Attribution of the cross-device performance gap in the uranium mine
Determine whether the discrepancy between the single screening-run cross-device performance and the repeated-training ensemble performance in the uranium mine is caused by training stochasticity or by an unlogged difference in the evaluation setup.
References
Whether the gap is therefore due to training stochasticity or to an unlogged difference in the evaluation set-up cannot be decided for frame B.
— Accountable and uncertainty-aware evaluation of sensor-based AI under distribution shift: devices, subjects, and nearly three years underground
(2609.09257 - Platte et al., 8 Sep 2026) in Section 8.2, 'Retrospective verdicts'
We attribute this to numerical differences propagating through hundreds of generated tokens, though we cannot exclude an unrecorded difference in the prompt.
— Assessing Readability with LLMs: The Role of Reasoning and Few-Shot Prompting
(2609.24650 - Thieffry et al., 21 Sep 2026) in Section 6, Limitations, item 1