What Should We Do After Fifty Percent?
At nine on Saturday morning, Gu Ning reopened the R17 record. On Thursday evening she had reported 0.5 for the event that complete confirmation of Version 1 would arrive within the original window. The window ended at 17:00 on Friday, and the available record was sufficient to adjudicate that the event had not occurred. The client finally said that the original receipt period would not work and asked to discuss it again the following week.
Tang Ke asked whether they should now stop. Ye Cheng thought that because the client remained willing to talk, the chance of continuing the following week was not low, so they could reserve some time. Faced with the same record, one leaned toward stopping and the other toward continuing. It looked as if one of them had failed to understand what fifty percent meant.
Gu Ning did not explain the mathematics of fifty percent again; Part 1 had already done that work. She first pointed out that the original 0.5 answered a question about an event that had ended. It was not a probability threshold for whether they should invest now. Even if a separate new event also happened to be estimated at 0.5, that number alone could not yield an action.
This chapter continues the fictional case of Chengwan Repair Station. Its people, resource quantities, and subsequent actions are thought experiments and do not represent operating data from the actual repair industry. Part 2 begins with a new action question: whether and how the station should commit resources to a possible next version of the terms. This question is labeled D17 and kept separate from the result of R17 Version 1 in Part 1.
A Closed Prediction Cannot Become a New Authorization
The original R17 event has failed to occur, and its 0.5 remains on the historical prediction page. If Gu Ning copied it onto Saturday's page and wrote, "It is still fifty percent, so continue," she would cross over two changes at once: the Version 1 window has closed, and no version for the following week has yet been formed.
The old report has not lost all value. It preserves how limited material was organized at the time and exposes weaknesses in the time conditions and the weights in the subjective model. A review can make the new action more careful, but it cannot make an old probability automatically describe a new event that has not yet been defined.
An action discussion must first decide what is being chosen. Is it sending one clarifying message, reserving a half-day work bay, buying materials in advance, or ending contact? These actions require different resources, alter different relations, and differ in how hard they are to exit. Calling all of them "continuing" makes one probability appear to authorize several depths of commitment at once.
Gu Ning therefore cited the old record on a new page without carrying over its number. At the top she wrote: "R17 Version 1 is closed; discussing it next week is a new proposal; for now, form the D17 action question without assuming that Version 2 exists." This does not isolate the past. It allows the past to enter the present under its correct identity.
Probability Describes Only the Uncertainty of an Event
An event probability can express the degree of an outcome within the current model. The number does not contain the actor's goal, nor does it state what failure would cost. The same 0.5 may describe a costless inquiry or a wager that exhausts all working capital.
If an action uses almost no resources and can be stopped at any time, a one-half chance may support trying it. If failure would prevent an existing commitment from being fulfilled, the same chance may be far too low. The meaning of 0.5 has not suddenly changed; the action question has acquired different consequences.
Conversely, a lower probability does not automatically mean taking no action. If a small action can obtain critical material and its cost is acceptable, an actor may reasonably try it. A higher probability does not automatically justify total commitment, especially when the minority failure case would cross the actor's survival line.
This does not make probability useless to action. Probability is one input to action judgment and helps assign weight to different outcomes. But that input must connect to goals, consequences, constraints, and time. With only a probability, the other places in the action function remain empty.
First Write Resources as Conditions of the Case
To make the discussion calculable, Gu Ning set clear boundaries for Chengwan's resources in the following week. The week contains ten half-day labor slots. Six have already been promised to existing jobs, while four remain available. A labor slot is a unit of measurement in the case; it does not imply that real work can always be divided neatly into half-days.
The station also has twelve units of working capital. At least eight must be retained for payments on existing jobs, basic operations, and possible rework. Only the remaining four can currently be discussed as deployable. These units are fictional normalized quantities and do not correspond to real currency amounts.
Four slots and four units are upper limits, not amounts that should be spent simply because they remain. They show which actions can enter the discussion and what a new commitment would displace. If someone proposes committing five units of working capital, the first issue is not whether the probability of success is high enough. The action already exceeds the current resource boundary.
These quantities are also not physical measurements of the margin in RC's ground of possibility. The ground retains possibilities that have not been locked in, but this does not mean that the repair station can draw unlimited funds from its accounts or create work bays from nothing. The available margin discussed in Part 2 must return to conditions that the subject can actually mobilize, preserve, or restore at the present time.
