Disclaimer

1. General

1. The Service gives you tools and information for early engineering work. It is not professional engineering, legal or regulatory advice.
2. Results depend on the inputs, data and settings you provide. Wrong or incomplete inputs give wrong results.

2. MADETOOL design results

1. MADETOOL produces conceptual and preliminary design estimates from your requirements and from public methods.
2. Results are not validated against a real aircraft or test. They are not certified, they are not an airworthiness approval and they do not show compliance with any regulation.
3. The DoD UAS group check uses the table named in the report (JP 3-30, 2019). It is not an FAA or other civil authorisation.
4. Example data shipped with MADETOOL are placeholders or synthetic values. Results built on them are illustrations only.
5. MADETOOL 1.0.0 was tested with GNU Octave 8.4. Five of the ten supported aircraft configurations have an example that converges; the other five have been tested for two sizing passes only, so treat first designs with them with extra care.
6. Configurations MADETOOL does not support are refused; they are never replaced by another configuration.
7. You must have every result checked by qualified engineers before any build, test or flight.

3. MASETOOL models

1. Each model is evaluated on a test set that is kept apart from training. Its metrics are results on the recorded test set of that job, not a promise of how it performs elsewhere.
2. No model is certified, universal, flight-ready or fit for safety-critical use.
3. Performance in the field depends on the camera, distance, light, weather, background and objects the model has not seen. You are responsible for validating a model on your own data, deploying and monitoring it, and using it lawfully (see Acceptable use in the Terms & Conditions).
4. Known limits of this version: - Training times, start times and finish times are planning estimates, not measurements. - For monochrome or thermal cameras, classes without real gray images can be trained on colour library images converted to gray, but only if you allow it. Converted colour images are not thermal images. The package reports when this was used. - Only classes in the MASETOOL catalogue are trained directly. Other classes need review first and may be refused. - Speed and latency figures are given only when they were measured on the stated hardware. - TensorRT engines must be built and tested on the target device.

4. Third-party software

The Service and the model packages use open-source software, including GNU Octave, Ultralytics YOLO, PyTorch and ONNX. Each is licensed under its own terms and comes without warranty from its authors. [PLACEHOLDER — lawyer to confirm: licence obligations for delivered models and code, in particular Ultralytics YOLO (AGPL-3.0 unless an enterprise licence applies)]

5. No reliance for safety-critical decisions

Do not rely on the Service alone for decisions that could affect safety, people or property. Our liability is set out in the Terms & Conditions.