feat(tradein/estimator): calibrated 80% prediction interval for expected_sold range (#1966)

The expected_sold_range_low/high were derived from the analog asking-IQR
(× asking→sold ratio): only ~55% of actual sold prices fell inside the stated
range — a 50%-ish interval mislabeled as the estimate's range. Replace with a
calibrated band around the expected_sold POINT, gated behind
estimate_calibrated_pi_enabled (default ON):

  range_low  = expected_sold_price × 0.649  (empirical p10 of sold/expected_sold)
  range_high = expected_sold_price × 1.392  (empirical p90, n=2366 prod deals)

This is a genuine ~80% prediction interval (verified 80.0% coverage on the
2366-deal sample). Headline asking range_low/high (market-asking IQR) untouched.

Frozen backtest baseline regenerated: range_coverage.overall 55.23 → 80.14
(n_covered 153 → 222); expected_sold POINT mape/bias and headline unchanged
(only the band moved). Tests asserting the old asking-IQR × ratio band updated
to the calibrated band (documented behavior change).

tradein-mvp/backend/app/core/config.py

@@ -96,6 +96,16 @@ class Settings(BaseSettings):
     estimate_sb_floor_sigma: float = 0.25
     estimate_sb_guardrail_tol: float = 0.05  # hard floor: est ≥ min(comp ppm²)×(1−tol)
     estimate_sb_mad_k: float = 3.5  # MAD-clip: drop comps с |ppm2−median| > k×MAD
+    # ── #1966: honest calibrated prediction-interval для expected_sold range ─────
+    # Старый expected_sold_range производился из IQR аналогов (asking-IQR × ratio):
+    # ~55% реальных продаж попадали в заявленный «диапазон оценки» (де-факто 50%-й
+    # интервал, выданный за полный). Эмпирически отношение actual_sold/expected_sold
+    # по 2366 прод-сделкам имеет p10=0.649, p90=1.392 → band
+    # expected_sold × [low_mult, high_mult] = настоящий ~80% prediction interval
+    # (проверено: 80.0% coverage на тех же 2366). OFF ⇒ точно старое IQR-поведение.
+    estimate_calibrated_pi_enabled: bool = True
+    estimate_pi_low_mult: float = 0.649  # empirical p10 of sold/expected_sold (#1966, n=2366)
+    estimate_pi_high_mult: float = 1.392  # empirical p90 of sold/expected_sold (#1966, n=2366)
     # ── #1795: premium headline anti-inflation (4 фикса, каждый за флагом) ──────
     # Диагноз: бизнес/премиум headline завышается ~2× vs медиана реальных ДКП
     # (Малышева 30 = 296k при median сделок 138k). Эконом/комфорт сходятся ±5%.

tradein-mvp/backend/app/services/estimator.py

@@ -2507,6 +2507,17 @@ def _price_from_inputs(
             effective_ratio = 1.0
         expected_sold_per_m2 = round(median_ppm2 * effective_ratio)
         expected_sold_price = round(median_price * effective_ratio)
+        if settings.estimate_calibrated_pi_enabled and expected_sold_price:
+            # #1966: калиброванный ~80% prediction interval вокруг ТОЧКИ expected_sold.
+            # Эмпирически отношение actual_sold_ppm2 / expected_sold_per_m2 по 2366
+            # прод-сделкам: p10=0.649, p90=1.392 → band [×low_mult, ×high_mult] даёт
+            # ~80% honest coverage. Старый IQR-производный band (asking-IQR × ratio)
+            # покрывал лишь ~55% реальных продаж при заявленном «диапазоне оценки».
+            # Множители low<1<high (дефолты) гарантируют low ≤ point ≤ high.
+            expected_sold_range_low = round(expected_sold_price * settings.estimate_pi_low_mult)
+            expected_sold_range_high = round(expected_sold_price * settings.estimate_pi_high_mult)
+        else:
+            # legacy: диапазон производный от IQR аналогов (asking-IQR × ratio).
             expected_sold_range_low = round(range_low * effective_ratio)
             expected_sold_range_high = round(range_high * effective_ratio)
 

