fix: create scratch dir and scope catalog ratio

scripts/generate_pack.py

@@ -1672,6 +1672,9 @@ def _normalize_zip_for_pack(
     """
     import tempfile as _tmp
 
+    # tmp/ is gitignored, so a fresh clone does not carry it. WSL keeps /tmp
+    # on a 4 GB tmpfs that MAME sets overflow, hence the repo-local scratch.
+    os.makedirs("tmp", exist_ok=True)
     tmp_fd, tmp_path = _tmp.mkstemp(suffix=".zip", dir="tmp")
     os.close(tmp_fd)
     try:

scripts/generate_readme.py

@@ -18,6 +18,7 @@ from datetime import datetime, timezone
 
 sys.path.insert(0, os.path.dirname(__file__))
 from common import (
+    GAME_DATA_TOPS,
     compute_composition,
     list_registered_platforms,
     load_database,
@@ -154,18 +155,32 @@ _CATALOG_LABELS = {"redump": "Redump", "no-intro": "No-Intro", "tosec": "TOSEC"}
 
 
 def _catalog_matched_line(db: dict) -> list[str]:
-    """Bullet line for files matched to dump-preservation catalogs."""
+    """Bullet line for files matched to dump-preservation catalogs.
+
+    Counted against the system files alone: those catalogs index console and
+    computer dumps, so arcade ROM sets and engine data sit outside their
+    scope and would only dilute the ratio.
+    """
     sources: set[str] = set()
     matched = 0
     for entry in db.get("files", {}).values():
         provenance = entry.get("provenance")
-        if provenance:
+        if not provenance:
+            continue
+        sources.update(provenance)
+        parts = entry.get("path", "").split("/")
+        top = parts[1] if len(parts) > 1 else ""
+        if top != "Arcade" and top not in GAME_DATA_TOPS:
             matched += 1
-            sources.update(provenance)
     if not matched:
         return []
+    system_files = compute_composition(db)["systems"]["files"]
     labels = ", ".join(_CATALOG_LABELS.get(s, s) for s in sorted(sources))
-    return [f"- **{matched:,} files** matched to dump-preservation catalogs ({labels})"]
+    return [
+        f"- **{matched:,} of {system_files:,} system files** matched to"
+        f" dump-preservation catalogs ({labels}); arcade sets and engine data"
+        f" fall outside what those catalogs index"
+    ]
 
 
 def generate_readme(db: dict, platforms_dir: str) -> str:

scripts/generate_site.py

@@ -54,6 +54,35 @@ ICON_CACHE_PATH = Path(".cache") / "system_icons.json"
 _icon_available: dict[str, bool] = {}
 
 
+def _content_check_ceiling(profiles: dict) -> str:
+    """How far content checking can reach across every profiled entry."""
+    hashed = sized = neither = 0
+    for profile in unique_emulator_profiles(profiles).values():
+        for f in profile.get("files", []) or []:
+            if any(
+                f.get(k)
+                for k in ("md5", "sha1", "crc32", "sha256", "known_hash_adler32")
+            ):
+                hashed += 1
+            elif any(f.get(k) for k in ("size", "min_size", "max_size")):
+                sized += 1
+            else:
+                neither += 1
+    total = hashed + sized + neither
+    if not total:
+        return ""
+    return (
+        f"Across every profiled entry, {hashed:,} of {total:,} "
+        f"({hashed / total * 100:.0f}%) carry a hash the code checks, "
+        f"{sized:,} ({sized / total * 100:.0f}%) only a size, and "
+        f"{neither:,} ({neither / total * 100:.0f}%) neither. That last share "
+        "is the ceiling of the method: where an emulator validates nothing, "
+        "reading its source establishes which file it loads, never whether "
+        "the content is the right dump. The provenance field answers that "
+        "other question, for the files a dump catalog indexes."
+    )
+
+
 def _icon_name(manufacturer: str, console_name: str) -> str:
     return f"{manufacturer} - {console_name}".replace("/", " ")
 
@@ -1884,6 +1913,8 @@ def generate_gap_analysis(
         "profiled emulator applies to the platform, whose own source is then "
         "the only authority.",
         "",
+        _content_check_ceiling(profiles),
+        "",
     ])
 
     # ---- Section 1b: platform lists vs emulator source ----