Add the LLM prose-composition path for the parcel-forecast chat, layered
over the deterministic Step-1 fallback which stays the safety net.
- chat/tools.py: 5 read-only section tools (exec_summary, product_recommendation,
forecast, risks, scenarios) — pure slices of the loaded report dict, no DB/
recompute, graceful on missing sections. market_now (raw analyze blob) and meta
are deliberately NOT exposed -> highest-PII data cannot reach the LLM.
- chat/safe_payload.py: the §19 gate — single place that builds the outbound
SafePayload from a section-aggregate allowlist; honors is_confidential hard-block.
- chat/orchestrator.py: manual tool-call loop with call-cap/termination, real
grounded_in provenance; any LLMResult.ok=False (disabled/timeout/rate_limited/
redaction_refused/call_cap/provider_error/empty) degrades to the deterministic answer.
- llm/prompts.py: versioned chat_system@v1 — answer only from sections, never
fabricate numbers, advisory tone, decline out-of-scope.
- api/v1/chat.py: branch on settings.llm_enabled; sync complete bridged via
run_in_threadpool. Default-off -> deterministic path, no provider built.
- Tests: fake provider only (no network), planted-secret redaction-boundary +
per-reason fallback + call-cap + numbers-from-report coverage.
Refs #957
Step 1 of #957. Answers parcel questions by reading the already-persisted
§22 SiteFinderReport (latest_run_for, schema "1.0", read-only) and returning
templated RU answers with engine numbers verbatim (§16, never fabricated).
Intent routing (explicit or RU keyword match) -> per-section renderers;
graceful on partial/missing sections and pending (no run / DB error) without
500s. Works with llm_enabled=False (llm_used always False); LLM composition
is Step 2. Mounted off /api/v1/chat so rbac_guard auto-requires an authed
known user.
Refs #957
Upgrade the §25.3 cannibalization index from a same-class-competitor proxy to
true own-portfolio overlap: score the recommended candidate segment against the
developer's own portfolio (get_own_portfolio, #1169 PR1) across axes —
audience/class (ordinal distance), price ₽/м² (interval overlap), unit-mix
(L1 similarity), timing (half-life decay) — geo-weighted by haversine proximity
to the parcel, aggregated by geo-weighted soft-max (the strongest nearby
cannibalizer dominates, not a diluting mean). Empty portfolio -> labelled proxy
fallback, confidence forced low, never presented as the true index.
Pure scoring fns unit-tested without DB; None-not-0 on missing axes; thin/
only-current portfolio -> low confidence + §26 note; deterministic (sorted
tie-breaks, no RNG). class+price+geo active now; unit-mix+timing plumbed via
optional params for a follow-up that wires them from the orchestrator horizon.
ruff + mypy clean; 151 special-index tests pass (964 forecasting dir, no regr).
Refs #1169
Foundation PR: unified "our projects" source the §25.3 overlap engine (PR2)
will consume. Two origins normalized to OwnProject (class/timing/price/unit-mix):
- current <- domrf_kn_objects filtered by settings.own_developer_ids (numeric
prefix of composite dev_id; empty -> [] graceful, no DB hit, no hardcoded id)
- future <- new manual-entry own_planned_project entity (migration 148)
Adds OWN_DEVELOPER_IDS config (comma-sep -> list[int], default []),
own_planned_project table (range/unit_mix CHECKs via IMMUTABLE helper, generated
geom), /api/v1/own-projects CRUD (created_by from X-Authenticated-User), and
get_own_portfolio(db). Per-source graceful degradation; psycopg-v3 CAST clean.
Does not touch special_indices.py or parcels.py (out of scope).
Refs #1169
The demand_index DB column comment still said "насыщающее преобразование"
(saturating) after #1167 switched it to city-relative normalization —
exactly the misleading-comment class that hid the original bug. Migration
147 updates COMMENT ON COLUMN (idempotent, comment-only, no data DDL).
Refs #948
demand_index used a fixed clamp01(velocity/50); on a live prod refresh
all 8 ЕКБ districts sold ≥50/mo so it saturated to 1.0 everywhere — zero
discrimination between districts. Redesign to mirror infra_index:
normalize each district's unit_velocity against the city reference (MAX
district velocity per refresh run), so demand always discriminates and
self-calibrates as the market grows (no magic constant to rot).
