- admin_scrape.py /failures limit: ge=0 → ge=1 (returning 0 rows is nonsensical for a list endpoint)
- orchestrator.py: check result.finish_reason after ok=True; 'length'/'content_filter' and any
other non-stop reason triggers deterministic fallback with reason=finish_reason:<value>
(LLMResult docstring mandated this but no consumer enforced it)
- Add window_months > 0 guard to vel_by_room dict comprehension (mirrors _monthly_rate)
- Correct outer comment in recommendation.py: honest-zero for known-zero buckets,
fallback only when velocity_by_room=None or bucket absent from _FORECAST_TO_METRIC_BUCKETS
- Add comment in as_dict() noting velocity_by_room is intentionally not serialized
(internal pipeline attr consumed directly by recommendation.py)
Introduce INFORMATIONAL_WARNING_CODES frozenset; verdict assembly now only
downgrades to «С ограничениями» when verdict-relevant warnings exist.
NO_ENGINEERING_NEARBY still appears in warnings[] for display but a clean
residential parcel (Ж-zone, no ЗОУИТ) correctly receives verdict_label=«Можно».
`_build_confidence` was calling `compute_report_confidence` without
`deal_count_months`, so the «за N мес» suffix never appeared in
production. Add `_deal_count_months(market_metrics)` extractor
(reads `window_months`, same key as `_history_months`) and pass it.
Also fix pre-existing UP038 violations (isinstance tuple → X | Y).
Add `velocity_by_room: dict[str, float] | None` to `MarketMetrics` — per-bucket
unit velocity (ед./мес) derived from the existing `sold_by_room` ROLLUP data that
`_query_sales_window` already returns. No new SQL required.
Thread per-bucket velocity through `_demand_only_overlay` via the new
`_FORECAST_TO_METRIC_BUCKETS` constant that maps each forecast bucket to its
market_metrics room-bucket keys. "80+ м²" sums "4" + "5+" keys. Fallback to
aggregate `unit_velocity` when `velocity_by_room` is None (thin-data path).
Previously `base_pace` was identical for all 5 room-buckets, so §9.4 norm and §9.2
base_pace cancelled out in pace/max_pace and ranking was driven purely by §9.5
macro_coef (segment steepness proxy). Now §9.2 reflects real per-bucket observed
demand from objective_lots.contract_date data.
Callers of `compute_market_metrics` that don't use `velocity_by_room` are unaffected
(the new field is additive to the frozen dataclass). All existing callers verified —
none construct `MarketMetrics` directly except the one production site.
Both metrics were querying prinzip_deals without any date filter,
returning all-time figures while the surrounding stats (leads_window,
converted_window, conv_pct_window) were scoped to the last N months.
Now both subqueries restrict to deals linked to leads in window_leads
(via deal_id IN (...)), making all «за период» figures consistent.
_GEO_WEIGHT_UNKNOWN was 0.1, which equals exp(−6.9/3)≈0.10 (weight of a
confirmed-far project at ~6.9 km). Projects beyond that distance got a
weight *below* 0.1, meaning unknown-coordinate projects outweighed
confirmed-far ones — an inversion of the documented intent.
Lowered to 0.05 (≈ exp(−3) = exp(−9 km / scale)), restoring the correct
hierarchy: confirmed-close > confirmed-far > unknown. Updated TestGeoWeight
(hardcoded 0.1 expectation) and TestCannibalizationTrueMode (overlap ×
floor comment/value) accordingly. Added two new assertions in TestGeoWeight
that enforce the hierarchy monotonically and verify unknown < exp(−6.9/3).
_INN_RE now requires an explicit ИНН/inn keyword anchor (case-insensitive) within
~20 chars before a 10/12-digit block. Bare digit sequences without the keyword are
no longer candidates — eliminates false-positives on large monetary amounts such as
1 200 000 000 (which coincidentally passes the ФНС checksum). Checksum validation
is kept as a second gate to avoid redacting e.g. «ИНН 1234567890» with bad digits.
_inn_repl updated to use match.group(1) (digit-only capture group) instead of the
full match that now also includes the keyword prefix.
7 new regression tests in test_redaction.py: bare large numbers not redacted,
keyword-cued real INNs (10/12 digit) still redacted, bad-checksum + keyword left
intact, latin «inn:» accepted.
Replace whole-HTML re.search for status with a 3-level block-scoped strategy:
1. CSS badge classes (status/badge/tag/chip/label) — highest precision.
2. Proximity to block labelled «Статус» via _find_text_near.
3. Full blocks scan where sold/reserved keywords always beat free —
preventing «в продаже» from nav/similar-flats sections from
misclassifying sold flats as free.
Add _STATUS_KW_RE and _classify_status_kw at module level with full
morphological coverage: продан/продана/продано, реализован[аоы]?,
забронирован[аоы]?, свободн[аоы]?.
