Before the Numbers Appear
Chengwan Repair Station receives a request to repair tables and chairs. The itemized list has already been sent out; the receiving time slot still needs to be discussed. Someone says the deal is very likely to be closed; someone else says to wait for a reply first. Both remarks are easy to understand, yet they may not be answering the same question.
Does "closing the deal" mean accepting the scope of the task, or accepting the payment terms? Does it mean receiving a complete confirmation, or actually paying and completing the repair afterward? Does "very likely" refer to today, or to next week? If these relations have no assigned place, a single later outcome can be fitted into several different stories.
This book begins here. We do not rush to write down a probability; we first make the question into an object that can be jointly adjudicated. The number enters afterward: what material it depends on, what range it compares, which conditions should trigger an update once they change, and how it faces an outcome that does not follow the direction of hope.
Probability is not a precise veneer applied to a vague sentence. It is a conditional representation that lets judgments under limited observation express degree, be provisionally adopted, and remain bound by later material. One cannot declare a judgment reliable merely because the number falls between zero and one.
Why It Is Called a Probability Game
The word "game" reminds us that judgment has explicit tasks and rules, yet reality cannot be compressed until only the rules remain. The event definition decides what counts as holding; the information cutoff decides what could be used at the time; the recording rules decide how comparison proceeds afterward. These are not optional decorations outside the number.
If the original rules can be changed after the outcome, the record keeper can always obtain a better-looking past. The window can be extended, the conditions can be loosened, the most favorable version of the forecast can be kept. Such a game is easy to win, but it loses the relation by which judgment is constrained by outcomes.
We want to preserve a different process: the question is made explicit first, the material has a traceable provenance, adoption has a time point, and the outcome is adjudicated by the original definition. Revision can happen, and the old judgment can still be returned to. Reliability is progressively examined through this kind of continuation; it is not conferred by a single lucky guess.
This book therefore promises no method for evading every error. It helps readers identify at which layer an error occurred and which work still retains value. Ambiguous material, a model's limited applicability, and an unfavorable realization need not be conflated into one and the same failure.
How a Fictional Case Works
Chengwan Repair Station, Gu Ning, Tang Ke, Ye Cheng, and all request records are fictional. Gu Ning maintains the forecasting relationship, Tang Ke handles communication, and Ye Cheng helps organize historical material. They are not real subjects interviewed by the author, nor a sample from some industry survey.
The case continues along R17 (the registration number of the table-and-chair repair request that opens the book) so that readers can see how material enters at different times. New replies on the main line will have their acquisition time points made explicit; separately arranged comparison scenarios will state that they have not joined the main line. The calculations can be checked, yet fictional quantities cannot become real statistics by virtue of the calculations.
Likewise, the characters' additional judgments are made public as assumptions. Readers may disagree with a model value or a weight; they may hold that it is currently more appropriate to continue without reporting a number. Raising such objections does not require first supplying an accurate replacement probability—only pointing out the relevant grounds and the gaps.
This gives the thought experiment an entrance for inspection. It displays the relations among formation, adoption, and reopening; it does not write every choice of the characters as an ideal answer. Making an assumption public does not mean the assumption has been shown sufficient; preserving a forecast does not mean the forecast has been shown accurate.
RC's Starting Point and the Constraints of the Domain
This book maintains RC's internal expository standpoint. Limited representation is the operationalization of RC's theory reduction—every theory is a reduced projection under a finite horizon, and this book applies that standpoint to the representation of judgment: it excerpts the current question from real relations and cannot exhaust all conditions. Processual completeness emphasizes that a judgment can connect to material, adoption, feedback, and reopening, rather than forming a permanent answer in one stroke.
These standpoints organize our questions; they do not obtain empirical validation for probability models on our behalf. The formula for conditional probability has its own mathematical premises, sample transplantation has its own scope requirements, and scoring has its own rule-bound character. Because they can be placed into the same narrative, one cannot say that one has proven the other.
Nor does this book quantify the available margin as a fixed unknown proportion or an interval width. Acknowledging that some conditions remain unobtained does not permit us to reject results that are already sufficient. When the original window does not hold, the non-holding should be acknowledged; the openness of reality cannot be used as a reason to keep the old model forever.
Feedback may promote calibration, or it may make the old explanation more rigid. The difference lies in whether adverse material can enter, whether the original report can be returned to, and whether the revision corresponds to concrete relations. Merely increasing the number of times the same claim is repeated accomplishes none of this work.
Four Parts Provide Four Kinds of Continuation
Part One begins with the object of judgment. Events, times, samples, comparison classes, the unknown, and model boundaries unfold in turn; what is delivered is a question into which a number can be placed, not a table of contents answering on the question's behalf.
Part Two treats how evidence changes judgment. Conditional grouping, reverse conditions, relative evidence, updating, source relations, and material quality each have their own duties. Agreement in computation is not independent verification, and multiple retellings do not automatically increase the number of originals.
Part Three examines the visible structure of confidence. Selection entry, unobserved states, competing explanations, comparison conditions, retrospective stories, and consensus can all alter the material that enters judgment. We check not only whether the number is right, but also what the number is currently standing for.
Part Four connects judgment to a continuing record. Actual reports, outcome adjudication, calibration, scoring, revision, provisional adoption, and handover allow a single event to enter subsequent learning. A complete ledger preserves the positions of the undecided and the erroneous; it cannot retain only successful displays.
How to Read and Use This Book
On a first reading, you can follow the case along without mastering every term in advance. When you see a fraction, first ask what the denominator is, what the conditions are, and whether it belongs to a historical frequency or a current estimate. When you see an update, first ask whether there is genuinely new content.
On a second reading, you can use the appendix along your own tasks. First write an event that can be adjudicated, then preserve the material time points and the actual adoption. Simple tasks can be recorded briefly; add fields when boundaries are disputed. There is no need to manufacture precision that has not been obtained just to fill in a form.
The formulas and links to primary sources in the book serve to check technical relations; they do not constitute a menu of tools detachable from conditions. Only after knowing the question and the inputs does one know whether a given formula may be used. The appendix likewise will not recommend the case's numbers as probabilities for the reader's real tasks.
Part 1 stops at the formation and calibration of judgment. Whether to act, how much to commit, whether failure can be borne, and how to exit must be carried forward by Part 2. A probability can become an input to action, yet it cannot by itself complete the choice among goals, constraints, and consequences.
We move from a phrase like "very likely" toward a judgment with somewhere to return. Returning is not endless hesitation; it is keeping clear, separately, what was known at the time, what was actually adopted, what the later outcome was, and what the next question is. Only then can numbers take part in the formation of common sense, rather than only in the expression of confidence.