diff --git a/gitgalaxy/standards/language_lens.py b/gitgalaxy/standards/language_lens.py
index b658710a..b1537eee 100644
--- a/gitgalaxy/standards/language_lens.py
+++ b/gitgalaxy/standards/language_lens.py
@@ -202,6 +202,13 @@ def inspect(
         
         # We explicitly add .txt and .log to the prose override list
         if ext in {'.md', '.mdx', '.rst', '.rtf', '.txt', '.log'}:
+            # ---> THE FIX: Catch disguised payloads before the early exit! <---
+            shebang_lang = self._tier_2_fingerprint_check(content_sample, ext)
+            if shebang_lang and shebang_lang != "undeterminable":
+                self.logger.warning(f"[{name}] IDENTITY CRISIS: Prose Ext '{ext}' contradicts Shebang '{shebang_lang}'")
+                result["anomaly_flags"].append(f"Identity Masking: Prose Extension ({ext}) vs Shebang ({shebang_lang})")
+                return self._forge_result("undeterminable", 0.0, 5, f"Identity Conflict ({ext} != {shebang_lang})", result, content_sample)
+                
             target_id = "markdown" if ext in {'.md', '.mdx'} else "plaintext"
             return self._forge_result(target_id, self.thresholds.get("PROSE_CONFIDENCE", 0.95), 1, f"Prose Extension ({ext})", result, content_sample)
         
diff --git a/tests/test_chronometer_timeout.py b/tests/test_chronometer_timeout.py
new file mode 100644
index 00000000..348642c8
--- /dev/null
+++ b/tests/test_chronometer_timeout.py
@@ -0,0 +1,73 @@
+import unittest
+import time
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+# Adjust the import path if necessary based on your actual module structure
+from gitgalaxy.physics.chronometer import Chronometer
+
+class TestChronometerTimeout(unittest.TestCase):
+    
+    @patch('gitgalaxy.physics.chronometer.subprocess.Popen')
+    @patch.object(Chronometer, '_calibrate_temporal_field') # Skip the heavy init sequence
+    def test_zombie_process_kill_switch(self, mock_calibrate, mock_popen):
+        """
+        Simulates an infinite, hanging Git stream.
+        Verifies that the Chronometer respects the timeout and successfully
+        reaps the zombie process at the OS level via kill() and communicate().
+        """
+        # 1. Setup the Infinite Stream
+        def infinite_git_log():
+            while True:
+                # Yields a valid line so the internal logic has to do work
+                yield "mock_hash|1700000000|TestAuthor\n"
+                yield "src/safe_file.py\n"
+                
+        # We attach the generator to a MagicMock so we can assert .close() is called on it later
+        mock_stdout = MagicMock()
+        mock_stdout.__iter__.return_value = infinite_git_log()
+        mock_stderr = MagicMock()
+        
+        # 2. Setup the Mock Process
+        mock_process = MagicMock()
+        mock_process.stdout = mock_stdout
+        mock_process.stderr = mock_stderr
+        mock_popen.return_value = mock_process
+        
+        # 3. Initialize Chronometer (calibration is bypassed)
+        chrono = Chronometer(Path("/mock/repo"))
+        
+        # 4. Ignite the escalator with a tiny timeout (50ms)
+        timeout_limit = 0.05 
+        start_time = time.time()
+        
+        processed_lines, reached_target = chrono._run_git_stream_escalator(
+            cmd=["git", "log", "mock_args"],
+            ignored_hashes=set(),
+            tracked_files=set(),
+            required_files=10000, 
+            timeout_limit=timeout_limit,
+            start_time=start_time
+        )
+        
+        # =====================================================================
+        # 5. INVARIANT ASSERTIONS (The Proof)
+        # =====================================================================
+        
+        # Check that it actually processed lines, but was interrupted by the timeout
+        self.assertTrue(processed_lines > 0, "The stream should have processed lines before timing out.")
+        self.assertFalse(reached_target, "The escalator should have aborted before reaching the file target.")
+        
+        # --- THE ZOMBIE KILL SWITCH VERIFICATION ---
+        # Did we send the SIGKILL?
+        mock_process.kill.assert_called_once()
+        
+        # Did we flush the pipes? (Crucial: kill() without communicate() leaves a zombie)
+        mock_process.communicate.assert_called_once()
+        
+        # Did we close the file descriptors to prevent FD leaks?
