge_only=True,
+        min_edge=0.0,
+        limit=20,
+    )
+
+    assert len(rows) == 1
+    assert rows[0]["bucket_label"] == "36°C"
+    assert rows[0]["side"] == "no"
+    assert rows[0]["side_probability"] == 0.90
+
+
 def test_ops_market_opportunities_requires_ops_admin(monkeypatch):
     def deny(_request):
         raise HTTPException(status_code=403, detail="ops only")

web/services/ops/market_opportunities.py

@@ -229,6 +229,60 @@ def _model_stats(row: Mapping[str, Any]) -> Tuple[Optional[float], Optional[floa
     return round(median, 1), round(max(values) - min(values), 1)
 
 
+def _local_hour(row: Mapping[str, Any]) -> Optional[int]:
+    text = str(row.get("local_time") or "").strip()
+    if not text:
+        return None
+    try:
+        hour = int(text.split(":", 1)[0])
+    except (TypeError, ValueError):
+        return None
+    return hour if 0 <= hour <= 23 else None
+
+
+def _option_near_value(option: Mapping[str, Any], value: Any) -> bool:
+    number = _finite_number(value)
+    if number is None:
+        return False
+    unit = str(option.get("unit") or "").upper()
+    tolerance = 2.0 if "F" in unit else 1.0
+    lower = _finite_number(option.get("lower"))
+    upper = _finite_number(option.get("upper"))
+    if lower is not None and upper is not None:
+        return lower - tolerance <= number <= upper + tolerance
+    if lower is not None:
+        return number >= lower - tolerance
+    if upper is not None:
+        return number <= upper + tolerance
+    return False
+
+
+def _is_late_priced_no_noise(
+    row: Mapping[str, Any],
+    option: Mapping[str, Any],
+    ask_prices_by_token: Mapping[str, Optional[float]],
+    tokens: Mapping[str, Optional[str]],
+    *,
+    no_ask: float,
+    model_median: Optional[float],
+) -> bool:
+    if no_ask > 0.05:
+        return False
+    hour = _local_hour(row)
+    if hour is None or hour < 17:
+        return False
+    yes_token = tokens.get("yes")
+    yes_ask = _finite_number(ask_prices_by_token.get(yes_token)) if yes_token else None
+    if yes_ask is not None and yes_ask < 0.80:
+        return False
+    anchors = (
+        row.get("current_max_so_far"),
+        row.get("deb_prediction"),
+        model_median,
+    )
+    return any(_option_near_value(option, anchor) for anchor in anchors)
+
+
 def _market_tokens(market: Mapping[str, Any]) -> Dict[str, Optional[str]]:
     outcomes = [str(item).strip().lower() for item in _json_list(market.get("outcomes"))]
     token_ids = [str(item).strip() for item in _json_list(market.get("clobTokenIds"))]
@@ -315,6 +369,15 @@ def build_market_opportunity_rows(
                 ask_number = _finite_number(ask)
                 if ask_number is None or ask_number <= 0 or ask_number >= float(max_price):
                     continue
+                if option_side == "no" and _is_late_priced_no_noise(
+                    scan_row,
+                    option,
+                    ask_prices_by_token,
+                    tokens,
+                    no_ask=ask_number,
+                    model_median=model_median,
+                ):
+                    continue
                 target_probability = (
                     model_probability
                     if option_side == "yes"