"Continue" Contains at Least Five Actions
Tang Ke unfolded the possible actions. The first was to make no further contact but preserve an entry point if the client returned. The second was to send one clarifying question about the desired scope, receipt period, and version for the following week, without reserving a work bay.
The third was to reserve one half-day slot before a reply arrived, but buy no materials. The fourth was to reserve the slot and prepare general-purpose materials. The fifth was to purchase specialized materials in the direction of the client's possible need, revise the existing schedule, and free two or more slots for a version that did not yet exist.
Every one of these actions might be called "continuing to follow up" in ordinary speech, but their depth of commitment increases step by step. The first adds almost no action. The second mainly spends communication time. The third begins to occupy an opportunity. The fourth and fifth bring money and scheduling into the commitment as well.
If Tang Ke supports the second action while Ye Cheng fears the fifth, then even when one says "continue" and the other says "do not continue," they are not disagreeing about the same action. Unfolding the verb before discussing action is not paperwork for its own sake. It prevents words of position from hiding the actual choice.
Inaction Also Requires an Accurate Description
Stopping is not a blank state either. The station can stop making inquiries while retaining the material; explicitly decline a next version; pause until complete terms arrive; or delete the request and refuse to reopen it. Different forms of stopping have different effects on future choice.
If Gu Ning chooses "occupy no work bay and wait for clear terms," she has neither committed to a next version nor rejected future cooperation. If she chooses "purchase immediately today," she changes how easily the station can later exit. Waiting and committing both shape later paths. Waiting cannot be recorded as if nothing happened.
Inaction may also have opportunity costs. A bay not reserved for R17 can accept another job; a clarifying message not sent may lose an earlier opportunity to obtain information. Later chapters will make opportunity cost explicit, but its existence does not make every instance of inaction equivalent to a loss.
Describing stopping precisely also prevents positions from being rewritten after the outcome. If the client returns on Monday, the record keeper cannot claim that the station had always decided to continue. If the client does not return, the keeper cannot claim that it had already issued a definite refusal. D17 must preserve the action that was actually taken, not merely the attitude that looks best later.
Goals Determine What Counts as Success
Tang Ke's main concern was not to miss a possible collaboration. Ye Cheng's was to keep the six existing slots unaffected. Gu Ning also wanted one limited contact to reveal what a next version would actually require. These goals are related, but they cannot all be summarized as "successfully winning R17."
Suppose the station sends one clarifying message and no new version is ultimately formed. The cooperation goal has failed, while the information goal may still be achieved. Suppose it reserves a bay in advance and ultimately secures the job. The income goal may be achieved while the stability goal is damaged because existing work is delayed.
One action can help one goal and harm another. If the order of priority among goals is not stated, any later outcome can be renamed a success: no cooperation saved resources, excessive commitment bought experience, and delay maintained the relationship.
The next chapter separates these goals in detail. For now, this chapter preserves a minimum agreement: D17 asks not only whether an agreement can be reached, but also which capacities the station wants to preserve. Only after goals enter the question does probability know which consequences it is weighting.
Consequences Belong to Combinations of Actions and Outcomes
The same outcome may have different consequences under different actions. If the client never forms Version 2 and the station has sent only one clarifying message, the additional loss is small. If the station has already bought specialized materials and delayed existing jobs, the absence of a version leaves a much larger loss.
Conversely, even if Version 2 is formed, formation does not mean it should be accepted when it requires more capacity than is available or its payment and delivery terms cross the station's limits. Event occurrence and action consequence cannot share a single affirmative label.
Gu Ning recorded consequences after each action, rather than only after possible future outcomes. The action determines what the station changes first; the external result determines which benefits or losses then materialize. Together they form the actual situation. Probability can assign degrees to uncertain external results, but it cannot erase exposure the station has chosen for itself.
This also explains why "the same probability" does not mean "the same risk." If risk is used to mean both the chance of occurrence and the loss after occurrence, it readily compresses two levels into one. This book will distinguish probability, consequence, exposure, and recovery conditions wherever possible, and will state the relations included when a collective term is needed.
An Independent Comparison Shows Why Choices Differ
Set the Chengwan case aside and consider two explicit thought experiments. A faces a choice whose desired outcome has a probability of one-half. Trying consumes only one unit that can readily be replenished; success yields a result useful to A's goal, and failure does not affect other commitments.
B faces the same one-half probability, but trying occupies all of B's available resources. Failure would prevent an already-promised basic task from continuing. Their probabilities are the same, and the outcome may even have the same name, but the consequences of action are different.
