When Was the Note Written?
Gu Ning had saved the first-version event card for request R17: by Friday 17:00, whether the repair station would receive the client's explicit confirmation of the specified task scope, payment conditions, and receipt period. At Thursday 10:00, that confirmation had still not arrived. Tang Ke said she was willing to ask once more later, to see whether the other party had any new intention.
This continuing scenario remains the fictional Chengwan. Chapter 1 settled what the event is; this chapter neither adds real samples in advance nor announces that the commission has been formed. We first look at how a prediction enters time, so that future messages can change the judgment while not silently changing the information the original judgment possessed.
Tang Ke asked: if a clearer reply arrived in the afternoon, could Gu Ning adjust the note? Gu Ning said of course; she wanted the judgment to follow the material. Tang Ke then asked: after the adjustment, when looking back that evening, could the note still represent the view held at 10:00 in the morning? Two legitimate needs had here come into conflict.
If revision is forbidden, the record blocks learning; if overwriting is allowed at any moment, the record can no longer say what was originally known. The problem is not whether to update, but how to preserve both the earlier and later judgments after updating. Time records give each its own place, without sacrificing one capacity to protect the other.
The Range of Information Available at the Time
As of 10:00, Gu Ning had the event card, the current task description, and Tang Ke's record of the previous day's communication. She knew the other party had inquired but had obtained no complete confirmation, and had not yet compiled the historical outcomes of a class of similar requests. This was the range of material at that moment, not a guarantee about future outcomes.
Available at the time does not mean all the material was reliable. Tang Ke's record might omit qualifiers; Gu Ning might misread a positive attitude. Preserving the information range is for knowing later what the original judgment was formed from; whether the material quality was sufficient remains to be examined in later chapters.
Likewise, that some message had already been generated elsewhere does not mean it was already able to support Gu Ning's judgment at this moment. The client might be deliberating; another assistant might already have heard something; but Gu Ning had not yet obtained it. Content that can only be found later must not be written into the morning's evidence list.
This book calls this range the information boundary of the prediction time point. It includes the material already obtained and usable for this judgment, and it also states what has not yet been obtained. The boundary is not the full map of knowledge, but the limited record that lets a judgment return to the conditions of its formation.
Probability values must later be formed on these conditions. If the same number rested first on vague inquiry and then on full acceptance, it has already been in different states. One cannot, merely because the number has not changed, hold that the judgment has had the same support throughout.
The Window's End and the Prediction's Start Are Not the Same Time
The event card's Friday 17:00 is the window's end for this outcome; Thursday 10:00 is the time point at which Gu Ning prepared to form her judgment. One states when the outcome is checked; the other states under what material the judgment was issued. Missing either invites later misreading.
For the same prediction of confirmation before Friday 17:00, a Thursday-morning judgment and a Friday-afternoon judgment face different remaining time and may possess different messages. Identical question endpoints do not mean identical prediction difficulty or information state.
As the deadline nears, Gu Ning has no reason to be automatically more confident of an affirmative outcome. If no reply ever arrives, the shrinking remaining time may itself become another piece of material; if the other party has said it will confirm at once, that too may change the judgment. The specific direction depends on conditions; one cannot infer from the mere passage of time that the probability must rise or fall.
Comparing two people's predictions likewise requires matching their issue times. If Tang Ke judged at the last moment after seeing a new message and Gu Ning judged the day before, Tang Ke guessing right alone does not show her original ability was necessarily stronger. Comparison should first ask what each faced at the time.
Time is therefore not an ornament beside the record. It enters the judgment's scope of application, letting representations of the same event at different observational positions be preserved separately. Without this distinction, it becomes easy later to put predictions made before knowing the outcome together with statements made after knowing it on a single scorecard.
Message Sent, Message Arrived, and Message Seen
To test the boundary, first set a comparison case: the client sends an acceptance message at Friday 16:55; the message arrives on the agreed channel at 16:58; Gu Ning views it only at 17:10. The three times are close, yet not the same.
Chapter 1's event concerned whether the repair station received explicit confirmation. Therefore, what must be checked in this comparison is whether the time and content of arrival on the agreed channel satisfy the condition — not when Gu Ning opened the record. If returnable material for the arrival time exists, a later viewing does not automatically make the event occur outside the window.
Set another comparison: the client says it sent the message before 17:00, yet the message arrives only after 17:00. Under the current event definition based on receipt, the sending statement cannot directly substitute for arrival within the window. If one asked instead whether the message was sent, that would be another event, requiring its own conditions and evidence.
This discussion does not address the reliability or legal force of any real communication system. Arrival time is material to be identified within the thought experiment. If the records cannot distinguish sending from arrival, Gu Ning should retain the uncertainty rather than supply a favorable time because the message content matches her hopes.
This set of comparisons shows that an outcome's occurrence and the observer's discovery of it can come apart. A record must state which layer it adjudicates. The time displayed by a system, the time remembered by a relayer, and the time claimed by a party may also not be the same kind of material; they cannot be interchanged merely because all resemble a time.
