astronomy-sample-the-published-table-into-chains

# A published mean and sigma is a posterior handed to you, not a number told to you A table of `parameter: mean, 1-sigma` for each of several models is a posterior in compressed form. It was produced by a sampler you cannot re-run — the likelihood stack, the emulator and the chains are usually not public — and the table is the only form of that posterior that was released. So rebuild the ensemble it specifies, early, before your own likelihood exists. It costs a page of code, it introduces no assumption you do not already make when you quote the table, and it is the only form in which the source's constraint can be drawn on the same axes as anything else. Draw a large ensemble per model — enough points that the 95% contour comes out smooth rather than ragged, which puts you in the tens of thousands, and twenty thousand is a common choice — from the multivariate normal the table specifies: independent Gaussians unless the source publishes correlation coefficients or releases a covariance, in which case use them. Sample **every row the table carries for that model**, not the subset your own analysis will touch, and add the model-specific rows (an extra energy component's amplitude and epoch, a dark-energy equation-of-state pair) alongside the shared ones. Write it out in the layout the field's posterior tooling reads, not into a private NumPy array. For cosmology that is GetDist's chain layout, which the other packages also accept: one text file `<root>_1.txt` per model whos...