A volunteer firehouse tags incident run reports. One overnight job changes the instruction set, the learning rate, and the base checkpoint together, and the crew cannot say which change helped. What is the design fault?
Select an answer to reveal the explanation.
Short Explanation
Instruction set, learning rate, and base checkpoint all moved in one job. The crew cannot name the cause. Change one factor per run and hold the rest. Kernels, test-envelope peeks, and a multimodal recipe are not the missing control.
Full Explanation
A comparison can be attributed only when one independent variable changes at a time. Mixing an instruction set, a learning rate, and a base checkpoint in a single job leaves the crew unable to name the cause of any score move. The repair is an ablation-style sequence: change one factor, hold the rest, then compare. Kernels, leaked test peeks, and multimodal system design are not the missing control.