fix(ingest): preserve brand casing when synthesizing device names

When a Wikipedia row's model string omits the brand, name synthesis prefixed
manufacturer.upper() / brand.title(), turning 'intel' into 'INTEL' (and
'oneplus' into 'Oneplus'). This reproduced the casing artifact that had to be
fixed by hand in the data. Use explicit per-brand display forms so Intel stays
'Intel' while genuinely all-caps brands (AMD, NVIDIA) keep their casing.

Author: Seungpyo1007

app/ingest/sources/wikipedia_cpu.py

@@ -43,6 +43,12 @@
     ("amd", "List_of_AMD_Threadripper_processors", "AMD Threadripper"),
 ]
 
+# Manufacturer keys are stored lowercase; these are their display forms used to
+# synthesize ``name`` when the model string omits the brand. Plain ``.upper()``
+# mangles "intel" → "INTEL" (an ingest casing artifact); AMD is genuinely
+# all-caps so it gets an explicit entry rather than title-casing.
+_BRAND_DISPLAY: dict[str, str] = {"intel": "Intel", "amd": "AMD"}
+
 # Lowercased header tokens → canonical field name. Order matters: the first
 # matching fragment per cell wins (so a "Cores/Threads" column maps to
 # ``cores`` rather than ``threads``).
@@ -139,7 +145,8 @@ def _build_candidate(
     process_node = row.get("process_node") or None
 
     segment = guess_cpu_segment(model)
-    name = model if model.lower().startswith(manufacturer) else f"{manufacturer.upper()} {model}"
+    brand = _BRAND_DISPLAY.get(manufacturer, manufacturer.title())
+    name = model if model.lower().startswith(manufacturer) else f"{brand} {model}"
 
     record: dict[str, object | None] = {
         "slug": slug,

app/ingest/sources/wikipedia_gpu.py

@@ -37,6 +37,12 @@
     ("intel", "List_of_Intel_graphics_processing_units", "Intel Graphics"),
 ]
 
+# Manufacturer keys are stored lowercase; these are their display forms used to
+# synthesize ``name`` when the model string omits the brand. Plain ``.upper()``
+# mangles "intel" → "INTEL" (an ingest casing artifact). NVIDIA and AMD are
+# genuinely all-caps so they get explicit entries rather than title-casing.
+_BRAND_DISPLAY: dict[str, str] = {"nvidia": "NVIDIA", "amd": "AMD", "intel": "Intel"}
+
 # Same matching strategy as CPU but with GPU-specific keyword sets.
 HEADER_RULES: dict[str, list[str]] = {
     "model": ["model", "card", "name"],
@@ -127,7 +133,8 @@ def _build_candidate(
     memory_type = (row.get("memory_type") or "").upper() or None
 
     segment = guess_gpu_segment(model)
-    name = model if model.lower().startswith(manufacturer) else f"{manufacturer.upper()} {model}"
+    brand = _BRAND_DISPLAY.get(manufacturer, manufacturer.title())
+    name = model if model.lower().startswith(manufacturer) else f"{brand} {model}"
 
     record: dict[str, object | None] = {
         "slug": slug,

app/ingest/sources/wikipedia_smartphone.py

@@ -37,6 +37,12 @@
     ("xiaomi", "List_of_Xiaomi_smartphones"),
 ]
 
+# Brand keys are stored lowercase; these are their display forms used to
+# synthesize ``name`` when the model string omits the brand. ``.title()`` is a
+# fine default (samsung→Samsung) but mangles intercapped brands, so those get
+# explicit entries (oneplus→OnePlus, not "Oneplus").
+_BRAND_DISPLAY: dict[str, str] = {"oneplus": "OnePlus"}
+
 HEADER_RULES: dict[str, list[str]] = {
     "model": ["model", "name"],
     "release_date": ["released", "release", "launched", "launch", "date"],
@@ -163,7 +169,7 @@ def _build_candidate(
     name = (
         model
         if model.lower().startswith(brand) or model.lower().startswith(brand.upper())
-        else f"{brand.title()} {model}"
+        else f"{_BRAND_DISPLAY.get(brand, brand.title())} {model}"
     )
 
     record: dict[str, object | None] = {