fix(chronometer): resolve git history for subdirectory scans

gitgalaxy/metrics/chronometer.py

@@ -107,14 +107,43 @@ def _initialize_history_scan(self):
         """Dispatches the survey engines to establish boundaries and churn cache."""
         t_start = time.time()
 
-        # Step A: Git Binary Verification
-        if (self.root / ".git").exists():
+        # Step A: Git Worktree Verification
+        # #2976: `.git` only exists at the repository ROOT, so the old check
+        # `(self.root / ".git").exists()` silently disabled the entire temporal
+        # pipeline for any scan rooted at a SUBDIRECTORY of a repo (galaxyscope
+        # path/to/repo/src, or a control-corpus language folder) -- the mtime
+        # fallback then fired, a fresh checkout's uniform mtimes tripped the
+        # TEMPORAL COLLAPSE guard below, and every file read the neutral
+        # (stability 50, churn 0). Ask git itself: rev-parse resolves upward
+        # from any subdirectory, and a linked worktree's `.git` FILE pointer
+        # passes the same way. A non-git directory still lands in the OS-walk
+        # fallback exactly as before.
+        #
+        # Determinism-sensitive harnesses (the golden-crucible pins scan a
+        # subdirectory of a pinned checkout whose git history is NOT part of
+        # the measured structure) opt out EXPLICITLY here -- before the fix
+        # they got temporal neutrality by accident of the broken root check.
+        if (
+            self.chrono_config.get("DISABLE_GIT_HISTORY", False)
+            or os.environ.get("GITGALAXY_DISABLE_GIT_HISTORY", "") == "1"
+        ):
+            # is_git_enabled stays False: Steps B/C below run the same OS-walk
+            # fallback (including the collapse guard) a non-git directory gets.
+            self.logger.info("Git history disabled by configuration. Using OS-walk fallback.")
+        else:
             try:
-                subprocess.run([_GIT_BIN, "--version"], capture_output=True, check=True)  # noqa: S603
-                self.is_git_enabled = True
-                self.logger.debug("Git binary verified. Commencing Deep Boundary Survey.")
-            except (subprocess.CalledProcessError, FileNotFoundError):
-                self.logger.warning("Git binary not found. Falling back to OS Walk.")
+                res = subprocess.run(  # noqa: S603 -- _GIT_BIN resolved absolute, fixed args
+                    [_GIT_BIN, "rev-parse", "--is-inside-work-tree"],
+                    cwd=self.root,
+                    capture_output=True,
+                    text=True,
+                    check=True,
+                )
+                if res.stdout.strip() == "true":
+                    self.is_git_enabled = True
+                    self.logger.debug("Git worktree verified. Commencing Deep Boundary Survey.")
+            except (subprocess.CalledProcessError, FileNotFoundError, OSError):
+                self.logger.warning("Not inside a git worktree (or git unavailable). Falling back to OS Walk.")
 
         # Step B: Establish Absolute Project Boundaries (Min/Max Time)
         self._determine_commit_bounds()
@@ -268,7 +297,12 @@ def _scan_git_history(self):
         # Axis 1 (Volume): Stop scanning once 50% of active files are mapped (max 5000).
         # Axis 2 (Time): Hard abort after 'timeout_limit' seconds.
         # ======================================================================
-        required_files = min(int(total_files * 0.50), 5000)
+        # #2976: a tiny subtree's year-log parses in milliseconds -- the 50% early
+        # exit only ever protected monorepo scans, and on a 6-file control-corpus
+        # folder it stopped after mapping 3 files, leaving the rest with no history
+        # at all. Full coverage at or below 200 files; the monorepo math is
+        # unchanged above that, and the Axis-2 time budget still applies to both.
+        required_files = total_files if total_files <= 200 else min(int(total_files * 0.50), 5000)
         timeout_limit = self.chrono_config.get("STREAM_TIMEOUT_SECONDS", 15.0)
 
         self.logger.info(f"Chronometer: Engaging 1-Year Historical Sweep. Budget: {timeout_limit}s")
@@ -277,13 +311,24 @@ def _scan_git_history(self):
 