A Late Observation Does Not Let the Prediction Acquire the Outcome Earlier
In the first comparison above, the event can hold within the window while Gu Ning learns of it only after the window. Her morning judgment could still rest only on morning material. The later confirmation that the original message indeed arrived in time cannot be written down as her having already obtained acceptance evidence at the time.
Preserving the original prediction and appending the outcome can both be accurate. The original judgment records its issue time and material; the outcome record records the arrival time, the viewing time, and the basis of the check. There is no need to force all records onto a single moment for the sake of temporal tidiness.
If the arrival time is later discovered to have been misrecorded, the outcome evidence can be corrected, with the correction's provenance stated. Correcting the outcome does not reissue the original prediction, nor does it mean the morning judgment thereby acquired more foreknowledge. Material updates have different roles at different layers.
This distinction especially protects a state that easily disappears: not knowing then, confirming later. It can provide valuable new material, but it must not be repackaged as having known all along. Judgment reliability must learn from the former; it cannot be obtained by renaming the former as the latter.
Conversely, Gu Ning's viewing it late also does not automatically prove the event did not occur within the window. If the observer's personal opening of the record replaces the original receipt condition, the event has had its object swapped. Late knowing and late arrival must be checked separately; one may not choose whichever explanation is most convenient.
Why Checking May Still Be Needed After the Deadline
When the window ends, Gu Ning does not necessarily immediately possess sufficient material. Messages may still await retrieval, time records may be disputed, and Tang Ke's summary may not have preserved the original text. These situations mean the outcome adjudication must continue to be checked; one cannot, merely because 17:00 has arrived, conjure a definite answer out of nothing.
The cutoff point limits the time range within which the event may hold; the adjudication time states when people completed their judgment about that range. A later adjudication can examine an earlier range without thereby extending it. The two times connect but cannot overwrite each other.
If the records completely show that no qualifying confirmation was received within the window, the event can be judged not to have held in that window. If material is missing, it can only be provisionally recorded as pending check. One cannot substitute one's own failure to see for having examined that range and found nothing, nor indefinitely use "pending" to evade material that was actually sufficient.
A pending check should state what is missing, who is to look for it, and what material might change the adjudication. It need not guarantee that all of the past will eventually be settled, but it must give the unknown an identifiable boundary. Otherwise the unknown may turn from an honest state into a hiding place with no outcome ever.
When evaluating predictions later, one must also keep track of which objects have been adjudicated and which remain under check. Deleting all pending entries may make the record tidier, but it changes the set of objects entering evaluation. Questions about samples and unobserved outcomes will continue in later chapters.
How to Update When New Material Arrives
Suppose on Thursday afternoon Tang Ke obtains a clear new message: the client wishes to verify the receipt period and has not yet accepted the whole first version. This comparison does not mean the main line has received a reply. It lets Gu Ning know that new material can enter the judgment without directly counting as full confirmation.
If Gu Ning adjusts her judgment on its basis, she should record a separate afternoon version, retaining the morning version. The new record states what was obtained, which condition changed, and whether the current event is still the same. Updating then genuinely follows the material, instead of merely erasing "very likely" on the note into "not so likely."
If the new material changes only her understanding of the event's tendency while the event definition is unchanged, different prediction versions of the same object can be preserved. If the client alters the task scope, a new event must be set up or the version relation explained; the altered object cannot be called a mere probability update.
An update may also preserve the original judgment. If the new material does not address the key gap, Gu Ning's temporarily not changing her estimate is a stateable choice; one cannot require that every incoming message move the number. The magnitude of change should have a basis and is not decided automatically by message count.
This book will later explain how evidence supports updating. This chapter first handles time discipline: judgments may be revised, prior states are not deleted, and new material's time of acquisition is not backdated. Without this discipline, even the most sophisticated updating method can lose reviewability in hindsight.
Repeated Predictions Do Not Mean More Outcomes Occurred
Gu Ning predicts once in the morning, once in the afternoon, and once the next day, perhaps leaving three records. They can show how the judgment changed, yet they concern the same event within the same window; writing three times did not produce three independent confirmation outcomes.
If after the fact only the correct one is kept, the other states are erased; if all three are treated as the scores of three different events, one outcome repeatedly enters the statistics. Whether they may be evaluated jointly requires a separate statement of object and method; one cannot conclude there are many samples merely from many rows.
The same issue arises with multiple predictors. Many people judging the same request supply representations from multiple observers, not multiple actual requests. Comparing them can be meaningful, but headcount cannot be treated directly as the number of outcome occurrences.
Therefore the record must at least identify the relations among event identifier, prediction version, and outcome identifier. Multi-time-point judgments of the same event can be chained together; outcomes across events are stored separately. With the structure clear, one later knows whether one is comparing updating ability, different observers, or the frequency of a class of events.
