![\"Beware](\"assets/evidence-viewer/miracle-priors-hard-coded-bias-historical-method.png\")
Miracle priors should be stated openly. They should not be smuggled into the room as a hidden veto. A skeptical prior can test a claim; it should not silently decide the claim before the witnesses, documents, rival explanations, and surrounding worldview evidence are allowed to speak.
\nThat is the problem with hard-coded bias. If a model begins by assigning a miracle claim something like a 10^-40 prior, it has not made a sober historical judgment. It has nailed the courtroom door shut, then congratulated itself for hearing no testimony. Mathematically, that is very close to sticking your fingers in your ears and saying, \"I cannot hear you,\" only with more decimal places.
\nI have seen this most plainly in an X AI / Grok run, where a preloaded anti-miracle prior was defended by appealing to the EU AI Act. That was the category mistake. The EU AI Act may matter for governing AI systems; it is not a metaphysical proof that dead men cannot rise. A regulation can tell a machine how to behave. It cannot decide, in advance, what history is allowed to contain. That kind of move does not weigh evidence. It changes the subject in a suit and tie.
\nThe fair question is not, \"Do miracles happen every Tuesday?\" Of course not. A miracle, if real, is not ordinary weather. The fair question is whether the claim is being judged under the right background assumptions.
\nIf strict naturalism is assumed from the start, then miracles will begin with a very low prior because the system has already defined them out of bounds. But if God is live, if theism is live, and if Christ-identity evidence is already in the field, then the prior is not the same question. The background has changed. The historical evidence still must be tested, but it should be tested in the right courtroom.
\nLewis would put the point more simply: before deciding whether God has acted, do not quietly assume that God cannot act. Lennox often makes the same kind of distinction in another key: science tells us what normally happens when nature is left to itself; it does not prove that the Author of nature can never do anything with His own book.
\nHume's pressure is not silly. Testimony can be weak. People can be mistaken. Religious excitement can outrun evidence. Rare claims deserve careful scrutiny. The Signal should not answer that by waving a pious hand and pretending the problem is gone.
\nBut Hume should not become a slogan-level veto either. If the prior is set so low that no realistic historical evidence could ever matter, then the method has stopped being historical. It is no longer asking what happened. It is protecting a rule that says what is allowed to have happened.
\nThe Resurrection claim should not be accepted because Christians want it to be true. It should not be rejected because a prior was chosen that makes it impossible before the evidence begins. Both moves are unfair. One is enthusiasm pretending to be evidence. The other is skepticism pretending to be mathematics.
\nThe Signal's job is to keep both temptations in the light. If the evidence is weak, let it be weak. If the evidence is stronger than a closed system expected, let it speak. A prior is a starting place, not a gag order.
\nThis item currently has no active Bayes factors. It is unweighted explanatory methodology for miracle-prior sensitivity and scoring governance. Its purpose is to prevent two opposite errors: smuggling in a high miracle prior without argument, or smuggling in a near-zero prior that functions as a hidden worldview veto.
\nUse the citation list attached to this evidence item for source audit. Key conversation partners include Hume on miracle testimony, Earman's critique of Humean veto reasoning, and McGrew/McGrew on Bayesian cumulative-case responses.
\n