Add prominent warnings for protection score proxy metric

README.md

@@ -47,6 +47,9 @@ The repository is organized around a practical study loop:
 
 ## Current Study Snapshot
 
+> **⚠️ Important Note on Protection Score:**
+> The `Protection Score` is a **proxy metric** that measures drift in ResNet18 feature space. It is **NOT validated** against real-world mimicry attacks like DreamBooth or LoRA. This score is useful for relative comparisons within this repository only, not as a universal guarantee of protection effectiveness. Real-world effectiveness has not been empirically validated.
+
 Current local report highlights:
 
 | Run | Protection Score | PSNR | SSIM | Notes |
@@ -57,8 +60,6 @@ Current local report highlights:
 | `blindfold` | `61.1` | `26.53` | `0.6114` | strongest current anti-readability preset, largest fidelity cost |
 | `collective n000050 / set_B` | `22.8` avg | `37.78` avg | `0.9666` avg | correct benchmark direction, objective still needs tuning |
 
-The `Protection Score` is an internal proxy derived from embedding and style similarity after robustness transforms. It is useful for relative comparisons inside this repository, not as a universal guarantee against all AI systems.
-
 ## Workflow
 
 1. **Prepare input**: choose a single artwork, a folder, or a subject split such as `set_B`.

src/auralock/cli.py

@@ -97,6 +97,11 @@ def _render_readability_table(report: dict[str, object]) -> Table:
         f"{report['protection_score']:.1f}/100",
     )
     table.add_row("Assessment", str(report["assessment"]))
+
+    # Add warning row if present
+    if "warning" in report:
+        table.add_row("⚠️  Warning", str(report["warning"]))
+
     return table
 
 
@@ -433,6 +438,13 @@ def protect(
     console.print(_render_protection_table(result))
     console.print(_render_quality_table(result.quality_report))
     console.print(_render_readability_table(result.protection_report))
+
+    # Print prominent warning about protection score
+    if "warning" in result.protection_report:
+        console.print(f"\n[yellow]{result.protection_report['warning']}[/yellow]")
+        if "disclaimer" in result.protection_report:
+            console.print(f"[dim]{result.protection_report['disclaimer']}[/dim]\n")
+
     if report is not None:
         payload = result.to_report_dict(output_path=output_path)
         if adaptive_requirements is not None:
@@ -485,6 +497,17 @@ def analyze(
 
     console.print(_render_quality_table(analysis_report["quality_report"]))
     console.print(_render_readability_table(analysis_report["protection_report"]))
+
+    # Print prominent warning about protection score
+    if "warning" in analysis_report["protection_report"]:
+        console.print(
+            f"\n[yellow]{analysis_report['protection_report']['warning']}[/yellow]"
+        )
+        if "disclaimer" in analysis_report["protection_report"]:
+            console.print(
+                f"[dim]{analysis_report['protection_report']['disclaimer']}[/dim]\n"
+            )
+
     if report_path is not None:
         _write_json_report(
             report_path,
@@ -578,6 +601,16 @@ def batch(
         progress.update(task, completed=True, description="Batch processing completed")
 
     console.print(_render_batch_table(summary))
+
+    # Print prominent warning about protection scores in batch
+    from auralock.core.metrics import (
+        PROTECTION_SCORE_DISCLAIMER,
+        PROTECTION_SCORE_WARNING,
+    )
+
+    console.print(f"\n[yellow]{PROTECTION_SCORE_WARNING}[/yellow]")
+    console.print(f"[dim]{PROTECTION_SCORE_DISCLAIMER}[/dim]\n")
+
     if report is not None:
         _write_json_report(report, summary.to_report_dict())
         console.print(f"[green]Report saved to:[/green] {report}")
@@ -646,6 +679,16 @@ def benchmark(
         progress.update(task, completed=True, description="Benchmark completed")
 
     console.print(_render_profile_summary_table(summary.profile_summaries))
+
+    # Print prominent warning about protection scores in benchmark summary
+    from auralock.core.metrics import (
+        PROTECTION_SCORE_DISCLAIMER,
+        PROTECTION_SCORE_WARNING,
+    )
+
+    console.print(f"\n[yellow]{PROTECTION_SCORE_WARNING}[/yellow]")
+    console.print(f"[dim]{PROTECTION_SCORE_DISCLAIMER}[/dim]\n")
+
     if report is not None:
         _write_json_report(report, summary.to_report_dict())
         console.print(f"[green]Report saved to:[/green] {report}")

src/auralock/core/metrics.py

@@ -24,6 +24,18 @@
     load_default_style_feature_extractor,
 )
 
