The problem. Trajectory methods (Monocle, Slingshot, PAGA from the single-cell day) order cells along lineages, but a progenitor doesn’t follow one predetermined path — it faces branching choices with uncertain outcomes. What you want isn’t just a position on a tree; it’s, for a given cell, the probability it ends up in each terminal state, and a measure of how much freedom it retains.

The idea. Palantir models differentiation as a Markov (stochastic) process over a nearest-neighbour graph of cells. From a chosen start cell it computes a high-resolution pseudotime and, crucially, the absorption probabilities into each terminal state — each cell’s likelihood of each fate. It also derives a differentiation-entropy per cell: high early when many fates remain open, falling as the cell commits. Applied to human bone-marrow data, it recovers hematopoietic branch points.

Why it matters. This reframes trajectories probabilistically, which is the more honest picture of fate decisions and connects to the stochastic-process thread that CellRank will push further. Entropy-as-plasticity is a genuinely useful readout. For developmental or regenerative questions — and tumor plasticity, near the STU’s interests — quantifying commitment beats a single ordering.

Verdict. A widely-used, elegant fate-probability method; results depend on start-cell choice and graph construction, and it assumes the sampled cells capture the continuum. Read it as differentiation told in probabilities rather than a fixed path.