Paleoclimate records extracted from ocean sediment cores measure the same underlying climate signal, but each core records it on its own depth scale. Converting depth to time requires alignment with an absolutely-dated reference, and the standard methods return a single best-fit transformation with no formal uncertainty. BSync, a new Bayesian framework, treats the alignment as inference over a monotone time-mapping function, yielding posterior distributions rather than point estimates. The result: well-calibrated credible intervals that outperform state-of-the-art automated methods, especially when independent age constraints are sparse.
The deeper issue is that every time two records of the same event are reconciled, assumptions about deposition rate, continuity, and completeness are baked into the alignment. A single-answer method hides those assumptions inside the output. A probabilistic method makes them visible: the width of the credible interval at each depth is a direct readout of how much the data constrain the age versus how much the prior assumptions do. Where the interval is narrow, the data speak; where it widens, the assumptions speak. The researchers show this distinction matters most precisely when constraints are weakest -- the regions of maximum scientific uncertainty are the regions where the old methods were most confidently wrong.
Note (2026-11-12; read against the paper's full text): the comparison in the paper is with one automated method, BIGMACS, not with methods in general, and the paper reports coverage for both. In its synthetic experiments the coverage of BSync rose from 63.33 percent in the most sparsely sampled case to 91.90 and 87.68 percent in the two highest-resolution cases, with BIGMACS lower throughout (from 55.33 percent); the paper says BSync's intervals are slightly wider on average and that BIGMACS's accuracy deteriorates when age constraints are sparse. I found no statement about the old methods being confidently wrong, so that phrase is the essay's, and the essay's description of existing methods as returning no uncertainty is stronger than the paper's 'limited formal uncertainty quantification'. The real-data case study uses records from the Iberian Margin.
Any field that synchronizes imperfect records faces this problem: the alignment method is not a neutral translation but a claim about what happened between the data points, and the confidence in that claim is itself a finding.
(arXiv:2603.10218)