domrf_kn_flats версионируется (UNIQUE(id, snapshot_date), м.50), scraper
UPSERT per snapshot — то же что для domrf_kn_objects (которое в L3 supply
после #1212 берём только latest). _AVG_PRICE_SQL фильтра snapshot_date НЕ
имел → AVG усреднял ИСТОРИЮ цен (stale на растущем рынке) → UI-поле
Competitor.avg_price_per_m2 + вход _price_similarity получали устаревшую
цену. COUNT '%прод%' множил sold ×N снапшотов → raw_sold/flat_count кратно
завышен → попадал в гард-нейтраль 0.5 или искажал stage_at_horizon как
×N-завышенный sold_pct.
Patch: WHERE f.snapshot_date = (SELECT MAX(snapshot_date) FROM domrf_kn_flats).
Зеркало паттерна best_layouts._SUPPLY_BATCH_SQL и _COMPETITORS_SQL DISTINCT ON
(уже было latest). 51/51 competitors-тестов зелёные.
Closes#1210
_ACTIVE_STATUSES = frozenset({"sales", "construction"}) — английский словарь
никогда не совпадал с domrf_kn_objects.site_status, который scraper берёт
СЫРЫМ из siteStatus дом.рф (domrf_kn.py:316). Реальные prod-значения
русские: «Строящиеся»/«Сданные».
Прод-аудит:
- data/sql/105_add_sales_started_flag.sql фильтрует по 'Строящиеся' (~1322 строки).
- partial index 66_indexes_recommend.sql использует те же.
- analytics_queries.py, MarketTab.tsx, CompetitorTable.tsx — все на русских.
Эффект: у ВСЕХ Competitor в POST /parcels/{cad}/competitors is_active=False
и CompetitorsSummary.active_count=0 при любых данных — типизированный
контракт систематически врал.
Patch: _ACTIVE_STATUSES = frozenset({"Строящиеся"}). Заодно обновил два
unit-теста которые кодировали баг (использовали "sales"/"construction"
в моках, тестировали логику против сломанного словаря). Теперь моки
матчат реальную prod-форму.
51/51 competitors-тестов зелёные. ruff clean.
Closes#1213
Cold §22 forecast measured ~215-233s on prod: §9.x layers re-execute the same
horizon/segment-invariant DB loads with identical args hundreds of times per
report (profiled: get_competitors x69, market_metrics x124, get_monthly_macro
x290). Add a per-report ContextVar cache (forecast_cache(), opened once in the
orchestrator) + @cached(key_builder) on the expensive §9.x loaders so each
unique load runs ONCE and reuses the same frozen, read-only instance.
Output is byte-identical (memoized producers are frozen dataclasses / read-only
Pydantic, callers never mutate; cache is per-report, discarded on exit; no-op
outside the report build). No concurrency, no signature changes.
- forecast_request_cache.py: ContextVar cache + cached() decorator (no-op
outside context, reentrant, _MISS sentinel for cached None)
- @cached on competitors/future_supply/market_metrics/macro_series/
sales_series/macro_coefficient/demand_normalization/regression loaders
- orchestrator: wrap build_site_finder_report in forecast_cache()
- 58 tests: key discrimination (call-counting regression guard), no-op-outside,
per-report isolation, reentrancy, frozen-producer canary, amplification proof
(real get_monthly_macro xN->1)
code-reviewer APPROVE (keys correct, mutation-safe, output identical). 1265
forecast/cache tests green. No new deps. Refs #1129.
weighted_avg_velocity was a naive mean despite the name — a 500-flat ЖК weighed
the same as a 20-flat one. Now count-weighted by flats_total (sql.md AVG
principle): Σ(velocity*flats_total)/Σ(flats_total). Competitors with unknown
flats_total are excluded from weights; if sizes are unknown for ALL, graceful
fallback to the simple mean (den>0 guard). Field name + API contract UNCHANGED
(zero consumer ripple — traced: only CompetitorsSummary, no frontend ref).
Tests: equal sizes → weighted==naive (existing 6.0 stays); NEW test with
500-flat@40 + 20-flat@2 → 38.54 (not naive 21.0), proving the weighting.
domrf_kn_flats.status is NULL in ~99.8% of rows, so WHERE status='sold'
always returned 0 rows and avg_price_per_m2 was always None. Drop the
filter; AVG over all rows with price_per_m2 IS NOT NULL is semantically
correct for a complex-level price estimate.
Adds regression test test_competitors_avg_price_populated (Issue #227).
