Formalize the "killer demo" of Bayesian + PAX — extract quantitative effect sizes from a knowledge-base pack's findings.json, convert them into informative priors on corresponding regression coefficients, and fit Bayesian linear regression on new data. Produces a posterior that formally blends prior literature with current evidence, with posterior predictive checks for model adequacy.
{
"confidence_min": "moderate",
"min_findings_with_effects": 3,
"required_fields": [
"construct_ids",
"effect_size_value",
"effect_size_se"
]
}{
"coefficient_prior_mean": "effect_size_value",
"coefficient_prior_sd": "effect_size_se",
"match_on": "construct_ids",
"scale_inflation": 1.5
}{
"intercept_prior": "normal(0, 10)",
"reference": "gelman_2006_variance_priors",
"sigma_prior": "half_cauchy(0, 2.5)",
"unmapped_coef_prior": "normal(0, 2.5)"
}{
"adapt_delta": 0.95,
"chains": 4,
"iter_sampling": 2000,
"iter_warmup": 1000,
"sampler": "no_u_turn_sampler"
}{
"max_rhat": 1.01,
"min_effective_sample_size": 400
}{
"flag_bayesian_p_extreme": 0.05,
"n_replications": 1000,
"test_quantities": [
"mean",
"sd",
"q05",
"q95"
]
}{
"report": "posterior_summary_comparison",
"scale_multipliers": [
0.5,
1,
2,
5
]
}{
"include_prior_comparison": true,
"statistics": [
"posterior_mean",
"posterior_median",
"credible_interval_95",
"credible_interval_50",
"probability_direction"
]
}