Merge pull request 'fix(forecasting): cap trend_ratio by v_rec magnitude when v_prior==0 (#1508)' (#1679) from fix/trend-ratio-vprior-zero-1508 into main

site-finder/09_macro_and_trend.py

@@ -9,9 +9,22 @@ Adds these tables/columns:
 Source for trend: per-flat register_date (objective_lots).
 Source for mortgage rate: prod cbr_mortgage_series (Sverdl region, latest "ставка ипотечная").
 """
-import sqlite3, pathlib, psycopg2, datetime as dt, statistics
+import sqlite3
+import pathlib
+import psycopg2
+import datetime as dt
 DB = pathlib.Path(__file__).parent / "analysis.db"
 
+# When v_prior == 0 we cannot compute a real ratio, so we synthesise one that
+# reflects the *magnitude* of v_rec rather than assigning a flat 2.0.
+# REF_VELOCITY: monthly flats/corpus that is considered "high activity" — above
+# this level the district earns a ratio near TREND_CAP_VPRIOR_ZERO; below it
+# the ratio slides toward neutral (1.0).  Calibrated from EKB corpus data where
+# an active corpus typically records 10–20 units/month; 10 is a conservative
+# "well-performing" benchmark.
+_REF_VELOCITY: float = 10.0        # monthly flats per corpus → full-cap reference
+_TREND_CAP_VPRIOR_ZERO: float = 1.5  # maximum ratio when v_prior == 0 (< hard 2.0 cap)
+
 EXTRA = """
 CREATE TABLE IF NOT EXISTS macro_context (
     key           TEXT PRIMARY KEY,
@@ -38,10 +51,13 @@ CREATE TABLE IF NOT EXISTS scoring_weights (
 def safe_alter(conn, sql):
     for stmt in sql.strip().split(";"):
         s = stmt.strip()
-        if not s: continue
-        try: conn.execute(s)
+        if not s:
+            continue
+        try:
+            conn.execute(s)
         except sqlite3.OperationalError as e:
-            if "duplicate column" not in str(e): raise
+            if "duplicate column" not in str(e):
+                raise
 
 def main():
     local = sqlite3.connect(DB)
@@ -88,11 +104,14 @@ def main():
          WHERE district IS NOT NULL AND register_date IS NOT NULL""").fetchall()
     by_d = {}
     for d, r, p, c in rows:
-        try: rd = dt.date.fromisoformat(r[:10])
-        except: continue
+        try:
+            rd = dt.date.fromisoformat(r[:10])
+        except ValueError:
+            continue
         bucket = "recent" if r >= cut_recent else ("prior" if r >= cut_prior else None)
-        if not bucket: continue
-        by_d.setdefault(d, {"recent":[], "prior":[], "corpuses":set()})
+        if not bucket:
+            continue
+        by_d.setdefault(d, {"recent": [], "prior": [], "corpuses": set()})
         by_d[d][bucket].append(rd)
         by_d[d]["corpuses"].add((p, c))
 
@@ -100,7 +119,15 @@ def main():
         n_corp = max(len(info["corpuses"]), 1)
         v_rec  = len(info["recent"])  / 6 / n_corp
         v_prior = len(info["prior"]) / 6 / n_corp
-        ratio  = (v_rec / v_prior) if v_prior > 0 else (None if v_rec == 0 else 2.0)
+        # v_prior > 0: real ratio; v_prior == 0 && v_rec == 0: neutral (None→1.0);
+        # v_prior == 0 && v_rec > 0: synthesise ratio scaled by v_rec magnitude so
+        # a tiny v_rec stays near 1.0 while a large v_rec approaches _TREND_CAP_VPRIOR_ZERO.
+        if v_prior > 0:
+            ratio: float | None = v_rec / v_prior
+        elif v_rec == 0:
+            ratio = None
+        else:
+            ratio = 1.0 + min(1.0, v_rec / _REF_VELOCITY) * (_TREND_CAP_VPRIOR_ZERO - 1.0)
         # Prod-style clamping
         trend_factor = max(0.7, min(2.0, ratio)) if ratio else 1.0
         # sat_factor: sold_pct in district. >50% = mature market (prod logic +30%