Update README.md

README.md

@@ -27,6 +27,12 @@ Dataset** product. It contains roughly **1 / 60th of the full dataset** at
 identical schema, attacker-tier distribution, and statistical fingerprint, so
 you can evaluate fit before licensing the full product.
 
+> 🤖 **Trained baseline available:**
+> [**xpertsystems/cyb002-baseline-classifier**](https://huggingface.co/xpertsystems/cyb002-baseline-classifier)
+> — XGBoost + PyTorch MLP for 10-class MITRE ATT&CK kill-chain phase
+> prediction, group-aware split by campaign, ablation evidence,
+> honest limitations in the model card.
+
 | File                    | Rows (sample) | Rows (full)   | Description                                  |
 |-------------------------|---------------|---------------|----------------------------------------------|
 | `network_topology.csv`  | ~651        | ~3,200        | Network segments and asset inventory         |
@@ -54,6 +60,28 @@ OT/ICS environments, with:
 - **Ransomware deployment, supply chain compromise, and exfiltration**
   outcome paths
 
+## Trained Baseline Available
+
+A working baseline classifier trained on this sample is published at
+**[xpertsystems/cyb002-baseline-classifier](https://huggingface.co/xpertsystems/cyb002-baseline-classifier)**.
+
+| Component | Detail |
+|---|---|
+| Task | 10-class MITRE ATT&CK kill-chain phase classification |
+| Models | XGBoost (`model_xgb.json`) + PyTorch MLP (`model_mlp.safetensors`) |
+| Features | 90 (after one-hot encoding); pipeline included as `feature_engineering.py` |
+| Split | **Group-aware by campaign_id** — train/val/test campaigns disjoint |
+| Demo | `inference_example.ipynb` — end-to-end copy-paste |
+| Headline metrics | XGBoost macro ROC-AUC 0.86; accuracy 47% (vs 19% always-majority baseline) |
+
+The model card documents the three columns excluded for label leakage
+(`technique_id`, `technique_name`, `tactic_category`), an ablation
+showing `timestep` carries most of the phase signal, and six explicit
+limitations including the gap between synthetic and real attack
+telemetry. Late-stage phases (collection / exfiltration / impact) are
+genuinely harder for a flat-tabular event-level model — the baseline
+exposes this honestly.
+
 ## Calibrated Benchmark Targets
 
 The full product is calibrated to 12 benchmark metrics drawn from
@@ -123,7 +151,8 @@ log and asset inventory schemas respectively.
 
 ## Suggested Use Cases
 
-- Training **kill-chain phase classifiers** (predict next ATT&CK phase)
+- Training **kill-chain phase classifiers** (predict next ATT&CK phase) —
+  [worked example available](https://huggingface.co/xpertsystems/cyb002-baseline-classifier)
 - Benchmarking **APT detection** algorithms (long dwell, low stealth_score)
 - **Campaign outcome prediction** (success / detected / blocked / aborted)
 - **MTTD / MTTR forecasting** under varying defender maturity
@@ -152,6 +181,10 @@ y = attacks["detected"].astype(int)
 y_outcome = campaigns["campaign_outcome"]
 ```
 
+For a worked end-to-end example with the 10-class kill-chain phase target,
+group-aware splitting, and feature engineering, see the inference notebook
+in the [baseline classifier repo](https://huggingface.co/xpertsystems/cyb002-baseline-classifier/blob/main/inference_example.ipynb).
+
 ## License
 
 This **sample** is released under **CC-BY-NC-4.0** (free for non-commercial