+        mock_stdout.close.assert_called_once()
+        mock_stderr.close.assert_called_once()
+
+if __name__ == '__main__':
+    unittest.main()
\ No newline at end of file
diff --git a/tests/test_gpu_recorder.py b/tests/test_gpu_recorder.py
new file mode 100644
index 00000000..b7f18532
--- /dev/null
+++ b/tests/test_gpu_recorder.py
@@ -0,0 +1,51 @@
+import unittest
+from gitgalaxy.recorders.gpu_recorder import GPURecorder
+
+class TestGPURecorderEviction(unittest.TestCase):
+    
+    def test_destructive_ram_eviction(self):
+        """
+        Verifies Stage 3.3: Destructive RAM Eviction.
+        Ensures the GPU Recorder physically destroys the input arrays via .pop()
+        to free memory, preventing OOM crashes on massive repositories.
+        """
+        recorder = GPURecorder(version="6.3.0")
+        
+        # 1. Create the dummy arrays (passed by reference)
+        mock_parsed_files = [
+            {"path": f"src/file_{i}.py", "lang_id": "python", "total_loc": 100, "telemetry": {}}
+            for i in range(5)
+        ]
+        
+        mock_unparsable = [
+            {"path": f"bin/payload_{i}.dll", "reason": "Binary"}
+            for i in range(2)
+        ]
+        
+        # Verify they actually have data before we start
+        self.assertEqual(len(mock_parsed_files), 5)
+        self.assertEqual(len(mock_unparsable), 2)
+        
+        # 2. Execute the GPU Recorder
+        result = recorder.record_mission(
+            parsed_files=mock_parsed_files,
+            unparsable_files=mock_unparsable,
+            summary={"unparsable_files": {}},
+            forensic_report={},
+            repo_name="test_repo"
+        )
+        
+        # =====================================================================
+        # 3. INVARIANT ASSERTIONS (The Proof)
+        # =====================================================================
+        
+        # A) Did it actually build the payload successfully?
+        self.assertIn("galaxy", result, "GPU Recorder failed to build the galaxy payload.")
+        self.assertTrue(len(result["galaxy"]["paths"]) == 5, "GPU Recorder missed files in the output.")
+        
+        # B) THE EVICTION CONTRACT: Are the original RAM arrays completely destroyed?
+        self.assertEqual(len(mock_parsed_files), 0, "FATAL: GPU Recorder failed to evict parsed_files from RAM!")
+        self.assertEqual(len(mock_unparsable), 0, "FATAL: GPU Recorder failed to evict unparsable_files from RAM!")
+
+if __name__ == '__main__':
+    unittest.main()
\ No newline at end of file
diff --git a/tests/test_language_lens.py b/tests/test_language_lens.py
new file mode 100644
index 00000000..5894c5ff
--- /dev/null
+++ b/tests/test_language_lens.py
@@ -0,0 +1,48 @@
+import unittest
+from gitgalaxy.standards.language_lens import LanguageDetector
+
+class TestLanguageLensSecurity(unittest.TestCase):
+
+    def test_identity_crisis_trap_blocks_disguised_payloads(self):
+        """
+        Simulates a file attempting to mask its identity (e.g., .txt extension but a bash shebang).
+        Verifies that the Identity Crisis Trap catches the contradiction, strips the file's 
+        identity to 'undeterminable', and forces it to Tier 5 (Absolute Distrust).
+        """
+        # 1. Provide minimal mock definitions so the detector knows what 'bash' and 'txt' are
+        mock_lang_defs = {
+            "shell": {
+                "extensions": [".sh", ".bash"],
+                "shebangs": ["bash", "sh"]
+            },
+            "plaintext": {
+                "extensions": [".txt", ".log"]
+            }
+        }
+        
+        # 2. Initialize the detector
+        detector = LanguageDetector(language_definitions=mock_lang_defs, comment_definitions={})
+        
+        # 3. Create the malicious file: claims to be text, but acts like a shell script
+        mock_path = "uploads/innocent.txt"
+        mock_content = "#!/bin/bash\nrm -rf /\n"
+        
+        # 4. Run the inspection
+        result = detector.inspect(mock_path, content_sample=mock_content)
+        
+        # =====================================================================
+        # 5. INVARIANT ASSERTIONS (The Proof)
+        # =====================================================================
+        
+        # 1. Identity MUST be stripped to prevent downstream execution math
+        self.assertEqual(result["lang_id"], "undeterminable", "Detector failed to strip the identity of the masked payload!")