Add 31 tests in test_domrf_catalog_parse.py covering all three
extraction levels plus regression for fem. word-form «Квартира продана».
_YEAR_PERIODS ('год') → month=1 collided with Q1 ('I квартал') → month=1
for the same region/year: both produced obs_date=YYYY-01-01, so the second
series in the XML silently overwrote the first in the in-memory by_key dict.
Fix: add _emiss_period_granularity() and extend the dedup key from
(region, obs_date) to (region, obs_date, granularity). 'year' and 'quarter'
are now distinct slots, so both rows survive the parser and reach the upsert.
Test: test_yearly_and_q1_both_survive_dedup verifies len==2 with both
values present; test_period_granularity covers all classification branches.
_count_full_years treated units=0 as a valid observation, so a series
where fill_month_grid zero-filled every month still accumulated 3 full
years and passed the _MIN_FULL_YEARS guard. Zero-filled months carry no
seasonal signal, so they must be skipped in the year counter — the same
way None values already were.
Fix: skip v==0 alongside v is None in _count_full_years.
Add four tests: zero-filled 36-month series → n_full_years=0/applied=False;
partial-coverage years (only 6 non-zero months/year) → not counted as full;
real non-zero series still passes guard; normalize_demand on zero-filled
SalesSeries returns series unchanged.
Add `deal_count_months: int | None = None` to `compute_report_confidence`.
When provided, threads it as suffix into `_factor_from_count` so the
deal_count ConfidenceFactor note reads «7 сделок за 6 мес — мало» instead
of the windowless «7 сделок — мало». Existing callers unaffected (default None).
Tests: two new cases in TestComputeReportConfidence — with/without window.
NSPD quarter-dump territorial_zones (layer 875838) does NOT store ПЗЗ letter
codes in properties — zone_code receives the cadastral reg number ("66:41-7.14")
or None, never a "Ж-1" style code. The old prefix check (startswith "Ж-"/"Ж1" …)
therefore never matched real NSPD data, silently marking every dump-sourced parcel
as non-residential (PZZ_NOT_RESIDENTIAL blocker).
Fix: three-tier detection in is_residential_zone:
1. Regex ^Ж on zone_code — handles PKK6/pzz_zones_ekb sources ("Ж-1", "Жс")
2. NSPD subcategory from raw_props — subcategory=2 (Жилые зоны, 221 ЕКБ objects)
and subcategory=3 (Смешанное использование, 126 objects) → residential;
subcategory=1 (ИЖС) excluded — МКД not permitted there
3. zone_name substring "жил" — fallback for non-standard sources
compute_gate_verdict now passes nspd_zoning["raw_props"] to is_residential_zone.
Backend now computes poi_weighted_score and per-POI score_contribution in
0..100 range. Normalization denominator = sum of top-7 category weights / 100
(straight-line mode) or sum of top-7 category weights (routing-decay mode).
Frontend stops reconstructing score from raw weight × 100.
Switch portfolio source from domrf_region_aggregates (room_count_type ONE/TWO/THREE/FOUR)
to domrf_flat_area_distribution (area_bucket FROM_0_TO_25…FROM_100), aggregated into the
same 5 area buckets as the deals series. Both series now share the same axis and are
directly comparable. Frontend portfolioMap simplified to direct bucket key lookup.
_compute_confidence docstring claimed "stub до G1/G2/D1/D2" — те эшелоны уже
отгружены, формулировка устарела. Это намеренно отдельная метрика: надёжность
site-finder скоринга (coverage POI/район/рынок), НЕ форсайтный §15
confidence_engine (надёжность прогноза спроса/цены над форсайт-входами,
которых в analyze hot-path нет). Поведение не меняется — резолв документацией.
Closes#1668
Replace bare metro_block placeholder ({"nearest_top3": None}, blocked on a
never-merged 22h metro scraper) with a direct KNN query against osm_poi_ekb
category='metro_stop' (loaded by poi_loader). Returns 3 nearest stations with
name + rounded distance_m; empty list when none nearby (vs None on error),
SAVEPOINT + try/except like adjacent SF-B5 blocks. No frontend consumer yet.
Closes#1667
report_docx и report_pptx рендерили bare _scenario_deficit_index под
захардкоженной шапкой «(12 мес)» — при fallback на другой горизонт подпись
лгала. Переключил оба на _scenario_deficit_cell (как в report_md): per-cell
аннотация «(гор. N мес)» при N != 12. Закрывает code-review major по #1590.
Внутренний recommendation→product_scoring контракт-ключ был мислейблом: величина —
темп распродажи нежилого (sell-through, прокси ликвидности/спроса), а НЕ доля нежилого
в объёме застройки. Переименован ключ + исправлены reason/docstring/комментарии у
потребителя _score_commercial. Числовая логика не изменена. Ключ внутренний (нет
frontend/schema/openapi-потребителей) → rename контракт-безопасен. pytest 171 passed.
Closes#1635
