Remove labeled prediction examples

README.md

@@ -99,7 +99,7 @@ deterministically at import time.
 Recommended public prediction format:
 
 ```json
-{"task_id": "v3_depth1_currency_lookup_t0", "answer": "USD", "model": "your-model-name"}
+{"task_id": "example_task_id", "answer": "<model-output>", "model": "your-model-name"}
 ```
 
 `RUN.md` has a fuller harness contract, adapter skeleton, prompt template, and a

RUN.md

@@ -36,7 +36,7 @@ The public split has no labels, so this produces predictions for inspection or
 private scoring. A useful prediction file format is JSONL:
 
 ```json
-{"task_id": "v3_depth1_currency_lookup_t0", "answer": "USD", "model": "your-model-name"}
+{"task_id": "example_task_id", "answer": "<model-output>", "model": "your-model-name"}
 ```
 
 Minimal adapter skeleton: