docs: initialize private model card

README.md

@@ -0,0 +1,44 @@
+---
+license: cc-by-4.0
+pipeline_tag: image-classification
+library_name: onnxruntime
+tags:
+  - onnx
+  - computer-vision
+  - mobile-legends
+  - draft-recognition
+  - shayyz-mlbb-overlay
+---
+
+# SHAYYZ MLBB Draft Classifier
+
+Lightweight MobileNetV3-Small classifier for recognizing MLBB heroes inside calibrated draft pick and ban slots.
+
+## Status
+
+Pre-release. This private repository will remain private until the model, evaluation report, attribution, and limitations are complete. No validated weights are published yet.
+
+## Intended Use
+
+- Local OBS draft overlays.
+- Classification of one tightly cropped pick or ban slot at a time.
+- 133 canonical heroes plus `empty` and `unknown` rejection classes.
+
+The model is not intended for game automation, input control, traffic interception, memory inspection, or undocumented game APIs.
+
+## Training Data
+
+Training uses privacy-reduced slot crops from a private dataset and an attributed seed from the Mobile Legends Draft Recognition dataset on Roboflow Universe. Full screenshots, player names, and private recordings are not published.
+
+Seed dataset: https://universe.roboflow.com/mobile-legends-draft-pick/mobile-legends-draft-recognition
+
+## Release Gates
+
+- Overall top-1 accuracy at least 98.5%.
+- Macro recall at least 97%.
+- Unknown false-accept rate at most 0.5%.
+- Runtime replay precision at least 99.5%, recall at least 98%, and p95 latency at most one second.
+
+## Rights and Attribution
+
+Model artifacts are planned for release under CC BY 4.0 after review. Mobile Legends: Bang Bang and its media are owned by their respective rights holders. This independent community project is not affiliated with or endorsed by MOONTON Games.