An engineering team develops an execution harness that significantly improves task performance on a specific model. Next, without modifying or retuning any control logic, they apply that exact harness to the provider's newly released successor model to evaluate whether the performance benefits persist. Which empirical test of harness scaling is being conducted?

Demonstrating Fixed-Model Improvement (Within-Model Lift) requires modifying or fine-tuning the model's underlying weights to support the new execution harness.

Explain the overarching objective of harness scaling and discuss how the three empirical evaluations build upon one another as progressively stronger tests of external execution controls.

Order the three empirical tests of harness scaling from the initial baseline evaluation to the progressively strongest evaluation.

Explain why the engineer's modifications invalidate the evaluation as a test of Cross-Generation Transfer (Frozen Transfer), and describe the protocol required to properly conduct this test.

An engineering team attempts to prove Beyond-Benchmark Generalization by evaluating their agent harness on held-out test splits from the same benchmark suite used during harness development. Why does this evaluation setup fail to satisfy the criteria for this test?

Match each empirical test of harness scaling to its core evaluation objective.

According to the definition of harness scaling, what primary objective do external execution controls aim to accomplish regarding a model's latent capabilities?

Analyze the methodological requirements and theoretical significance of Cross-Generation Transfer (Frozen Transfer). In your discussion, address what remains unchanged during this evaluation, why modifying control logic is impermissible, and what a successful outcome demonstrates about the harness.

Match each empirical test of harness scaling to the primary constraint or condition defining its evaluation protocol.

Beyond-Benchmark Generalization evaluates whether harness control principles improve performance on completely different, held-out task ___ beyond the development benchmark.

Arrange the procedural steps required to conduct a Cross-Generation Transfer (Frozen Transfer) evaluation in the correct chronological sequence.

Explain why the engineer's proposed adjustments violate the criteria for testing Beyond-Benchmark Generalization, and identify what the evaluation must test instead.

The Fixed-Model Improvement evaluation measures whether a refined execution harness can improve a fixed model's end-to-end task ___ rate.