From faf6beb536c977c0d674afb610d8bd9a5b6da9d2 Mon Sep 17 00:00:00 2001
From: CC6 Worker <cc6@rinet.one>
Date: Tue, 5 May 2026 01:30:16 +0200
Subject: [PATCH] =?UTF-8?q?M12.6=20SF:=20sport-aware=20enrichment=20+=20fe?=
 =?UTF-8?q?deration=20map=20(HBS,=20HKS,=20HRS,=20HOS,=20HVS,=20HPS,=20HBS?=
 =?UTF-8?q?=20bocanje=E2=80=A6)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- data/sport_federations.json: 24 Croatian sport federations + aliases +
  PGŽ local media (Novi list, Glas Istre, Rijeka.danas).
- enrich_router._sport_fed/_normalize_sport/_load_sport_feds: cached
  loader that picks up file changes via mtime.
- _research_links() now sport-aware: when row.sport maps to a known fed,
  the dynamic links list shows that fed (national + PGŽ regional) plus the
  three PGŽ local-media search URLs in place of the static HNS Semafor +
  transfermarkt fallback.
- scrape_sport_federation(sport, ime, prezime): generic profile-page
  scraper (slug pattern OR search-results crawl) → returns
  {profile_url, slika_url, datum_rodenja, mjesto_rodenja, klub_naziv}.
- _propose_for_sportas() now routes through the federation scraper before
  HNS Semafor; HNS path is gated to nogomet or rows already linked.
- _load_row(sportas) JOINs klubovi to fall back to klub.sport when
  c.sport is empty.
- Tested on 1024 Marijan Alkić (boćanje): proposed profile_url +
  datum_rodenja from hrvatski-bocarski-savez.hr; /apply persisted them.
- Tested on 3335 Toni Jelenković (košarka) and 3379 Niko Miknić
  (plivanje): research_links surface HKS/KS PGŽ and HPS respectively.

Worker:
- _pick_sportas now selects on coverage<70 across ALL sports (sport
  set OR known external linkage), not just hns_*.
- _SOURCE_WEIGHTS extended with 16 federation hosts at 0.88-0.92.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 workers/enrichment_worker.py | 63 +++++++++++++++++++++++++++---------
 1 file changed, 47 insertions(+), 16 deletions(-)

diff --git a/workers/enrichment_worker.py b/workers/enrichment_worker.py
index 859bbc9..ce005bf 100644
--- a/workers/enrichment_worker.py
+++ b/workers/enrichment_worker.py
@@ -95,30 +95,46 @@ _SPORTAS_KEYS = ('sport','profile_url','slika_url','hns_igrac_id','biografija',
                  'datum_rodenja','mjesto_rodenja','broj_dresa')
 
 
-def _coverage_expr(table_keys: tuple[str, ...]) -> str:
-    """Postgres expression that returns 0..100 coverage % for the row."""
+def _coverage_expr(table_keys: tuple[str, ...], prefix: str = '') -> str:
+    """Postgres expression that returns 0..100 coverage % for the row.
+
+    `prefix` is e.g. 'c.' when the SQL uses a table alias.
+    """
     parts = []
     for k in table_keys:
-        parts.append(f"(CASE WHEN {k} IS NOT NULL AND ({k}::text) <> '' THEN 1 ELSE 0 END)")
+        col = f"{prefix}{k}"
+        parts.append(f"(CASE WHEN {col} IS NOT NULL AND ({col}::text) <> '' THEN 1 ELSE 0 END)")
     total = len(table_keys)
     return f"((({' + '.join(parts)})::numeric * 100) / {total})"
 
 
 def _pick_sportas(limit: int = 50) -> list[int]:
-    """Athletes with coverage<COVERAGE_MAX, randomly ordered."""
-    cov = _coverage_expr(_SPORTAS_KEYS)
+    """Athletes with coverage<COVERAGE_MAX, randomly ordered.
+
+    Selection is sport-agnostic now: the router decides which federation to
+    query based on c.sport (or klubovi.sport via the JOIN). We require either
+    sport to be set on the row OR a known external linkage so we don't burn
+    cycles on rows the router can't enrich.
+    """
+    cov = _coverage_expr(_SPORTAS_KEYS, prefix='c.')
     sql = f"""
-      SELECT id FROM pgz_sport.clanovi
-      WHERE aktivan = TRUE
+      SELECT c.id
+      FROM pgz_sport.clanovi c
+      LEFT JOIN pgz_sport.klubovi k ON k.id = c.klub_id
+      WHERE c.aktivan = TRUE
         AND {cov} < %s
         AND (
-              source IN ('hns_semafor','hns_family','manual','godisnjak')
-              OR jsonb_exists(vanjski_id, 'hns_comet')
-              OR (source_url ILIKE '%%semafor.hns.family%%')
-              OR profile_url ILIKE '%%semafor.hns.family%%'
+              c.sport IS NOT NULL
+              OR k.sport IS NOT NULL
+              OR c.source IN ('hns_semafor','hns_family','manual','godisnjak','hbs_savez','hks_savez')
+              OR jsonb_exists(c.vanjski_id, 'hns_comet')
+              OR (c.source_url ILIKE '%%semafor.hns.family%%')
+              OR (c.profile_url ILIKE '%%semafor.hns.family%%')
+              OR (c.source_url ILIKE '%%hrvatski-bocarski-savez.hr%%')
+              OR (c.profile_url ILIKE '%%hrvatski-bocarski-savez.hr%%')
         )
-        AND ((metadata->>'enriched_at') IS NULL
-             OR (metadata->>'enriched_at')::timestamptz < now() - interval '7 days')
+        AND ((c.metadata->>'enriched_at') IS NULL
+             OR (c.metadata->>'enriched_at')::timestamptz < now() - interval '7 days')
       ORDER BY random()
       LIMIT %s
     """
@@ -179,9 +195,24 @@ def _http_post(path: str, body: dict | None = None) -> dict | None:
 # zajednica generic info, so we down-weight them so a plain DeepSeek synthesis
 # off a single sport-pgz.hr source falls below the gate.
 _SOURCE_WEIGHTS = {
-    'semafor.hns.family': 0.95,
-    'wikipedia.hr':       0.80,
-    'sport-pgz.hr':       0.55,
+    'semafor.hns.family':           0.95,
+    'hrvatski-bocarski-savez.hr':   0.92,
+    'hns-cff.hr':                   0.90,
+    'hks-cbf.hr':                   0.90,
+    'hrs.hr':                       0.90,
+    'hos-cvf.hr':                   0.90,
+    'hvs.hr':                       0.90,
+    'hps.hr':                       0.90,
+    'atletika.hr':                  0.90,
+    'htsavez.hr':                   0.90,
+    'judo-savez.hr':                0.88,
+    'karate.hr':                    0.88,
+    'veslacki-savez.hr':            0.88,
+    'gimnastika.hr':                0.88,
+    'stolni-tenis.hr':              0.88,
+    'kuglanje.hr':                  0.88,
+    'wikipedia.hr':                 0.80,
+    'sport-pgz.hr':                 0.55,
 }
 # Fields that are safe to auto-write even from low-confidence sources because
 # they come from the entity's own structured page (URLs, IDs).