=

The passage around this formula

The challenger adds $G_{tr}$ , its uncertainty-aware estimate, and a preregistered interaction with task horizon: $\operatorname{logit}\Pr(Y_{imtr}=1)=\text{baseline}+ \gamma_1G_{tr}+\gamma_2G_{tr}\log(1+h_t)$. The primary scoreboard is not an in-sample coefficient. It is leave-one-repository-out predictive log loss, calibration slope, and Brier score, each computed before inspecting a graph of agents. The transfer scoreboard withholds whole task families or repositories, then asks which model better predicts the outcome from the descriptions and annotations already frozen. A positive in-sample $\gamma_1$ is not a result; a redaction variable is almost designed to correlate with difficulty. Only lower held-out error and better calibration earn the claim that it carries…

Read this part in the article →