+        
+        # 2. Must be banished to Tier 5 (Absolute Distrust)
+        self.assertEqual(result["lock_tier"], 5, "Detector did not relegate the deceptive file to Tier 5!")
+        
+        # 3. Anomaly Flag must be registered in RAM so the SecurityLens can pick it up
+        self.assertTrue(len(result["anomaly_flags"]) > 0, "No anomaly flag was registered in the telemetry!")
+        self.assertIn("Identity Masking", result["anomaly_flags"][0], "The specific Identity Masking flag was missing.")
+
+if __name__ == '__main__':
+    unittest.main()
\ No newline at end of file
diff --git a/tests/test_neural_auditor.py b/tests/test_neural_auditor.py
new file mode 100644
index 00000000..dd02f200
--- /dev/null
+++ b/tests/test_neural_auditor.py
@@ -0,0 +1,60 @@
+import unittest
+import tempfile
+import struct
+import os
+from gitgalaxy.physics.neural_auditor import NeuralAuditor
+
+class TestNeuralAuditorHeaders(unittest.TestCase):
+    
+    def setUp(self):
+        # Initialize the auditor once for all tests in this class
+        self.auditor = NeuralAuditor()
+
+    def test_truncated_safetensors_file(self):
+        """
+        Simulates a file that is too small to even contain the 8-byte 
+        integer required by the safetensors format specification.
+        """
+        # Create a temporary file with a .safetensors extension
+        with tempfile.NamedTemporaryFile(suffix='.safetensors', delete=False) as temp_file:
+            # Write exactly 4 bytes of garbage data (struct.unpack requires 8)
+            temp_file.write(b'\x01\x02\x03\x04')
+            temp_path = temp_file.name
+
+        try:
+            # If the auditor is robust, this will NOT crash. 
+            # It should hit the broad except block and return the safe fallback.
+            result = self.auditor.audit_model(temp_path)
+            
+            self.assertEqual(result["architecture"], "Corrupted/Unknown")
+            self.assertEqual(result["parameters"], "Error")
+            self.assertEqual(result["quantization"], "Error")
+        finally:
+            # Always clean up the temporary file
+            os.remove(temp_path)
+
+    def test_corrupted_json_header(self):
+        """
+        Simulates a file that has a valid 8-byte size integer, 
+        but the subsequent bytes are corrupted/invalid JSON.
+        """
+        with tempfile.NamedTemporaryFile(suffix='.safetensors', delete=False) as temp_file:
+            # 1. Pack the number '10' into an 8-byte little-endian unsigned long long
+            header_size_bytes = struct.pack('<Q', 10)
+            temp_file.write(header_size_bytes)
+            
+            # 2. Write 10 bytes of invalid, un-parsable JSON
+            temp_file.write(b'{GARBAGE_}')
+            temp_path = temp_file.name
+
+        try:
+            # This should trigger a json.decoder.JSONDecodeError internally,
+            # which must be safely caught by the auditor.
+            result = self.auditor.audit_model(temp_path)
+            
+            self.assertEqual(result["architecture"], "Corrupted/Unknown")
+        finally:
+            os.remove(temp_path)
+
+if __name__ == '__main__':
+    unittest.main()
\ No newline at end of file
diff --git a/tests/test_redos_poison.py b/tests/test_redos_poison.py
new file mode 100644
index 00000000..7d660d4f
--- /dev/null
+++ b/tests/test_redos_poison.py
@@ -0,0 +1,74 @@
+import unittest
+import time
+from gitgalaxy.core.detector import LogicSplicer
+from gitgalaxy.core.prism import Prism
+
+class TestReDoSPoisoning(unittest.TestCase):
+    
+    def test_logic_splicer_balanced_end_pathological(self):
+        """
+        Simulates a pathologically nested file (50,000+ opening braces).
+        Verifies that the C-backed .find() loop survives and exits instantly.
+        """
+        # Initialize with a dummy language config
+        splicer = LogicSplicer("cpp", {})
+        
+        # Create a massive string of entirely unbalanced opening braces
+        poison_string = "{" * 60000 
+        
+        start_time = time.time()
+        # The search should hit the HANDSHAKE_LOOKAHEAD_LIMIT and exit cleanly
+        end_idx = splicer._find_balanced_end(poison_string, 0, "{", "}")
+        duration = time.time() - start_time
+        
+        # 1. Did it respect the lookahead limit? (Should stop at 50,000 max)
+        self.assertTrue(end_idx <= 50000, "Splicer failed to clamp the lookahead limit!")
