fix(coverage): drop footnote and decimal artifacts from upstream slugs

The auto-generated coverage report listed non-SKU garbage such as '1200-4'
(from a '[4]' reference marker) and '1-25' (from a '1.25' clock cell). Strip
<sup> reference nodes and leftover '[...]' text in wikitable_first_cells, and
reject decimal-derived '<digits>-<digits>' slugs via a shared
is_probable_model_slug filter now used by both the CPU and GPU sources.

Author: Seungpyo1007

app/coverage/normalize.py

@@ -38,6 +38,22 @@
 _SEPARATOR_RE = re.compile(r"[^a-z0-9]+")
 _COLLAPSE_RE = re.compile(r"-+")
 
+# Decimal-derived artifacts like "1-25" (from a "1.25" clock cell) — two pure
+# numeric groups joined by a hyphen. Real CPU/GPU SKUs are a single numeric run
+# ("6276") or carry letters ("core-i9-14900k"), never bare "<digits>-<digits>".
+_DECIMAL_ARTIFACT_RE = re.compile(r"^\d+-\d+$")
+
+
+def is_probable_model_slug(slug: str, *, min_len: int = 4) -> bool:
+    """Heuristic: does ``slug`` look like a real CPU/GPU model (vs a stray cell)?
+
+    Rejects too-short slugs, slugs with no digit, and decimal-derived artifacts
+    such as ``"1-25"`` that come from non-model cells (clock speeds, footnotes).
+    """
+    if len(slug) < min_len or not any(c.isdigit() for c in slug):
+        return False
+    return not _DECIMAL_ARTIFACT_RE.match(slug)
+
 
 def slugify(name: str, manufacturer: str | None = None) -> str:
     """Normalize a vendor-style name to a kebab-case slug.

app/coverage/sources/wikipedia.py

@@ -6,6 +6,7 @@
 
 from __future__ import annotations
 
+import re
 from collections.abc import Iterator
 
 import httpx
@@ -14,6 +15,10 @@
 WIKI_REST_HTML = "https://en.wikipedia.org/api/rest_v1/page/html/{title}"
 USER_AGENT = "TechEngine-Coverage/0.1 (+https://github.com/GetTechAPI/TechEngine)"
 
+# Leftover footnote/reference markers (e.g. "[ 4 ]", "[c]") that survive as plain
+# text after <sup> nodes are stripped. They otherwise leak into slugs as "-4".
+_FOOTNOTE_RE = re.compile(r"\[[^\]]*\]")
+
 
 def fetch_wikipedia_html(page_title: str, *, timeout: float = 30.0) -> str:
     """Download the parsed HTML for a Wikipedia page."""
@@ -38,6 +43,10 @@ def wikitable_first_cells(html: str) -> Iterator[str]:
             cell = row.find(["th", "td"])
             if not cell:
                 continue
-            text = cell.get_text(" ", strip=True)
+            # Drop Wikipedia reference/footnote markers (rendered as <sup>), which
+            # otherwise leak into the model name as "[4]" and pollute the slug.
+            for sup in cell.find_all("sup"):
+                sup.decompose()
+            text = _FOOTNOTE_RE.sub("", cell.get_text(" ", strip=True)).strip()
             if text:
                 yield text

app/coverage/sources/wikipedia_cpu.py

@@ -4,7 +4,7 @@
 
 from collections.abc import Iterator
 
-from ..normalize import slugify
+from ..normalize import is_probable_model_slug, slugify
 from .base import CoveragePoint
 from .wikipedia import fetch_wikipedia_html, wikitable_first_cells
 
@@ -48,9 +48,8 @@ def fetch(self) -> Iterator[CoveragePoint]:
     def _extract(html: str, manufacturer: str, page: str) -> Iterator[CoveragePoint]:
         for raw_name in wikitable_first_cells(html):
             slug = slugify(raw_name, manufacturer=manufacturer)
-            # Filter obvious non-models. Real CPU slugs always contain a digit
-            # and are at least a few characters long.
-            if len(slug) < 4 or not any(c.isdigit() for c in slug):
+            # Filter obvious non-models (too short, no digit, decimal artifacts).
+            if not is_probable_model_slug(slug):
                 continue
             yield CoveragePoint(
                 category="cpu",

app/coverage/sources/wikipedia_gpu.py

@@ -4,7 +4,7 @@
 
 from collections.abc import Iterator
 
-from ..normalize import slugify
+from ..normalize import is_probable_model_slug, slugify
 from .base import CoveragePoint
 from .wikipedia import fetch_wikipedia_html, wikitable_first_cells
 
@@ -36,7 +36,7 @@ def fetch(self) -> Iterator[CoveragePoint]:
     def _extract(html: str, manufacturer: str, page: str) -> Iterator[CoveragePoint]:
         for raw_name in wikitable_first_cells(html):
             slug = slugify(raw_name, manufacturer=manufacturer)
-            if len(slug) < 4 or not any(c.isdigit() for c in slug):
+            if not is_probable_model_slug(slug):
                 continue
             yield CoveragePoint(
                 category="gpu",

tests/unit/test_coverage_wikipedia_parser.py

@@ -35,3 +35,37 @@ def test_wikipedia_cpu_extract_filters_obvious_non_models() -> None:
     assert "core-i7-14700k" in slugs
     # "Codename: Raptor Lake" has no digit and should be filtered.
     assert all("raptor" not in s for s in slugs)
+
+
+# Mirrors the AMD list pages that produced garbage slugs in the auto report:
+# footnote markers (<sup> + leftover "[ c ]") and a decimal clock cell.
+_HTML_NOISY = """
+<html><body>
+  <table class="wikitable">
+    <tr><th>Model</th></tr>
+    <tr><td>1200<sup class="reference">[4]</sup></td></tr>
+    <tr><td>1200 (AF) [ 16 ] [ c ]</td></tr>
+    <tr><td>1.25</td></tr>
+    <tr><td>1210</td></tr>
+  </table>
+</body></html>
+"""
+
+
+def test_wikitable_first_cells_strips_footnote_markers() -> None:
+    cells = list(wikitable_first_cells(_HTML_NOISY))
+    # <sup> reference and bracketed leftover markers are gone.
+    assert "1200" in cells
+    assert all("[" not in c and "]" not in c for c in cells)
+
+
+def test_wikipedia_cpu_extract_drops_decimal_and_footnote_artifacts() -> None:
+    points = list(WikipediaCpu._extract(_HTML_NOISY, "amd", "List_of_AMD_Ryzen_processors"))
+    slugs = {p.slug for p in points}
+    # Footnote suffixes no longer appear.
+    assert "1200-4" not in slugs
+    assert "1200-af-16-c" not in slugs
+    # Decimal clock cell "1.25" -> "1-25" is rejected as a non-model artifact.
+    assert "1-25" not in slugs
+    # A genuine bare-numeric SKU still survives.
+    assert "1210" in slugs