Adds deterministic horizon-aware relevance_weight ∈[0,1] to get_competitors
(ТЗ §9.1): weighted mean of geo_proximity (0.30), class_similarity (0.25),
price_similarity (0.20), stage_at_horizon (0.25). stage_at_horizon projects
each competitor's sell-out to horizon_months (request param, default 12) so a
competitor sold-out by our launch ranks lower. class_similarity uses a domrf
class-ladder vs the local dominant class (no target class yet). Orders by
relevance_weight DESC (distance still present + tie-break). relevance_breakdown
(4 sub-scores) exposed for explainability (§16).
Additive: all legacy Competitor fields intact, new fields Optional; zero
frontend consumers of /competitors (UI uses /analyze). sold-count folded into
existing price query (no extra db.execute, #227 regression green). Graceful:
thin data → neutral 0.5. 57 new pure-helper tests. Completes #949.
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).
