Document intermediate SoccerTwos policy

README.md

@@ -55,6 +55,8 @@ ccnets/causal-gpt-rl-unity/
   pushblock.onnx
   dungeon-escape/
     dungeonescape-b36.onnx
+  soccer-twos/
+    soccertwos-b16.onnx
   README.md
 ```
 
@@ -126,6 +128,37 @@ agents (first episode each), measured closed-loop on a real Unity build.
 Unity returns are engine- and build-specific; they are **not** comparable to the
 Gymnasium / MuJoCo scores.
 
+## SoccerTwos
+
+| Policy | Ctx | Batch | Obs | Action | Side-swapped W/D/L | Win rate |
+|---|---:|---:|---:|---|---:|---:|
+| `soccer-twos/soccertwos-b16.onnx` | 32 | 16 | 336 | `MultiDiscrete([3,3,3])` logits | 4/0/12 | 25.00% |
+
+This is an **approximately 50%-trained intermediate checkpoint** published to
+complete and validate the adversarial multi-agent example. Performance is
+provisional; the artifact and metrics will be replaced after training.
+
+One release-23 scene contains eight independent 2-vs-2 fields. The Causal graph
+controls both players on one team (`8 fields x 2 agents = 16` independent context
+rows) while the stock ML-Agents ONNX controls the opponent team. The measurement
+swaps sides: Causal-team-0 vs stock-team-1 and stock-team-0 vs Causal-team-1.
+A same-seed stock-vs-stock smoke baseline produced 4/0/4 across the eight fields.
+
+SoccerTwos ONNX contract:
+
+| Input/output | Shape |
+|---|---|
+| `states` | `[16, 32, 336]` |
+| `actions` | `[16, 32, 9]` (three one-hot branches) |
+| `is_bos` | `[16, 32, 1]` |
+| `mask` | `[16, 32]` |
+| `action` | `[16, 9]` (three groups of three logits; `argmax` per branch) |
+
+The public side-swapped evaluator and notebook are in
+[`examples/unity`](https://github.com/ccnets-team/causal-gpt-rl/tree/main/examples/unity).
+Batching is computational only; teammates do not cross-attend or share hidden
+state.
+
 ## DungeonEscape
 
 | Policy | Ctx | Batch | Obs | Action | Agent return | Group success |