Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper critiques the conventional approach to synchronizing multimodal recordings in medical world models, highlighting that such synchronization can inadvertently erase critical task-relevant information. The authors introduce a four-regime taxonomy to assess where evidence of target events survives during this synchronization process and operationalize it through a pretraining audit called EndoClock. Their findings, illustrated through echocardiography examples, reveal that essential measurement events can be lost due to synchronization, emphasizing the need to maintain native observation processes to avoid information loss.
Synchronizing medical data can erase crucial evidence, as shown by the loss of key echocardiography events during data acquisition.
Medical world models commonly learn from multimodal recordings synchronized onto a fixed-rate grid. This preprocessing resamples each native stream onto a shared time axis. Each stream has an observation clock that governs when observations are emitted or updated. When this clock depends on the latent or acquisition state, it is endogenous. In such settings, synchronization may not be neutral and can erase task-relevant evidence before the model sees the data. We introduce a four-regime taxonomy that characterizes where the evidence needed to distinguish a target event or state survives. The relevant witness may remain in the sampled values, in grid-cell update patterns, in native timing, or only in an external acquisition channel. EndoClock operationalizes this taxonomy as a conservative pretraining audit. It reports the lowest witness-bearing representation supported by the available evidence, or unresolved when no regime can be established. We illustrate this failure in echocardiography, where B-mode video write-outs cease during pulsed-wave Doppler acquisition while the corresponding measurement events remain recorded only in an external acquisition log. This work is a preliminary failure alert and executable audit. Its practical message is to preserve the native observation process long enough to determine whether synchronization has erased information required by the intended task.