Update README.md

README.md

@@ -16,10 +16,10 @@ environments (Wiki, News, Shop) rendered in six UI eras.
 | `test.csv` | 103 | Same schema, held-out split. |
 | `test.raw.json` | 231 | The full runnable benchmark (train + test): each task carries its `intent`, `sites`, `start_url`, and a **deterministic evaluator** spec. |
 
-## From goals to tasks
+## Connecting Goals to Tasks
 
-This dataset stores **231 goals** (128 train + 103 test), not the full task set.
-Each goal — with its reference answer and plan — is written to be independent of
+This dataset contains **231 goals** (128 train + 103 test), not the full task set.
+Each goal (with its reference answer and plan) is written to be independent of
 any particular UI era, so at evaluation time the environment renders every goal in
 all **six UI versions**, giving **231 × 6 = 1,386 tasks**. You only see one entry
 per goal here.