+        
+        # 2. Did it run in O(N) time? (Should take less than 0.1 seconds)
+        self.assertTrue(duration < 0.5, f"LogicSplicer hung for {duration:.2f} seconds!")
+
+    def test_prism_nested_peel_limit(self):
+        """
+        Simulates a pathologically nested block comment (Rust/Swift style).
+        Verifies that NESTED_PEEL_LIMIT stops infinite loops.
+        """
+        comment_defs = {
+            "mechanical_families": {
+                "nested_c": {
+                    "delimiters": ["//", "/*", "*/"]
+                }
+            }
+        }
+        prism = Prism(comment_definitions=comment_defs, language_definitions={})
+        
+        # Create a string with 2,000 nested opening block comments and one closer
+        poison_comment = "/* " * 2000 + " */"
+        
+        start_time = time.time()
+        # _refract_nested uses a while-peel loop that must clamp to NESTED_PEEL_LIMIT
+        code, lits = prism._refract_nested(poison_comment)
+        duration = time.time() - start_time
+        
+        # If the peel limit (500) works, this exits almost instantly rather than looping 2000 times
+        self.assertTrue(duration < 0.5, f"Prism nested peel hung for {duration:.2f} seconds!")
+        
+    def test_prism_refract_unbalanced_quotes(self):
+        """
+        Simulates massive unbalanced quotes and braces sent to the main refract function.
+        Verifies the balanced scoping implementation doesn't get trapped by escaped quotes.
+        """
+        prism = Prism(comment_definitions={}, language_definitions={})
+        
+        # 20,000 alternating escaped quotes and braces
+        poison_string = "{\\\" " * 10000 
+        
+        start_time = time.time()
+        # Pass it through the main entry point to hit _find_balanced_end with quote-tracking
+        result = prism.refract(poison_string, "cpp")
+        duration = time.time() - start_time
+        
+        # The character-by-character scan should still process 20k chars in milliseconds
+        self.assertTrue(duration < 0.5, f"Prism refract hung for {duration:.2f} seconds!")
+
+if __name__ == '__main__':
+    unittest.main()
\ No newline at end of file
diff --git a/tests/test_zero_dependency.py b/tests/test_zero_dependency.py
index 299e771c..6ad1d4ad 100644
--- a/tests/test_zero_dependency.py
+++ b/tests/test_zero_dependency.py
@@ -44,6 +44,42 @@ def test_fallback_does_not_crash_signal_processor(self):
             
         except TypeError as e:
             self.fail(f"Zero-Dependency Mode crashed the Signal Processor! Error: {e}")
+            
+    @patch('gitgalaxy.security.security_auditor.ML_AVAILABLE', False)
+    def test_fallback_does_not_crash_security_auditor(self):
+        """
+        Simulates a user running GalaxyScope without 'xgboost' or 'pandas' installed.
+        Ensures that the ML Threat Inference gracefully bypasses without crashing,
+        while still executing the dependency graph resolution.
+        """
+        # We must import it inside the test or at the top of the file
+        from gitgalaxy.security.security_auditor import SecurityAuditor
+        
+        # 1. Initialize the blinded auditor
+        auditor = SecurityAuditor(model_path="dummy_path.json")
+        
+        # 2. Create a mock star
+        mock_stars = [{
+            "path": "src/safe_file.py",
+            "telemetry": {},
+            "raw_imports": ["src/other_file.py"]
+        }]
+        
+        try:
+            # 3. Force the auditor to process the stars
+            result_stars = auditor.audit_galaxy(mock_stars)
+            
+            # 4. INVARIANT ASSERTIONS
+            self.assertEqual(len(result_stars), 1, "Auditor should return the exact same number of stars.")
+            
+            # Verify the ML threat assessment was safely skipped
+            self.assertFalse(result_stars[0].get("is_ml_threat", False), "Blinded auditor should not flag ML threats.")
+            
+            # Verify the first half of the function (dependency graph mapping) STILL worked!
+            self.assertIn("dependency_network", result_stars[0], "Auditor failed to run the dependency graph resolution fallback.")
+            
+        except Exception as e:
+            self.fail(f"Zero-Dependency Mode crashed the Security Auditor! Error: {e}")
 
 if __name__ == '__main__':
     unittest.main()
\ No newline at end of file