tradein-mvp/backend/tests/fixtures/backtest_baseline.json

@@ -7,16 +7,16 @@
       "n_covered": 0
     },
     "low": {
-      "coverage_pct": 55.27,
+      "coverage_pct": 80.0,
       "mape_pct": 18.63,
       "n": 275,
-      "n_covered": 152
+      "n_covered": 220
     },
     "medium": {
-      "coverage_pct": 50.0,
+      "coverage_pct": 100.0,
       "mape_pct": 19.18,
       "n": 2,
-      "n_covered": 1
+      "n_covered": 2
     }
   },
   "confidence_order": [
@@ -125,9 +125,9 @@
   },
   "range_coverage": {
     "overall": {
-      "coverage_pct": 55.23,
+      "coverage_pct": 80.14,
       "n": 277,
-      "n_covered": 153
+      "n_covered": 222
     },
     "per_confidence": {
       "high": {
@@ -136,19 +136,19 @@
         "n_covered": 0
       },
       "low": {
-        "coverage_pct": 55.27,
+        "coverage_pct": 80.0,
         "n": 275,
-        "n_covered": 152
+        "n_covered": 220
       },
       "medium": {
-        "coverage_pct": 50.0,
+        "coverage_pct": 100.0,
         "n": 2,
-        "n_covered": 1
+        "n_covered": 2
       }
     }
   },
   "sharpness": {
-    "median_rel_width": 0.4638,
+    "median_rel_width": 0.743,
     "n": 277
   }
 }

tradein-mvp/backend/tests/test_estimator_expected_sold.py

@@ -228,11 +228,24 @@ def test_expected_sold_applied_when_ratio_present() -> None:
     assert est.median_price_rub == expected_headline
     assert est.median_price_per_m2 == 150_000
 
-    # Parallel expected_sold = asking × ratio (round()).
+    # Parallel expected_sold point = asking × ratio (round()).
     assert est.expected_sold_price_rub == round(est.median_price_rub * ratio)
     assert est.expected_sold_per_m2 == round(est.median_price_per_m2 * ratio)
-    assert est.expected_sold_range_low_rub == round(est.range_low_rub * ratio)
-    assert est.expected_sold_range_high_rub == round(est.range_high_rub * ratio)
+    # #1966: expected_sold range is now a calibrated ~80% PI around the POINT
+    # (point × [p10, p90] of sold/expected_sold), not the old asking-IQR × ratio band.
+    from app.core.config import settings
+
+    assert est.expected_sold_range_low_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_low_mult
+    )
+    assert est.expected_sold_range_high_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_high_mult
+    )
+    assert (
+        est.expected_sold_range_low_rub
+        <= est.expected_sold_price_rub
+        <= est.expected_sold_range_high_rub
+    )
     assert est.asking_to_sold_ratio == ratio
     assert est.ratio_basis == "per_rooms"
 
@@ -342,8 +355,15 @@ def test_expected_sold_fires_on_anchor_only_no_radius_comps() -> None:
     assert est.expected_sold_price_rub > 0
     assert est.expected_sold_price_rub == round(est.median_price_rub * ratio)
     assert est.expected_sold_per_m2 == round(est.median_price_per_m2 * ratio)
-    assert est.expected_sold_range_low_rub == round(est.range_low_rub * ratio)
-    assert est.expected_sold_range_high_rub == round(est.range_high_rub * ratio)
+    # #1966: calibrated ~80% PI around the expected_sold point (not asking-IQR × ratio).
+    from app.core.config import settings
+
+    assert est.expected_sold_range_low_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_low_mult
+    )
+    assert est.expected_sold_range_high_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_high_mult
+    )
 
 
 def test_expected_sold_none_when_median_price_zero() -> None:

