Update README.md

README.md

@@ -28,4 +28,18 @@ Benchmarked on 120,000 samples across six input magnitude ranges. Metrics: MAE,
 
 ## Limitations
 
-Precision degrades for extremely large magnitude inputs (e.g., >500,000), indicated by increased MAE/MSE, although R2 remains high.
+Precision degrades for extremely large magnitude inputs (e.g., >500,000), indicated by increased MAE/MSE, although R2 remains high.
+
+## Important Disclaimer, Safety Warning, & Liability Limitation
+
+This model is provided strictly "as is" and "with all faults," without warranty of any kind, express or implied.
+
+*   **Statistical Approximation vs. Exact Arithmetic:** This is a statistical machine learning model, not a hardware-level Arithmetic Logic Unit (ALU). Although optimized using double precision (`torch.float64`), it operates via continuous regression approximation. It **does not** guarantee exact mathematical results. Even with the reduced error rates in v2.0, the absolute error (MAE of 0.706 at high ranges) represents an incorrect mathematical sum. Under no circumstances should this model be deployed in accounting, financial ledgers, navigation, structural engineering, or medical dosage systems where exact addition is mandatory.
+*   **Strict Double Precision (`torch.float64`) Requirement:** This model's weights are configured specifically for double precision. Feeding standard single-precision tensors (`torch.float32`) will result in runtime casting errors, model exceptions, or severe degradation of accuracy. Users are solely responsible for ensuring input compliance.
+*   **Scale Limits and Degradation:** Despite optimization, the model continues to exhibit accuracy degradation as input scale increases, particularly beyond 500,000. It is highly susceptible to extrapolation errors when presented with out-of-distribution values.
+*   **Metric Misinterpretation ($R^2$ Illusion):** The $R^2$ value of `1.000` is a mathematical side effect of high variance at scale and does not mean the model's predictions are error-free. Absolute error metrics (MAE/MSE) must be reviewed to understand actual operational deviations.
+*   **Deserialization Security Risk:** Like all PyTorch models saved as serialized state dictionaries, there is an inherent risk of arbitrary code execution upon loading. When importing the model parameters, users are strongly advised to enforce strict security practices by specifying `weights_only=True` in the PyTorch load call:
+    ```python
+    torch.load("model.pth", map_location=device, weights_only=True)
+    ```
+*   **Assumption of Risk:** The end-user assumes all liability, financial or otherwise, arising from the utilization, deployment, or integration of this model into any hardware or software system.