Thank you for the response.
You are correct in regards to us trying to avoid forward predicting simulated proxies to save costs. Given the caveats you pointed out and our release timeline, it’s too risky to make considerable engine changes.
Our overall goal with this post was to glean for more quick performance gains from the referenced article, but it’s understandable if there are none left due to engine updates since then. If time permits, we may consider revisiting the suggested approach in the future.