A may have reason to try and B may have reason to refuse. This comparison proves neither that they must choose this way nor an expected value for the gain and loss. It excludes only one hasty inference: that there is a single probability threshold above which every subject should take the same action.
B's choice may change if B gains more resources or a reliable alternative path after failure. If A's outcome is irrelevant to A's goal, a low cost need not justify trying. Reasons for action change with actual relations. They are not permanent personality labels for courage or caution.
Different Subjects Can Reasonably Choose Different Actions
When the same external probability is given to different subjects, their existing commitments, available resources, and capacity to bear failure may differ. A station with a spare bay may reserve half a day, while a station with only one bay may have to wait for clear terms.
This does not make the better-resourced station necessarily more correct or the less-resourced station deficient in judgment. A probability representation can be shared while the conditions for action remain situated in each subject. Requiring everyone to act at the same threshold often treats one subject's consequence structure as a universal condition.
Shared projects also raise questions of responsibility. A person who decides to reserve a bay but does not bear the delay to existing work may be more willing to commit. If the person who bears the rework did not participate in the judgment, the action plan has omitted where the loss falls. Later chapters connect assessment to responsibility; for now, this chapter places the subject's position on the card.
Gu Ning, Tang Ke, and Ye Cheng belong to the same repair station but have different observational responsibilities. Their views may all enter the discussion, yet the final action page must state who can allocate labor slots, who handles communication, and which existing commitments are affected. A majority in favor cannot substitute for alignment of authority and responsibility.
Time Changes the Meaning of the Same Action
Sending a clarifying question on Saturday and sending the same words on Wednesday near delivery may consume different remaining choices. If the station obtains terms early, it can still compare schedules. If it obtains them late, the information may be identical while fewer switching paths remain.
Reserving one labor slot also needs an expiry. If held indefinitely, it is no longer a small action but a continuing occupation. If it will be released unless terms are formed by noon on Monday, its depth of commitment is limited. Time is more than a date outside probability; it changes how long an action locks something in.
Waiting also has a deadline. Waiting until complete conditions appear can prevent unsupported commitment. Waiting until every other bay is occupied may eliminate the ability to accept the job. There is no unconditional rule that it is always better to wait for fuller information. The value of information must be considered together with the path changes produced by waiting.
D17 therefore needs a decision point. Gu Ning decided that Saturday would produce only the initial action card. Any decision to reserve a bay or buy materials must be reviewed before the existing schedule is confirmed on Monday. This is a working arrangement inside the case, not an industry convention.
A Probability Threshold Needs a Consequence Structure
People often say to act above seventy percent and decline below thirty percent. Such a rule can simplify a fixed task, but its threshold must be tied to an action, gains, losses, and constraints. When the threshold is carried into another task, the conditions that once made it useful may no longer exist.
Even within one task, sending a clarifying message and purchasing everything in advance should not automatically share a threshold. The main effect of the former may be to obtain information; the latter already assumes delivery exposure. When the action changes, the question of what probability makes it worthwhile changes too.
A threshold may also come from a hard constraint rather than average benefit. A nonzero probability of failure may be unacceptable when the consequence reaches a legal duty, a threat to life, or a subject's explicit survival line. The case in this book does not concern such high-stakes real-world domains, but it must still preserve the structural distinction: a hard boundary is not merely a more pessimistic probability.
Gu Ning did not write a universal threshold for D17. She first required every action to state what it would lock in, what it could affect in the worst case, and when it could be exited. Only when these relations are clear does numerical comparison have an object.
A Choice Can Be Provisionally Definite Without Pretending to Be Optimal
An actor cannot wait forever until every consequence has an exact probability. Monday's scheduling will arrive, and Chengwan must choose under limited material. Acknowledging limited information does not make every action equally reasonable. It requires the station to exclude actions that cross known boundaries and state the assumptions behind the remaining actions.
The four available labor slots and four deployable units of working capital provide a first limit. Version 2 does not yet exist, and its specialized materials and formal capacity requirements are unknown, so an early deep commitment lacks an object. Whether one clarifying message is worthwhile still requires comparison between its goal and additional cost.
This analysis can yield a provisional adoption without claiming a global optimum. If optimality requires listing every future path, a limited actor will often be unable to establish it. Sustainable decision-making first asks whether a plan preserves critical capacities, can receive feedback, and leaves a next step after failure.
