From 12c3bc3e15a1e777482ed913a292fcb235ab224d Mon Sep 17 00:00:00 2001
From: Henrik Jess Nielsen <lrihni@egmont.com>
Date: Sun, 24 May 2026 19:35:43 +0200
Subject: [PATCH] Add data quality normalization for cars: mileage in thousands
 and leasing prices
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- detect km < 500 on non-new cars (e.g. '130' instead of '130.000')
- detect prices < 5.000 kr (likely monthly leasing rate not sale price)
- flags shown in AI prompt → forces low score (1-3)
- orange warning box in UI distinct from AI warnings
---
 score.py             | 62 ++++++++++++++++++++++++++++++++++++++++++++
 templates/index.html |  5 ++++
 2 files changed, 67 insertions(+)

diff --git a/score.py b/score.py
index f772336..d81c429 100644
--- a/score.py
+++ b/score.py
@@ -143,6 +143,57 @@ def extract_structured_fields(raw: str) -> dict:
     return fields
 
 
+def normalize_listing(item: dict, category: str) -> dict:
+    """
+    Detect and flag data quality issues for cars:
+    - Mileage written as thousands (130 instead of 130.000)
+    - Suspiciously low price (likely a leasing monthly rate, not full price)
+
+    Adds 'data_quality_flags' list to item (in-place) and returns item.
+    """
+    if category != "brugte biler":
+        return item
+
+    flags = []
+    raw = item.get("details", {}).get("raw_text", "")
+    fields = extract_structured_fields(raw)
+
+    # ── Mileage check ─────────────────────────────────────────────────────────
+    km_str = fields.get("km", "")
+    if km_str:
+        km_digits = re.sub(r"[^\d]", "", km_str)
+        if km_digits:
+            km_val = int(km_digits)
+            # Likely written in thousands: 130 km instead of 130.000 km
+            # Cars below 500 km that aren't brand new (year < current-1) are suspicious
+            year_str = fields.get("year", "")
+            current_year = datetime.now().year
+            car_age = (current_year - int(year_str)) if year_str.isdigit() else 99
+            if km_val < 500 and car_age > 1:
+                corrected_km = km_val * 1000
+                flags.append(
+                    f"⚠️ Kilometertallet ser forkert ud: sælger har skrevet '{km_val} km' "
+                    f"— sandsynligvis ment {corrected_km:,} km. "
+                    f"Annoncen nedprioriteres pga. vildledende km-angivelse."
+                )
+
+    # ── Price check ───────────────────────────────────────────────────────────
+    try:
+        price = float(str(item.get("price_dkk", 0)).replace(",", "."))
+        if 0 < price < 5000:
+            flags.append(
+                f"⚠️ Prisen på {int(price):,} kr. er usandsynlig lav for en bil — "
+                f"sandsynligvis en månedlig leasingydelse, ikke salgspris. "
+                f"Annoncen nedprioriteres pga. misvisende prisangivelse."
+            )
+    except (ValueError, TypeError):
+        pass
+
+    if flags:
+        item["data_quality_flags"] = flags
+    return item
+
+
 def listing_summary(item: dict, idx: int) -> str:
     """Compact text representation of a listing for the AI prompt."""
     raw    = item.get("details", {}).get("raw_text", item.get("description", ""))
@@ -165,11 +216,18 @@ def listing_summary(item: dict, idx: int) -> str:
 
     meta_line = " | ".join(meta_parts)
 
+    # Include any data quality flags so AI factors them into scoring
+    quality_issues = item.get("data_quality_flags", [])
+    quality_block = ""
+    if quality_issues:
+        quality_block = "\nDATA KVALITETSPROBLEM (giv lav score 1-3):\n" + "\n".join(quality_issues) + "\n"
+
     return (
         f"--- Annonce #{idx + 1} (ID: {item['id']}) ---\n"
         f"Navn: {item['name']}\n"
         f"Pris: {item['price_dkk']} DKK\n"
         + (f"{meta_line}\n" if meta_line else "")
+        + quality_block
         + f"{text}\n"
     )
 
@@ -301,6 +359,10 @@ def score_listings(
     category = detect_category(items)
     phash = prefs_hash(prefs)
 
+    # ── Normalize data quality issues before any scoring ────────────────────
+    for item in items:
+        normalize_listing(item, category)
+
     # ── Split: persistent score cache → in-file cache → needs AI scoring ────
     to_score, cached = [], []
     now = datetime.now().isoformat(timespec="seconds")
diff --git a/templates/index.html b/templates/index.html
index 9fea81c..77e0986 100644
--- a/templates/index.html
+++ b/templates/index.html
@@ -425,6 +425,10 @@
         const pct   = Math.round(score * 10);
         const warn  = item.ai_warnings
           ? `<p class="text-xs mt-2" style="color:#dc2626">↑ ${item.ai_warnings}</p>` : "";
+        const qualityFlags = (item.data_quality_flags || []).length > 0
+          ? `<div class="mt-2 rounded-lg px-3 py-2" style="background:#fff7ed;border:1px solid #fed7aa">
+              ${item.data_quality_flags.map(f => `<p class="text-xs" style="color:#c2410c">${f}</p>`).join("")}
+             </div>` : "";
         const tag   = rankTag(score);
 
         const card = document.createElement("div");
@@ -447,6 +451,7 @@
               </div>
               <p class="text-xs leading-relaxed" style="color:#78716c">${item.ai_reason || ""}</p>
               ${warn}
+              ${qualityFlags}
             </div>
             <div class="shrink-0 text-right ml-2">
               <p class="font-semibold text-sm tabular-nums">${Number(item.price_dkk || 0).toLocaleString("da-DK")}<span class="text-xs font-normal ml-0.5" style="color:#a8a29e">kr</span></p>