         # 3. The Command: Limit Git to the last year of commits.
         # This generates massive churn spikes without getting bogged down in decade-old bedrock.
+        # #2976: `--relative -- .` scopes the stream to the scanned subtree and names
+        # files relative to cwd. Without it, a subdirectory scan streams the WHOLE
+        # repo's log (burning the time budget on unrelated files) and `--name-only`
+        # emits repo-relative paths while `git ls-files` above emits cwd-relative
+        # ones -- so churn_map/mtime_map would be keyed differently from both the
+        # tracked-files denominator and the scanner's scan-root-relative rel_path,
+        # and every lookup would miss. At the repository root both flags are no-ops:
+        # the output is byte-identical to the previous command.
         cmd = [
             _GIT_BIN,
             "log",
             "--since=1.year",
             "--name-only",
+            "--relative",
             "--pretty=format:@@GIT_COMMIT@@|%H|%at|%an",
             "--no-merges",
+            "--",
+            ".",
         ]
 
         # Execute the stream

gitgalaxy/standards/gitgalaxy_config.py

@@ -699,6 +699,12 @@
 # Consumed by: chronometer.py
 # ------------------------------------------------------------------------------
 CHRONOMETER_CONFIG = {
+    # #2976: force the OS-walk fallback even inside a git worktree. For
+    # determinism-sensitive harnesses (the golden-crucible pins) whose
+    # corpus git history is not part of the measured structure. The
+    # GITGALAXY_DISABLE_GIT_HISTORY=1 environment variable is the same
+    # switch for subprocess invocations.
+    "DISABLE_GIT_HISTORY": False,
     # The absolute ceiling for OS-level fallback scanning
     "FALLBACK_SCAN_LIMIT": 25000,
     # Process management

tests/core_engine/test_chronometer.py

@@ -37,14 +37,13 @@ def test_chronometer_no_git_fallback(mock_getmtime, mock_walk, mock_run, tmp_pat
 def test_chronometer_git_boundaries(mock_run, tmp_path):
     """Proves the boundary scanner correctly extracts min/max times from git logs."""
 
-    # Simulate the .git directory existing so the hardware check fires
-    (tmp_path / ".git").mkdir()
-
     def git_side_effect(cmd, **kwargs):
         m = MagicMock()
         m.stdout = ""
-        if "--version" in cmd:
-            m.stdout = "git version 2.40.0\n"
+        if "rev-parse" in cmd:
+            # #2976: the gate asks git itself rather than testing (root/.git).exists(),
+            # so a subdirectory of a repo (and a linked worktree) passes too.
+            m.stdout = "true\n"
         elif "log" in cmd and "-1" in cmd and "HEAD" not in cmd and len(cmd) == 4:
             m.stdout = "5000\n"  # Max Time
         elif "rev-list" in cmd:

tests/dead_key_audit.py

@@ -95,6 +95,7 @@
     # --- External YAML/env config ---
     "galaxyscope": "top-level section name in a user's .galaxyscope.yml project config file",
     "GITGALAXY_LICENSE_KEY": "environment variable (os.environ.get), not a repo-produced dict",
+    "GITGALAXY_DISABLE_GIT_HISTORY": "environment variable (os.environ.get), not a repo-produced dict (#2976)",
     "vulnerability_density_min": "optional risk_tuning YAML key (signal_processor.py risk-equation-style tuning)",
     "asymptotic_dampener": "optional risk_tuning YAML key (signal_processor.py)",
     "quarantine": "STATIC_ARCHETYPES app-config constant, read with a graceful string fallback",

tests/test_golden_crucible.py

@@ -66,7 +66,15 @@ def test_golden_crucible_matches_baseline(tmp_path):
         ],
         check=True,
         timeout=180,
-        env={**os.environ, "GITGALAXY_LICENSE_KEY": "COMMUNITY_FREE_TIER"},
+        env={
+            **os.environ,
+            "GITGALAXY_LICENSE_KEY": "COMMUNITY_FREE_TIER",
+            # #2976: the golden masters pin a corpus whose git history is not part
+            # of the measured structure. Before the chronometer's subdir fix they
+            # got temporal neutrality by accident (the scan root data/ has no .git);
+            # this makes the same neutrality explicit and deterministic.
+            "GITGALAXY_DISABLE_GIT_HISTORY": "1",
+        },
     )
 
     actual_path = output_dir / "data_galaxy_audit.json"