- normalize_demand(velocity, *, city_reference_velocity), pure + graceful
(None stays None; reference<=0 -> honest 0.0, no ZeroDivisionError)
- refresh_locations now two-pass: collect velocities (one
compute_market_metrics per district, no O(n^2)), derive city reference,
normalize + upsert; SAVEPOINT-per-row and counters preserved
- remove _DEMAND_SATURATION_UPM constant; log city_reference_velocity
- tests: rewrite demand normalization + add end-to-end city-relative
suite incl. discrimination regression guarding the all-1.0 prod bug
Refs #948
Promote district to a first-class `location` entity (ТЗ §8.2), ADDITIVE — no
district->FK refactor. New `location` table keyed by district_name (joinable by
string), carrying 4 normalized [0,1] indices (NULL when no data, never 0):
- infra: _district_poi_score / _city_avg_poi_score (per-district POI aggregate)
- competition: market_metrics.overstock_index (available/stuck competing supply
— orthogonal to demand; NOT sell-through, which is market-heat correlated w/ demand)
- demand: market_metrics.unit_velocity (saturating /50)
- future_supply: future_supply_pressure.index (passthrough, already 0..1)
- data/sql/146_location.sql: idempotent table + UNIQUE(district_name) + range
CHECK + centroid GIST
- services/site_finder/locations.py: compute_location_indices (reuses forecast
per-district fns) + refresh_locations (SAVEPOINT per-row, CAST, ON CONFLICT)
- workers/tasks/location_refresh.py + beat (Mon 07:00 MSK, after supply-layers)
- api/v1/locations.py: read-only GET list + GET by name (analyst+admin via rbac,
frontend pilot-gated)
- tests: 34 (normalization 0..1/null, competition⊥demand orthogonality, idempotent
upsert, read API list/by-name/404)
Part of #948 (Part A insight shipped #1164).
Insight = аналитик пишет free-form intel про локацию/участок с пометкой
«непублично» (is_confidential, §7.13/§8.9). CRUD под /api/v1/insights;
created_by из X-Authenticated-User (не из тела — спуфинг автора невозможен).
Зеркало custom_pois: raw-SQL service + Pydantic v2 schemas, sync def handlers
(threadpool, off event loop).
- data/sql/145_insight.sql: table insight (idempotent, geom GENERATED из lon/lat,
индексы district/cad_num/is_confidential/created_by/GIST); аддитивно расширяет
CHECK audit_log.action += insight_write (тот же constraint ck_audit_log_action,
to_regclass-guarded, superset — не ломает #962-аудит; НЕ правит applied 144).
- audit_middleware.classify_path: method-aware insight_write для POST/PUT/DELETE
/api/v1/insights (GET-чтения не аудируются); обратно совместимо с parcels.
- ACL: backend rbac_guard hard-блокирует только /api/v1/admin/*; pilot отсекается
frontend RouteGuard + Caddy (как parcels/custom_pois). is_confidential — стораемая
пометка без per-record backend-гейта (analyst видит, per #962-политике).
- tests: 28 insight (CRUD+filters+required+#261 commit) + 5 audit. 86 зелёных, ruff.
Part B (Location first-class, §8.2) — отдельный follow-up.
Refs #948.
EPIC18/§19. analyst sees everything (deals, insights, exports, site-finder,
analytics, concept) EXCEPT admin/data-management. Enforcement is backend-hard
(the existing rbac_guard already 403s any non-admin role on /api/v1/admin/*, so
adding the role auto-blocks it) + frontend (deny_paths via /me) + audit.
§19 audit: new best-effort HTTP middleware logs the sensitive actions
(analyze / forecast / forecast-export) to a new audit_log table after the
response. Audit failures never break or delay-fail the request (2-layer
try/except + finally close). Registered INNER to rbac_guard so only authorized
requests are audited (a 403 short-circuits before audit). classify_path matches
export before forecast (anchored).
- auth/roles.yaml: analyst role (paths /**, deny admin-mgmt) + analysttest QA user
- core/auth.py (+ tradein mirror): Role Literal += analyst
- core/audit_middleware.py (new) + main.py registration
- data/sql/144_audit_log.sql (idempotent; auto-applies)
- tests: analyst rbac (403 admin / 200 parcels) + 11 audit cases
No data-level ACL (analyst sees data per policy) -> no #948 dependency. No new
deps. parcels.py untouched. Real analyst logins still need adding to
caddy/users.caddy.snippet (devops).
Refs #962.
fedstat ИПЦ is reCAPTCHA-blocked; CBR publishes inflation openly. Add
fetch_inflation + parse_inflation_xlsx (CBR UniDbQuery DownloadExcel/132934,
monthly % г/г, region=rf, source=cbr) to cbr_macro.py; upsert
indicator_type=inflation_yoy via the existing cbr_macro_sync task (per-series
guard, SAVEPOINT-per-row, CAST not ::, ON CONFLICT on the PK).
