gendesign/tradein-mvp/backend/tests/fixtures/backtest_baseline.json
bot-backend 86cb294949
All checks were successful
CI Trade-In / changes (pull_request) Successful in 13s
CI / changes (pull_request) Successful in 13s
CI Trade-In / browser-tests (pull_request) Has been skipped
CI Trade-In / frontend-checks (pull_request) Has been skipped
CI / backend-tests (pull_request) Has been skipped
CI / frontend-tests (pull_request) Has been skipped
CI / openapi-codegen-check (pull_request) Has been skipped
CI Trade-In / backend-tests (pull_request) Successful in 4m6s
test(tradein/backtest): эталон гейта после снятия квартала соседа
Гейт `test_backtest_regression_gate` реплеит замороженную фикстуру (277 сделок)
прямо через `_price_from_inputs` с `target_house_cadnum=None`, минуя
`estimate_quality`. То есть на фикстуре меряется ровно снятие поправки по кварталу
СОСЕДА (координатный резолв в реплей не попадает — его делает вызывающий код).

Затронуто 30 сделок из 277 (10.8% — сходится с прод-оценкой поверхности ~10%);
множитель цены стало/было: min 0.704, медиана 1.019, max 1.661 (17 вверх / 13 вниз).

Сдвиг метрик (эталон → стало):
  expected_sold.overall.mape_pct   13.18 → 12.63  (−0.55)
  calibration.low.mape_pct          13.2 → 12.67  (−0.53)
  студии mape                      19.38 → 16.96  (−2.42)
  3к mape                           9.34 →  7.79  (−1.55)
  2к mape                          18.26 → 17.39  (−0.87)
  комфорт mape                     11.80 → 10.14  (−1.66)
  бизнес mape                      14.65 → 13.65  (−1.00)
УХУДШИЛОСЬ:
  эконом mape                      13.71 → 14.20  (+0.49), bias 2.54 → 3.33
  4+ mape                          16.30 → 16.38  (+0.08), bias 3.34 → −1.32
  хвосты: студии p75 34.32 → 38.82, бизнес p25 −22.93 → −27.15

Покрытие (82.09/82.16), sharpness (0.743), n по всем бакетам и headline не менялись.
Атрибуция: на origin/main регенерация тем же путём даёт эталон байт-в-байт,
расхождение появляется ровно от правки резолва квартала.
2026-08-12 20:35:52 +05:00

154 lines
3 KiB
JSON

{
"calibration": {
"high": {
"coverage_pct": null,
"mape_pct": null,
"n": 0,
"n_covered": 0
},
"low": {
"coverage_pct": 82.09,
"mape_pct": 12.67,
"n": 276,
"n_covered": 220
},
"medium": {
"coverage_pct": 100.0,
"mape_pct": 6.99,
"n": 1,
"n_covered": 1
}
},
"confidence_order": [
"high",
"medium",
"low"
],
"expected_sold": {
"overall": {
"mape_pct": 12.63,
"median_bias_pct": -3.74,
"n": 269,
"n_no_analogs": 0,
"p25_pct": -15.17,
"p75_pct": 8.67
},
"per_rooms": {
"0": {
"label": "студия",
"mape_pct": 16.96,
"median_bias_pct": 16.96,
"n": 35,
"n_no_analogs": 0,
"p25_pct": 1.5,
"p75_pct": 38.82
},
"1": {
"label": "1к",
"mape_pct": 11.06,
"median_bias_pct": -4.61,
"n": 93,
"n_no_analogs": 0,
"p25_pct": -14.43,
"p75_pct": 8.15
},
"2": {
"label": "2к",
"mape_pct": 17.39,
"median_bias_pct": -11.71,
"n": 74,
"n_no_analogs": 0,
"p25_pct": -23.07,
"p75_pct": -0.36
},
"3": {
"label": "3к",
"mape_pct": 7.79,
"median_bias_pct": -4.05,
"n": 43,
"n_no_analogs": 0,
"p25_pct": -10.26,
"p75_pct": 3.82
},
"4": {
"label": "4+",
"mape_pct": 16.38,
"median_bias_pct": -1.32,
"n": 24,
"n_no_analogs": 0,
"p25_pct": -14.91,
"p75_pct": 15.41
}
},
"per_segment": {
"бизнес": {
"mape_pct": 13.65,
"median_bias_pct": -10.54,
"n": 46,
"p25_pct": -27.15,
"p75_pct": -1.31
},
"комфорт": {
"mape_pct": 10.14,
"median_bias_pct": -5.01,
"n": 101,
"p25_pct": -15.49,
"p75_pct": 4.37
},
"премиум": {
"mape_pct": 68.92,
"median_bias_pct": -68.92,
"n": 1,
"p25_pct": -68.92,
"p75_pct": -68.92
},
"эконом": {
"mape_pct": 14.2,
"median_bias_pct": 3.33,
"n": 115,
"p25_pct": -8.69,
"p75_pct": 27.3
},
"элит": {
"mape_pct": 33.2,
"median_bias_pct": -33.2,
"n": 6,
"p25_pct": -42.75,
"p75_pct": -22.4
}
}
},
"headline": {
"ask_median_ppm2": 145883.6593586467,
"deal_median_ppm2": 125063.0,
"spread_pct": 16.65
},
"range_coverage": {
"overall": {
"coverage_pct": 82.16,
"n": 269,
"n_covered": 221
},
"per_confidence": {
"high": {
"coverage_pct": null,
"n": 0,
"n_covered": 0
},
"low": {
"coverage_pct": 82.09,
"n": 268,
"n_covered": 220
},
"medium": {
"coverage_pct": 100.0,
"n": 1,
"n_covered": 1
}
}
},
"sharpness": {
"median_rel_width": 0.743,
"n": 269
}
}