Org card: dataset and replay kit are public

README.md

@@ -16,7 +16,7 @@ General Machines is an applied AI research lab building open datasets, evals and
 
 ## Releases
 
-- **CartTrace-100k**: 101,416 journeys of an AI shopping agent (Qwen3.8-27B) through checkout and payment over five commerce protocols (UCP, ACP, AP2, MPP and WebMCP), across 1,044 stand-in stores in 25 categories. Every journey is scored step by step by a judge. Over all 101,416 journeys the agent succeeds 68.7% of the time (95% CI 68.1–69.2), and 65.6% when every claim it makes must be grounded. All journeys run in a sandbox; nothing was bought on a real store. The data (CC BY 4.0) and the replay kit (Apache-2.0) open here soon. [Release page](https://generalmachines.ai/datasets/carttrace-100k) · [Explorer](https://www.machinecommerce.co/carttrace).
+- **[CartTrace-100k](https://huggingface.co/datasets/generalmachines/CartTrace-100k)**: 101,416 journeys of an AI shopping agent (Qwen3.8-27B) through checkout and payment over five commerce protocols (UCP, ACP, AP2, MPP and WebMCP), across 1,044 stand-in stores in 25 categories. Every journey is scored step by step by a judge. Over all 101,416 journeys the agent succeeds 68.7% of the time (95% CI 68.1–69.2), and 65.6% when every claim it makes must be grounded. All journeys run in a sandbox; nothing was bought on a real store. Data CC BY 4.0, replay kit Apache-2.0. [Release page](https://generalmachines.ai/datasets/carttrace-100k) · [Explorer](https://www.machinecommerce.co/carttrace) · [Replay kit](https://github.com/generalmachines/cartTrace-100k).
 
 Next: Commerce-1, a planned small model trained on CartTrace that predicts when a shopping agent is about to fail. The replay kit runs any model, open or frontier, through the same stores.