Surface inflation_yoy in MonthlyMacro (frozen, carry-forward) and ACTIVATE the
reserved §9.5 inflation channel (macro_coefficient f_inflation: level-vs-4%-target
nudge, non-positive to avoid double-counting f_rate, excluded from
_RATE_DRIVEN_FACTORS). Channel was DEGRADED (no data) -> now BACKED + consumed;
_CONF_HIGH_MIN_BACKED 4->5. Deterministic (§16/§26); renorm claims the reserved
0.08 slice as designed. Live-verified (2026-04 5.58%); 194 macro + 902 forecasting
tests green. No migration, no new deps.
Refs #946.
The legacy resolve_cian_zhk_url hit /zhk/<id>/ which now 404s, leaving the
318 geo-matched cian houses (house_sources.ext_id set, cian_zhk_url NULL)
unfetchable -> the newbuilding-enrichment backfill matched 0 houses.
Add resolve_cian_zhk_url_via_search(nb_id): fetch the cat.php newbuilding SERP
and extract the canonical zhk-<slug>.cian.ru url, ANCHORED on the
<h1 data-name="Title"> header anchor (NOT naive first-match — the SERP carries
promo/recommendation zhk-* links before the title that would otherwise resolve
the wrong ЖК and silently corrupt enrichment). Validated against the real
2.81MB prod SERP + an adversarial poisoned-recommendation test.
Wire into newbuilding_enrich_backfill: broaden selection to "has ext_id OR
cian_zhk_url", resolve+persist the url under a SAVEPOINT before enriching,
rate-limited + resumable + idempotent. Keep the old resolver (deprecated).
Bounded prod proof (5 houses, direct): 5/5 urls resolved+persisted, 3/5 fully
enriched (+18 price_dynamics, +3 reliability); the 2 misses were direct-mode
anti-bot on the 2nd fetch. Full 318-run gated on the cian mobile proxy
(mproxy.site) being restored. code-reviewer APPROVE (SQL/idempotency) +
resolver hardened against wrong-ЖК. 19 tests green.
Refs #972.
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.
950-E1: app/tasks/newbuilding_enrich_backfill.py selects houses linked to
ext_source='cian_newbuilding' with a fetchable cian_zhk_url and runs the existing
fetch_newbuilding + save_newbuilding_enrichment over each, populating
houses_price_dynamics + house_reliability_checks (+ house_reviews when present).
- Idempotent + resumable: skips already-enriched houses; force= re-run dup-safe
(price_dynamics UPSERT via dim_key, reliability dedup guard with id-DESC
tiebreaker, reviews ON CONFLICT (source,ext_review_id)).
- Anti-bot aware: get_scraper_delay('cian') + jitter, SAVEPOINT-per-house
(one captcha/failure logs + continues, never aborts the batch).
- limit= for bounded proof runs; sizing counters (total/fetchable/pending).
Fix cian_newbuilding._extract_transport_rate: cian transportAccessibilityRate
drifted to a nested dict and broke save_newbuilding_enrichment's CAST bind on
LIVE data (cannot adapt type 'dict') — coerce dict/float/bad -> int|None
(bool excluded before int). Fixes the live enrichment crash, not just backfill.
house_reviews stays ~0 for now: cian reviews are in a separate XHR, not the ЖК
initialState — extending fetch_newbuilding for them is a documented follow-up.
Tested: 8 new unit tests; isolated-DB proof-run landed 6 price_dynamics + 1
reliability row, idempotency proven across re-runs. Refs #972.
Run the merged frontend vitest suite on every frontend-touching PR so the
tests don't rot. Mirrors the #1032 backend gate: per-job dorny/paths-filter
(new `frontend` output, `frontend/**`), node:20 + npm cache on the lockfile,
`npm ci --legacy-peer-deps --no-audit --no-fund` (exact frontend/Dockerfile
flags), then `npm run test`. Additive — backend-tests job unchanged; no
top-level paths filter so frontend-only PRs already trigger the workflow.
Making frontend-tests a required check is a follow-up in the Forgejo UI.
Refs #999.
Surface report.meta provenance as a compact muted caption above the advisory
disclaimer: «Прогноз рассчитан: <дата> · горизонты N / M мес · схема X». The
freshness date sources from the run's persist time (envelope created_at, always
set) with meta.generated_at as fallback — the deterministic assembler leaves
generated_at null, but created_at is the real compute timestamp. Until now this
provenance lived only in exports; horizons appeared only via the chart axis.