+# Protection Score Warning Constants
+PROTECTION_SCORE_WARNING = (
+    "⚠️  Protection Score is a proxy metric. "
+    "Real-world effectiveness is NOT validated."
+)
+PROTECTION_SCORE_METRIC_TYPE = "proxy_unvalidated"
+PROTECTION_SCORE_DISCLAIMER = (
+    "This score measures drift in ResNet18 feature space, "
+    "NOT actual mimicry prevention against real-world attacks like DreamBooth or LoRA. "
+    "Use for relative comparisons within this repository only."
+)
+
 
 def calculate_psnr(
     original: torch.Tensor | np.ndarray,
@@ -311,11 +323,11 @@ def get_protection_readability_report(
     )
 
     if protection_score >= 45:
-        assessment = "Strong"
+        assessment = "Strong (proxy space)"
     elif protection_score >= 25:
-        assessment = "Moderate"
+        assessment = "Moderate (proxy space)"
     else:
-        assessment = "Weak"
+        assessment = "Weak (proxy space)"
 
     return {
         "style_similarity": style_similarity,
@@ -324,6 +336,9 @@ def get_protection_readability_report(
         "robust_embedding_similarity": robust_embedding_similarity,
         "protection_score": protection_score,
         "assessment": assessment,
+        "metric_type": PROTECTION_SCORE_METRIC_TYPE,
+        "warning": PROTECTION_SCORE_WARNING,
+        "disclaimer": PROTECTION_SCORE_DISCLAIMER,
         "transform_style_similarities": transform_style_similarities,
         "transform_embedding_similarities": transform_embedding_similarities,
     }

src/tests/test_metrics.py

@@ -131,3 +131,70 @@ def test_quality_poor(self):
         report = get_quality_report(img1, img2)
 
         assert report["overall_quality"] == "Poor"
+
+
+class TestProtectionReadabilityReport:
+    """Tests for protection readability report with warning metadata."""
+
+    def test_protection_report_includes_warnings(self):
+        """Test that protection report includes warning metadata."""
+        from auralock.core.metrics import (
+            PROTECTION_SCORE_DISCLAIMER,
+            PROTECTION_SCORE_METRIC_TYPE,
+            PROTECTION_SCORE_WARNING,
+            get_protection_readability_report,
+        )
+
+        original = torch.rand(1, 3, 100, 100)
+        modified = original + torch.randn_like(original) * 0.1
+        modified = torch.clamp(modified, 0, 1)
+
+        report = get_protection_readability_report(original, modified)
+
+        # Check that warning metadata is present
+        assert "warning" in report
+        assert "metric_type" in report
+        assert "disclaimer" in report
+
+        # Check that values match constants
+        assert report["warning"] == PROTECTION_SCORE_WARNING
+        assert report["metric_type"] == PROTECTION_SCORE_METRIC_TYPE
+        assert report["disclaimer"] == PROTECTION_SCORE_DISCLAIMER
+
+    def test_protection_report_assessment_includes_proxy_space(self):
+        """Test that assessment labels include '(proxy space)' suffix."""
+        from auralock.core.metrics import get_protection_readability_report
+
+        original = torch.rand(1, 3, 100, 100)
+        modified = original + torch.randn_like(original) * 0.1
+        modified = torch.clamp(modified, 0, 1)
+
+        report = get_protection_readability_report(original, modified)
+
+        # Check that assessment includes proxy space indicator
+        assert "(proxy space)" in report["assessment"]
+
+    def test_protection_report_has_all_expected_keys(self):
+        """Test that protection report has all expected keys including new warning fields."""
+        from auralock.core.metrics import get_protection_readability_report
+
+        original = torch.rand(1, 3, 100, 100)
+        modified = original + torch.randn_like(original) * 0.1
+        modified = torch.clamp(modified, 0, 1)
+
+        report = get_protection_readability_report(original, modified)
+
+        # Original keys
+        assert "style_similarity" in report
+        assert "embedding_similarity" in report
+        assert "robust_style_similarity" in report
+        assert "robust_embedding_similarity" in report
+        assert "protection_score" in report
+        assert "assessment" in report
+        assert "transform_style_similarities" in report
+        assert "transform_embedding_similarities" in report
+
+        # New warning keys
+        assert "metric_type" in report
+        assert "warning" in report
+        assert "disclaimer" in report