Update README.md

README.md

@@ -13,7 +13,7 @@ tags:
 
 # SpotVortex TFT Dual-Head Risk Model (ONNX)
 
-Production ONNX artifact used by SpotVortex for spot instance risk inference. Outputs dual risk scores, capacity and runtime per instance type and availability zone.
+Production ONNX artifact used by SpotVortex for spot instance risk inference. Outputs dual risk scores — capacity and runtime — per instance type and availability zone.
 
 ---
 
@@ -36,4 +36,46 @@ Training data is sourced from **SpotLake** spot market telemetry:
 | Gen | Compute (`c`) | General Purpose (`m`) | Memory (`r`) | Burstable (`t`) |
 |:---:|---|---|---|---|
 | **5** | `c5` `c5a` `c5ad` `c5d` `c5n` | `m5` `m5a` `m5ad` `m5d` `m5dn` `m5n` `m5zn` | `r5` `r5a` `r5ad` `r5b` `r5d` `r5dn` `r5n` | `t2` `t3` `t3a` |
-|
+| **6** | `c6a` `c6g` `c6gd` `c6gn` `c6i` `c6id` `c6in` | `m6a` `m6g` `m6gd` `m6i` `m6id` `m6idn` `m6in` | `r6a` `r6g` `r6gd` `r6i` `r6id` `r6idn` `r6in` | — |
+| **7** | `c7a` `c7g` `c7gd` `c7gn` `c7i` `c7i-flex` | `m7a` `m7g` `m7gd` `m7i` `m7i-flex` | `r7a` `r7g` `r7gd` `r7i` `r7iz` | `t4g` |
+
+---
+
+## Artifact Contents
+
+| File | Description |
+|------|-------------|
+| `tft.onnx` | ONNX computation graph |
+| `tft.onnx.data` | External tensor data sidecar |
+| `MODEL_MANIFEST.json` | Checksums and model scope metadata |
+| `supported_instance_families.txt` | Flat list of supported families for quick inspection |
+| `LICENSE` | Non-commercial proprietary terms |
+
+---
+
+## Model Interface
+
+### Input
+
+| Property | Value |
+|----------|-------|
+| Tensor type | `float32` |
+| Shape | `[batch, timesteps, 10]` |
+| Features per timestep | 10 |
+
+### Outputs
+
+| Tensor Name | Description |
+|-------------|-------------|
+| `capacity_score` | Predicted spot capacity availability risk |
+| `runtime_score` | Predicted spot instance interruption/runtime risk |
+
+---
+
+## Usage & Restrictions
+
+| Topic | Detail |
+|-------|--------|
+| Runtime integration | Use the SpotVortex agent repository |
+| Commercial use | ❌ Prohibited without prior written permission — see `LICENSE` |
+| Data internals | Private data-preparation pipelines and tuning details are excluded |