[{"data":1,"prerenderedAt":73},["ShallowReactive",2],{"mental-model-bayesian-updating-en":3},{"title":4,"description":5,"summary":6,"sections":7,"visual":43,"related":51,"sources":64},"Bayesian Updating","Adjust confidence when new evidence arrives. Learn the logic of Bayes' theorem, why context matters and how to revise a training belief without false precision.","Begin with a provisional belief and change your confidence when evidence arrives. The important question is how surprising that evidence would be under competing explanations.",[8,14,20,26,32,37],{"heading":9,"kind":10,"paragraphs":11},"How much should one failed attempt change your mind?","idea",[12,13],"You expect a pass to work against a particular frame, then it fails. Should you abandon it? Not necessarily. A single failure might be common even when the pass is useful. But if the same problem keeps appearing under the right conditions, your confidence should move. Bayesian updating gives this ordinary question a more precise logic.","Bayes' theorem is a mathematical relationship between conditional probabilities, associated with Thomas Bayes and later development by Pierre-Simon Laplace. Using it to update a belief requires a starting probability and assumptions about how likely the evidence is under different explanations. Skill First uses the reasoning as a practical habit. Most training situations do not supply trustworthy numbers, so qualitative updates are often more honest than a percentage invented after a round.",{"heading":15,"kind":10,"paragraphs":16},"A starting belief, evidence and a new estimate",[17,18,19],"The prior is your confidence before the new observation. The likelihood describes how probable that observation would be if an explanation were true. The posterior is your revised confidence after considering the evidence. The crucial comparison is not whether the evidence fits your favourite explanation at all. It is whether it fits that explanation better than the alternatives.","Here is a made-up numerical example to show the calculation, not a prediction about real athletes. Before a positional round, suppose you assign equal chances to a partner usually choosing response A or response B. You estimate that they would choose A on the first exchange 80% of the time if A is their usual response, and 20% if B is. You observe A. Bayes' rule gives an 80% revised probability that A is their usual response: 0.5 × 0.8 divided by the sum of 0.5 × 0.8 and 0.5 × 0.2.","Notice what the calculation does not say. It does not prove the partner will choose A next time. It also depends entirely on the estimates you supplied. If those estimates are poor, the clean arithmetic does not rescue them. A different partner, rule or starting position can make the original assumptions unsuitable.",{"heading":21,"kind":22,"paragraphs":23},"Use the idea without pretending to be a calculator","practice",[24,25],"A grappler believes their preferred pass is reliable against a certain response. They keep a short record of relevant attempts with several partners. Repeated success in the expected situation strengthens confidence there. Repeated failure against a particular frame reduces confidence for that condition, while leaving other applications open.","Ask what else explains the observation. Was the entry different? Did the partner offer a new grip? Was the round unusually cooperative? Evidence should change the claim at the level it actually tests. A poor attempt while exhausted says less about the technique in normal conditions than a consistent failure during controlled, well-executed attempts.",{"heading":27,"paragraphs":28,"kind":31},"Changing confidence is not yet choosing an action",[29,30],"Suppose a partner increasingly seems likely to step back after your jab. That is an update about their behaviour, not an instruction to chase them every time. The appropriate action also depends on what happens if your estimate is wrong. A less likely response may deserve attention when it would leave you badly positioned. Keep an option for recovery rather than treating the most likely outcome as the only possible one.","This separates two questions that often get blurred in training: what do I now expect, and what should I do given those expectations? Your coach can help consider the consequences and available alternatives. You might become more confident about the partner's preference while keeping the same cautious first action. Or a small change in confidence might justify trying a different response in a controlled drill. Updating should improve the choice available to you, not automatically produce a bigger gamble.","caution",{"heading":33,"kind":31,"paragraphs":34},"Do not count the same lesson five times",[35,36],"Five clips from one exchange are not five independent attempts. Advice repeated by several accounts may come from the same original source. Evidence becomes less informative when you count correlated observations as separate confirmation. Also look for examples that challenge your claim; otherwise updating becomes a formal-sounding name for confirmation bias.","A prior is not a licence to keep believing whatever you already prefer. Make your starting assumption visible and allow new information to move it. You rarely need to announce certainty. It is enough to say the pass looks promising here, this reaction now seems more likely, or we need a better test.",{"heading":38,"kind":39,"paragraphs":40},"Leave the claim ready for the next round","experiment",[41,42],"Write one belief, the conditions where it applies and what would increase or reduce confidence. Gather observations during appropriate practice and review the pattern with a coach. Update the wording before choosing the next test. Keep uncertainty visible instead of changing every opinion after one dramatic result.","Probabilistic thinking considers possible outcomes. Bayesian updating changes confidence as evidence arrives. Prediction error compares expectation with observation, and the scientific method helps make that comparison testable. A useful update leaves you with a better question for the next session, not a number that looks authoritative but has no reliable basis.",{"description":44,"labels":45},"Confidence shifts from an initial estimate through new evidence to a revised estimate while uncertainty remains visible.",[46,47,48,49,50],"Prior","New evidence","Compare explanations","Posterior","Next test",[52,55,58,61],{"id":53,"reason":54},"probabilistic-thinking","Updating works with degrees of confidence and several possible outcomes.",{"id":56,"reason":57},"confirmation-bias","Evidence should move your belief, including when it contradicts a preference.",{"id":59,"reason":60},"prediction-error","A mismatch between expectation and observation can prompt an update.",{"id":62,"reason":63},"scientific-method","A clear test helps determine which observation actually bears on the claim.",[65,69],{"title":66,"url":67,"note":68},"Stanford Encyclopedia of Philosophy: Bayes' Theorem","https:\u002F\u002Fplato.stanford.edu\u002Fentries\u002Fbayes-theorem\u002F","Mathematical relationship and history; the numerical scenario is an invented teaching example.",{"title":70,"url":71,"note":72},"Stanford Encyclopedia of Philosophy: Bayesian Epistemology","https:\u002F\u002Fplato.stanford.edu\u002Fentries\u002Fepistemology-bayesian\u002F","Explains priors, conditionalisation and assumptions involved in updating degrees of belief.",1791220422256]