Test-time adaptation lets time series forecasters update after deployment as new observations arrive. According to the paper, existing adaptation objectives focus primarily on prediction errors at individual time steps, and the authors argue this is not enough for financial data.

The proposed solution, DPAL, is presented as a loss function designed for this setting. The abstract is truncated, so the exact formulation and experimental results are not fully described in the announcement.

The work is available on arXiv as a new submission, with the identifier 2610.14376. Further details would require reading the full paper.