From b613ac6827fc2f369b1f0b146b4d138fcd353521 Mon Sep 17 00:00:00 2001
From: goon-foss <goon-foss@tutamail.com>
Date: Fri, 16 Oct 2026 11:32:37 +0200
Subject: [PATCH] fix(scenes): sceny o identycznym tytule zawsze w puli
 kandydatow

find_blocking_candidates OR-owal dokladny tytul z blokiem studio/data pod
jednym LIMIT 50 bez kolejnosci. Samo okno dat to ok. 45 tys. scen, wiec
scena o identycznym tytule trafiala do puli w 17% przypadkow (probka 120).
Teraz tytul idzie osobnym zapytaniem po indeksie, a blok studio+data od
najblizszej daty. Sredni czas spadl z 12 do 4 ms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 app/resolve/scene_match.py | 39 +++++++++++++++++++++++++++++---------
 1 file changed, 30 insertions(+), 9 deletions(-)

diff --git a/app/resolve/scene_match.py b/app/resolve/scene_match.py
index ff338c6..438f9f3 100644
--- a/app/resolve/scene_match.py
+++ b/app/resolve/scene_match.py
@@ -4,7 +4,7 @@ from __future__ import annotations
 import uuid
 from datetime import date, timedelta
 
-from sqlalchemy import and_, or_, select, text
+from sqlalchemy import and_, func, select, text
 from sqlalchemy.orm import Session
 
 from app.config import get_settings
@@ -165,12 +165,29 @@ def find_blocking_candidates(
       - jeśli mamy studio + date → studio_id == X AND date BETWEEN ±window_days
       - jeśli mamy tylko date → date BETWEEN ±window_days
       - jeśli mamy tylko studio → studio_id == X
-      - dodatkowo, jeśli `title_normalized` podany, OR-uj exact title match
-        (przydaje się gdy date/studio brakuje)
+      - dodatkowo, jeśli `title_normalized` podany, sceny o identycznym tytule
+
+    Sceny o identycznym tytule ida osobnym zapytaniem i zawsze trafiaja do puli. Do
+    2026-10-16 byly OR-owane z blokiem pod jednym LIMIT bez kolejnosci, a samo okno dat
+    to ok. 45 tys. scen, wiec dokladny duplikat potrafil wypasc (ten sam blad co przy
+    filmach, find_movie_candidates). Blok studio+data idzie od najblizszej daty (12 ms
+    dla najwiekszego studia); kolejnosc po podobienstwie tytulu kosztowala 0,4-3 s na
+    scene, za drogo na ingest.
     """
     if window_days is None:
         window_days = get_settings().date_window_days
 
+    out: list[Scene] = []
+    if title_normalized:
+        out = list(
+            session.execute(
+                select(Scene)
+                .where(Scene.title_normalized == title_normalized)
+                .order_by(Scene.created_at.desc())
+                .limit(limit)
+            ).scalars().all()
+        )
+
     conds = []
     if studio_id is not None and release_date is not None:
         conds.append(
@@ -191,11 +208,15 @@ def find_blocking_candidates(
     elif studio_id is not None:
         conds.append(Scene.studio_id == studio_id)
 
-    if title_normalized:
-        conds.append(Scene.title_normalized == title_normalized)
-
     if not conds:
-        return []
+        return out
 
-    stmt = select(Scene).where(or_(*conds)).limit(limit)
-    return list(session.execute(stmt).scalars().all())
+    stmt = select(Scene).where(*conds)
+    if studio_id is not None and release_date is not None:
+        stmt = stmt.order_by(func.abs(Scene.release_date - release_date))
+    seen = {s.id for s in out}
+    for s in session.execute(stmt.limit(limit)).scalars().all():
+        if s.id not in seen:
+            seen.add(s.id)
+            out.append(s)
+    return out