analysis: resolve SHA references and measure timestamp anonymisation

Two checks the human asked for after Phase 1: whether the mapping list
from the 2026-08-07 history rewrite still resolves older references, and
how far the anonymisation rule actually reaches.

sha_refs.tsv verifies all 251 mapping rows against the repos and
resolves every SHA cited in a management doc.
timestamp_anonymisation.tsv separates the project's own commits from
upstream fork history before counting non-compliant timestamps.

analysis/scripts/common.py

@@ -82,11 +82,19 @@ def assert_untouched():
             "field-test mandate forbids:\n  " + "\n  ".join(offending))
 
 
-def git(repo, *args):
-    """Run git in repo, return stdout as text. Raises on non-zero exit."""
+def git(repo, *args, env_tz=None):
+    """Run git in repo, return stdout as text. Raises on non-zero exit.
+
+    env_tz pins the timezone git renders dates in. Date rules in this
+    project are stated in UTC; rendering them in the machine's local zone
+    silently misclassifies every commit.
+    """
+    env = None
+    if env_tz:
+        env = dict(os.environ, TZ=env_tz)
     res = subprocess.run(
         ["git", "-C", str(repo), *args],
-        capture_output=True, text=True, errors="replace",
+        capture_output=True, text=True, errors="replace", env=env,
     )
     if res.returncode != 0:
         raise RuntimeError(f"git {' '.join(args)} in {repo}: {res.stderr.strip()}")