New ForecastMetaLine.tsx: three independently-guarded segments (RU long date
with optional non-midnight time, horizons, most-muted schema); graceful null
when meta empty / 202-pending. Advisory disclaimer stays last.
Part of #958.
Surface the §16-traceability evidence behind the supply forecast that
previously lived only in exports: future_market.future_supply (§9.3
supply-pressure breakdown — open/hidden stock, absorption, months of
pressure, future units, pressure index) and future_market.future_competitors
(§9.7 relevance-weighted pipeline, sorted by relevance desc). New
ForecastFutureSupplyBlock as subsection 6.6 (+ section-6-6 nav),
complementing 6.1's per-horizon aggregates.
- Fix FutureSupply/FutureSupplyBreakdown/FutureCompetitor optionality in
forecast.ts to match the backend (FutureSupplyPressure/Competitor): null
metrics typed + rendered «нет данных», never a fabricated 0.
- Supply panel gates on computed metrics (index/pressure), NOT open/hidden
unit stocks — so the empty-supply_layers state (pre Monday-worker, every
parcel today) hides the panel instead of falsely showing "0 ед."; the
competitor table still renders. hasSupplySignal shared in forecast-helpers.
- RU labels reuse report_pdf/excel export wording. Graceful on partial/202.
Part of #958; surfaces #950 future-supply data.
Add «Скачать .docx» (Word) and «Скачать .pptx» (PowerPoint) buttons to
ForecastExportButtons, alongside the existing .md/.json/Telegram controls.
The backend /forecast/export already renders both formats; they were just
not reachable from the UI. Both reuse the existing response.blob() →
triggerDownload path (correct for binary zip-based OOXML — never .text(),
which would corrupt the archive). Per-format loading state; 404 →
«Отчёт ещё не готов». No new deps (icons from lucide-react).
Part of #958.
Surface the product-score + special-index breakdown behind the overall KPI
as two token-bar groups in Section 6 (6.5), sidebar anchor section-6-5.
Product scores use a 0..1 quality gradient anchored on the documented 0.5
balance point (>=0.55 strong / 0.45-0.55 mid / <0.45 weak; risk scores are
pre-inverted so higher=better); special indices use a neutral accent
(value = signal strength, direction non-uniform). Null values render as
«нет данных», never a 0-bar that implies a real zero. Graceful on
absent/partial/202-pending.
- types/forecast.ts: ReportScoring + nested (dict-keyed scores/indices)
- ForecastScoringBlock.tsx: two bar groups, a11y, null-distinct rendering
- forecast-helpers.ts: scoreVariant/scoreBarWidthPct + RU label maps
- Section6Forecast + sidebar nav: wire 6.5
Part of #958.
Adds «6.4 Рекомендация по продукту» to Section6Forecast — the §13.4
product recommendation (class, unit-mix, commercial, USP, §16 reasons)
that previously existed only in the export renderers. Unit-mix is shown
as token-colored horizontal bars; since the assembler emits no per-format
share (pct), bars are honestly labeled as a deficit-by-format signal, not
a build-share. Graceful on thin/202-pending/partial reports.
- types/forecast.ts: ReportProductTz + nested interfaces, optional on ForecastReport
- ForecastProductTzBlock.tsx: per-field guards, role=img bars, reasons disclosure
- forecast-helpers.ts: deficitBarWidthPct
- Section6Forecast + sidebar nav: wire 6.4
Part of #958.
Add Конкуренты / Будущие проекты / Зоны риска Leaflet layers to the Site Finder
analysis map (MiniMap→SiteMap) with per-layer toggles in CpLayerControlPanel
(Конкуренты on by default; pipeline + risk off). Competitors/pipeline render as
distinct CircleMarkers from analyze lat/lon (#1074); risk zones as GeoJSON
polygons parsed from geom_wkt via a small inline WKT parser (no new deps).
Graceful on missing coords / empty arrays / thin responses. Coordinate order
verified (GeoJSON lon-lat for <GeoJSON>, Leaflet lat-lon for CircleMarker).
3 of 4 layers; future-ЖК (newbuilding) pending newbuilding_listings API exposure
(#1108 built the table). Part of #999.
ECharts (via ChartShell, SSR-safe) chart of deficit_index across horizons per
scenario (база/агр/конс, --viz-1..3) + dashed rate-path on a secondary axis +
баланс/целевой markLines, tooltip with units. Renders above the 6.1 table;
returns null on thin reports. Makes the §22 forecast signal glanceable for demos.