"Provisional" cannot become a license for arbitrariness. An adopted action must record its boundary, period of validity, and reopening conditions. Otherwise every step can be called provisional while accumulating into a commitment that can no longer be exited.
How Prospective Focus Enters Action
RC describes prospective practice as a sequence of expectation, reality, and renewed expectation. In D17, expectation is not an idea suspended outside action. It participates in choosing what to send, what to preserve, and what not to lock in yet. These actions may then change the material and paths available later.
Accuracy of judgment and reasonableness of action are connected but distinct. A well-grounded probability cannot, by its quality alone, authorize an action whose consequences are unclear and whose commitment crosses resource boundaries. A favorable later outcome likewise cannot prove retrospectively that every depth of commitment was reasonable.
Whether feedback promotes calibration depends on whether the original action and its outcome can be revisited. If every success expands what "continue" supposedly meant and every failure shrinks it to "we only meant to ask," the action record cannot be constrained by consequences. When the actual action is clear, new material can identify which level needs to change.
Here positive and negative feedback are not used as terms from control engineering. The book speaks concretely of sequences that promote calibration and sequences that harden judgment. The former let differences change the plan; the latter absorb every result into the old narrative. The action page will preserve this distinction.
Advice Cannot Omit the Adviser's Position
If Gu Ning asks Tang Ke only, "Do you recommend continuing?" the answer will be shaped by the scope of Tang Ke's role. She handles communication, so sending a clarification is a small action she can perform. Reserving a bay and changing purchasing require others to bear the result. The loss visible within one person's responsibility may not cover the whole plan.
An adviser may state observations and preferences, but the formal action card must translate the advice into a concrete action and return it to the people and resources affected. Otherwise "I only suggested contacting them" may become "everyone agreed to schedule the job" as it is passed along, giving one opinion a depth of commitment that was never authorized.
This does not require a vote of the entire staff for every choice. Authority can be arranged beforehand, and low-impact actions can be performed by the responsible role. A wider check is required when an action crosses protected resources, alters existing commitments, or creates consequences that are hard to reverse. The authority boundary must be visible before action, rather than locating someone to blame after a bad result.
An outside adviser who supplies an excellent probability occupies a similar position. The adviser may help judge how likely an outcome is but may not know all of the subject's constraints. The person adopting the advice must add their own consequence structure; professional judgment does not arrive with action authority that was never granted.
Four Places Constitute an Action Question
Gu Ning condensed the discussion into four places. The first is the goal: what the station seeks to obtain and what it must preserve. The second is the action: which concrete changes are available now and how deeply each commits the station.
The third is consequence: when external results combine with the action, who gains or loses what, and whether recovery is possible. The fourth is deadline: when a decision is needed, how long the action lasts, and what new material invalidates the current choice.
Probability enters the consequence paths and expresses degrees for outcomes not yet known. Resource boundaries enter actions and consequences and limit which plans can be adopted. Subject and responsibility run through all four places, preventing one person from assessing while another silently bears the loss.
These four places do not yet make a complete decision model. Kinds of benefit, expected value, irreversible failure, and staged commitment will be developed in later chapters. For now they form a minimum framework that prevents "should we continue?" from being fought over by an old probability and two vague positions.
The D17 Action Question Card
At 9:40 on Saturday, Gu Ning saved D17's first action question card. Its object is whether and how to invest in discussion of a possible next version of R17, not to predict Version 1 again. There is currently no Version 2, no new reply, and no specialized process or delivery condition.
The card records five kinds of action, ranging from making no contact to deep early commitment. Current resources consist of four schedulable labor slots and four deployable units of working capital. Six slots of existing work and eight protected units may not be reallocated without explanation. Every number belongs to the fictional case.
The goals provisionally listed are obtaining clear conditions, pursuing suitable cooperation, fulfilling existing commitments, and preserving the capacity to adjust. Their priority has not yet been decided. The consequence field reserves places for commitment, occupation, missed opportunities, and recovery, without filling in probabilities or amounts in advance.
The decision deadline is provisionally set before Monday's schedule is confirmed. Any new reply, version formation, resource change, or disruption to an existing job triggers a review. The old 0.5 appears only as a historical report linked from Part 1, not as authorization for D17.
This chapter reaches the following judgment: the same probability can support different choices under different goals, actions, consequences, and deadlines. Probability has not lost its function; it has returned to its proper place. The next chapter separates goals and examines how income, fulfillment, learning, and preserved capacity can conflict, preventing action from selecting the most convenient definition of success only after the outcome.