feat(parcels): best-layouts endpoint + service (#113 PR C) #196

Merged

lekss361 merged 3 commits from feat/113-best-layouts-endpoint into main

2026-05-16 09:10:07 +00:00

Author SHA1 Message Date

Author	SHA1	Message	Date
lekss361	8488012f05	fix(best-layouts): address review-bot minor items (#113 PR C) 3 carryover 🟡 items из re-review verdict на `a88917f`: 1. Coverage % denominator now post-filter (was mixing pre-filter count с post-filter intersection → artificially low confidence when user passed exclude/filter list) 2. MAX(snapshot_date) pre-computed via db.scalar() + bind :latest_snap instead of subquery in WHERE (avoid re-execution per spatial scan) 3. LAYOUT_CONFIDENCE_HIGH_PCT / LAYOUT_CONFIDENCE_MEDIUM_PCT as module-level constants instead of hardcoded literals Integration test gap (#4 в re-review) — отдельный follow-up issue, не в scope этого PR. Tests: 10/10 best-layouts pass, 43/43 regression OK.	2026-05-16 12:01:09 +03:00
lekss361	a88917f0cf	fix(best-layouts): replace non-existent geom_3857 with on-the-fly geography (#113 PR C) domrf_kn_objects has no geom_3857 column (only latitude/longitude NUMERIC). Switch _COMPETITORS_IN_RADIUS_SQL and _SUPPLY_BATCH_SQL to use ST_SetSRID(ST_MakePoint(longitude, latitude), 4326)::geography — consistent with competitors.py pattern. Rename centroid params x_3857/y_3857 → center_lon/center_lat. Remove dead _round_to_5 function. Fixes review-bot 🔴 BLOCK on `fdc6456` PR #196. Tests: 10/10 pass (mock keys updated к center_lon/center_lat), 43/43 regression OK.	2026-05-16 11:52:08 +03:00
lekss361	fdc64569d2	feat(parcels): best-layouts endpoint + service (#113 PR C) POST /api/v1/parcels/{cad_num}/best-layouts — top layout patterns (room_bucket × area_bin) у конкурентов в радиусе с ranking by velocity из mv_layout_velocity + unit-mix recommendation для архитектурного ТЗ. Service (backend/app/services/site_finder/best_layouts.py, 563 LOC): - centroid lookup с fallback (cad_parcels → cad_quarters) - competitors-in-radius через ST_DWithin - aggregate по mv_layout_velocity per room_bucket - supply batch query (один SQL вместо N) для sold_pct_of_supply - Hamilton largest-remainder normalizer (sum_pct=100 invariant) Endpoint (backend/app/api/v1/parcels.py, +24 LOC) после /competitors. Tests (backend/tests/api/v1/test_parcel_best_layouts.py, 10 mock tests): 404, empty competitors, MV empty, exclude/filter, min_velocity, time_window scaling, obj_class_filter, target_total_flats, 3-bucket ranking. Phase 2.1 (без layout_type / balcony_count — отсутствуют в БД, ждут B2B Объектив #52). Reuses PR A schemas/helpers, PR B MV. NOTE: competitors.py:143 status='sold' carryover bug НЕ fixed — отдельный follow-up issue. PR C использует MV (через objective_corpus_room_month) что работает корректно.	2026-05-16 11:42:20 +03:00

lekss361

8488012f05

fix(best-layouts): address review-bot minor items (#113 PR C)

3 carryover 🟡 items из re-review verdict на a88917f:

1. Coverage % denominator now post-filter (was mixing pre-filter count
   с post-filter intersection → artificially low confidence when
   user passed exclude/filter list)
2. MAX(snapshot_date) pre-computed via db.scalar() + bind :latest_snap
   instead of subquery in WHERE (avoid re-execution per spatial scan)
3. LAYOUT_CONFIDENCE_HIGH_PCT / LAYOUT_CONFIDENCE_MEDIUM_PCT as
   module-level constants instead of hardcoded literals

Integration test gap (#4 в re-review) — отдельный follow-up issue,
не в scope этого PR.

Tests: 10/10 best-layouts pass, 43/43 regression OK.

2026-05-16 12:01:09 +03:00

lekss361

a88917f0cf

fix(best-layouts): replace non-existent geom_3857 with on-the-fly geography (#113 PR C)

domrf_kn_objects has no geom_3857 column (only latitude/longitude NUMERIC).
Switch _COMPETITORS_IN_RADIUS_SQL and _SUPPLY_BATCH_SQL to use
ST_SetSRID(ST_MakePoint(longitude, latitude), 4326)::geography — consistent
with competitors.py pattern. Rename centroid params x_3857/y_3857 →
center_lon/center_lat. Remove dead _round_to_5 function.

Fixes review-bot 🔴 BLOCK on fdc6456 PR #196.

Tests: 10/10 pass (mock keys updated к center_lon/center_lat), 43/43 regression OK.

2026-05-16 11:52:08 +03:00

lekss361

fdc64569d2

feat(parcels): best-layouts endpoint + service (#113 PR C)

POST /api/v1/parcels/{cad_num}/best-layouts — top layout patterns
(room_bucket × area_bin) у конкурентов в радиусе с ranking by velocity
из mv_layout_velocity + unit-mix recommendation для архитектурного ТЗ.

Service (backend/app/services/site_finder/best_layouts.py, 563 LOC):
- centroid lookup с fallback (cad_parcels → cad_quarters)
- competitors-in-radius через ST_DWithin
- aggregate по mv_layout_velocity per room_bucket
- supply batch query (один SQL вместо N) для sold_pct_of_supply
- Hamilton largest-remainder normalizer (sum_pct=100 invariant)

Endpoint (backend/app/api/v1/parcels.py, +24 LOC) после /competitors.

Tests (backend/tests/api/v1/test_parcel_best_layouts.py, 10 mock tests):
404, empty competitors, MV empty, exclude/filter, min_velocity, time_window
scaling, obj_class_filter, target_total_flats, 3-bucket ranking.

Phase 2.1 (без layout_type / balcony_count — отсутствуют в БД,
ждут B2B Объектив #52). Reuses PR A schemas/helpers, PR B MV.

NOTE: competitors.py:143 status='sold' carryover bug НЕ fixed —
отдельный follow-up issue. PR C использует MV (через
objective_corpus_room_month) что работает корректно.

2026-05-16 11:42:20 +03:00