astronomy-the-joint-posterior-is-the-parameter-result

# One set of axes, every model, every parameter the source tabulates When the claim under test is that two models absorb the same data by shifting the parameters in *different directions*, the claim does not live in any marginal. It lives in the joint distribution: which way each model moves in the expansion-rate–matter-density plane, whether a shift runs along a degeneracy or across it, how far apart two models' contours are in the plane where they are closest, whether they overlap at all. The exhibit that carries that is one figure — the triangle plot, every model overlaid on one set of axes, filled 68% and 95% contours below the diagonal, the one-dimensional marginals on it — and it is the first thing a reader in this field looks for. `the-canonical-figure` gives the general rule; this is the version with the decisions in it. **The parameter list is the source's, not your sampler's.** Take the axes from the rows the source tabulates *for every model in the comparison*, in the source's order and under the source's symbols. A parameter only one model carries cannot be a shared axis — it belongs in its own panel or a second figure — and where the source's caption says which subset it drew, that subset is the grid. Two consequences follow and both get made backwards by default. A parameter your own likelihood never constrains stays on the grid, carrying the published constraint alone — see `astronomy-sample-the-published-table-into-chains` for where its samples come from — a...