tradein-mvp/backend/tests/test_estimator_expected_sold_clamp.py

@@ -133,14 +133,25 @@ def test_expected_sold_clamped_to_headline_when_ratio_above_1(
     assert (
         est.expected_sold_per_m2 <= est.median_price_per_m2
     ), "expected_sold_per_m2 превышает median_price_per_m2"
+    # #1966: expected_sold range is now a calibrated ~80% PI around the point
+    # (point × [p10, p90] of sold/expected_sold), so it is NO LONGER bounded by the
+    # asking-IQR band — the high arm (point × 1.392) legitimately exceeds range_high.
+    # The invariant we keep is the calibrated band itself + ordering low ≤ point ≤ high.
+    from app.core.config import settings
+
     assert est.expected_sold_range_high_rub is not None
-    assert (
-        est.expected_sold_range_high_rub <= est.range_high_rub
-    ), "expected_sold_range_high > range_high"
+    assert est.expected_sold_range_high_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_high_mult
+    )
     assert est.expected_sold_range_low_rub is not None
+    assert est.expected_sold_range_low_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_low_mult
+    )
     assert (
-        est.expected_sold_range_low_rub <= est.range_low_rub
-    ), "expected_sold_range_low > range_low"
+        est.expected_sold_range_low_rub
+        <= est.expected_sold_price_rub
+        <= est.expected_sold_range_high_rub
+    )
     assert est.expected_sold_price_rub == est.median_price_rub
     assert est.expected_sold_per_m2 == est.median_price_per_m2
     clamp_logs = [r for r in caplog.records if "clamped" in r.getMessage()]
@@ -154,8 +165,15 @@ def test_expected_sold_not_clamped_when_ratio_below_1() -> None:
     assert est.median_price_rub > 0
     assert est.expected_sold_price_rub == round(est.median_price_rub * ratio)
     assert est.expected_sold_per_m2 == round(est.median_price_per_m2 * ratio)
-    assert est.expected_sold_range_low_rub == round(est.range_low_rub * ratio)
-    assert est.expected_sold_range_high_rub == round(est.range_high_rub * ratio)
+    # #1966: calibrated ~80% PI around the expected_sold point (not asking-IQR × ratio).
+    from app.core.config import settings
+
+    assert est.expected_sold_range_low_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_low_mult
+    )
+    assert est.expected_sold_range_high_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_high_mult
+    )
     assert est.expected_sold_price_rub < est.median_price_rub
 
 

tradein-mvp/backend/tests/test_estimator_imv_blend.py

@@ -413,8 +413,15 @@ def test_expected_sold_consistent_with_blended_median() -> None:
     # expected_sold выведен из POST-blend значений — no stale.
     assert est.expected_sold_price_rub == round(est.median_price_rub * ratio)
     assert est.expected_sold_per_m2 == round(est.median_price_per_m2 * ratio)
-    assert est.expected_sold_range_high_rub == round(est.range_high_rub * ratio)
-    assert est.expected_sold_range_low_rub == round(est.range_low_rub * ratio)
+    # #1966: expected_sold range — калиброванный ~80% PI вокруг точки, не asking-IQR × ratio.
+    from app.core.config import settings
+
+    assert est.expected_sold_range_high_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_high_mult
+    )
+    assert est.expected_sold_range_low_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_low_mult
+    )
     # Sanity: sold < asking (ratio<1), но НЕ абсурдная «скидка» от stale 6М-базы.
     assert est.expected_sold_price_rub == round(12_000_000 * ratio)
 

tradein-mvp/backend/tests/test_same_building_anchor.py

@@ -565,8 +565,16 @@ def test_estimate_expected_sold_distinct_after_anchor() -> None:
     # поэтому сравниваем с допуском ±1 на округление float→int.
     assert est.expected_sold_price_rub == round(est.median_price_rub * ratio)
     assert abs(est.expected_sold_per_m2 - round(est.median_price_per_m2 * ratio)) <= 1
-    assert est.expected_sold_range_high_rub == round(est.range_high_rub * ratio)
-    assert est.expected_sold_range_low_rub == round(est.range_low_rub * ratio)
+    # #1966: expected_sold range — калиброванный ~80% PI вокруг точки (point × [p10,p90]
+    # sold/expected_sold), не asking-IQR × ratio.
+    from app.core.config import settings
+
+    assert est.expected_sold_range_high_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_high_mult
+    )
+    assert est.expected_sold_range_low_rub == round(
+        est.expected_sold_price_rub * settings.estimate_pi_low_mult
+    )
     # Distinct: sold строго ниже asking (ratio < 1) — фикс двойных идентичных чисел.
     assert est.expected_sold_price_rub < est.median_price_rub
     assert est.expected_sold_per_m2 < est.median_price_per_m2