This does not demand an immediate complex database. Simply writing clearly which event a given prediction targets, and which record the original prediction was, already removes basic confusions. The more complex the tool, the more it needs these relations; more fields cannot substitute for identification of objects.
How a Moving Window Produces Another Question
Gu Ning could instead ask each day whether a confirmation will arrive within the next day. This has its own use, yet it differs from the fixed window always ending at Friday 17:00. Asking on Thursday about the next day, then asking on Friday about the next day: the two questions already cover different time segments.
If a new window extends to the weekend, a confirmation occurring over the weekend can answer the new question but not the original Friday-deadline question. Without this distinction, every non-occurrence allows the window to slide forward until one occurrence, upon which the original judgment is declared correct.
Windows may also overlap. The same confirmation may fall within two different ranges at once; the two outcomes do not thereby have fully independent origins. Overlaps can be preserved; whether merging them for comparison is appropriate still depends on identifying the use.
Neither the fixed nor the moving window is naturally superior. The former makes it easy to return to a definite range; the latter can help attend continuously to new time segments. The key is to state clearly which one is adopted, and not to choose the more favorable range after outcomes arrive.
If the original window has ended, that question can be accurately closed and a new window established. Closing does not mean never observing the commission again; it gives continued observation a new starting point, without maintaining attention through indefinitely extending an old judgment.
Do Not Let After-the-Fact Explanations Alter the Raw Material
Suppose the outcome ultimately did not hold within the window, and Tang Ke later explains that the other party was obviously still hesitating. This explanation may have material, or may be a story formed after the outcome appeared. It needs to be kept separate from the morning record, not to overwrite the positive inquiry that was recorded.
Knowing things afterward is useful; it can help refine future observational focus. It does not automatically prove that Gu Ning should have possessed the same information at the time, still less that she then held the opposite judgment. Evaluating the original prediction means returning to the material obtainable then and the conditions already adopted.
Likewise, once an outcome holds, people easily say they knew it all along. If the morning said only "there is a chance" and the evening recollection becomes "it would certainly be confirmed," the confidence level has been rewritten by the outcome. The original record lets such rewriting become visible without arguing each time over whose memory is more trustworthy.
This does not turn record-keeping into a trial of character. People may sincerely reorganize experience in later explanations without intending to fabricate. Processual completeness requires returnable material precisely so that learning does not depend only on the recollection most favorable to oneself.
Later chapters on narrative and conviction will continue to analyze how stories acquire authority. This chapter retains the most basic temporal distinctions: what was seen then, what was judged then, the later outcome, and the later explanation each have their own provenance and cannot borrow one another's issue time.
When Actions Change the Process, the State Must Also Be Preserved
Tang Ke's preparing to ask the client again is an action. The inquiry may bring new information, and it may also participate in the client's later confirmation process. If Gu Ning's prediction was formed before this action was decided, later evaluation needs to know whether the action occurred.
This is not used here to prove the inquiry caused the confirmation, still less to supply a communication success rate. It only points out that arrangements during the observation period may change. If the record treats all outcomes as occurring under entirely unchanged conditions, it later becomes hard to know what the prediction actually depended on.
One can preserve whether a new description was proposed, whether receipt arrangements changed, and whether additional confirmation was obtained, without immediately computing the magnitude of their effects. What happened and what it caused are different questions. Preserving the former first gives later interpretation a chance to check the latter.
Predictions are sometimes explicitly conditioned on some action occurring, and sometimes formed only under the existing state. The two should be kept apart. One cannot, because no statement was made originally and the action later obtained a favorable result, write it in as the full plan already contained in the original judgment.
This volume still does not decide for the repair station whether to press for a reply or reserve equipment. Action configuration is left to the next volume. Preserving action state here protects the before-and-after conditions of the judgment's object, so that the event's time record does not pretend the process was wholly untouched by participants.
What Has Already Happened and What Is Not Yet Known
There is one further boundary to preserve. Suppose that at Friday 16:00 a complete confirmation has already arrived and been viewed by Gu Ning, and she then writes "a confirmation will be received before 17:00." Under the original event's condition of receipt, the establishing material is already available; the affirmative statement is no longer in the same information state as Thursday morning's prediction.
If this statement is placed on the original prediction's scorecard, it may fabricate a remarkably accurate record. It does not show how one judges before obtaining the outcome; it writes knowledge already obtained into a future-facing sentence form. Keeping the word "will" in the sentence does not turn known material back into unknown.
Nor does the original event revert to "not yet held" because a cancellation might occur after 17:00. It concerns whether confirmation was received within the window, not how long the cooperation would last. If what is truly of interest is "not yet canceled at the deadline," a separate condition must be set; one cannot swap it in after seeing subsequent changes.
Another case: a confirmation may already have arrived while Gu Ning has not yet viewed it. She then has an informational unknown, yet is not necessarily facing an object that has not yet occurred. To express a judgment here, she should state whether she is assessing whether the corresponding material already exists, or whether a new confirmation will still arrive within the remaining window. The two questions point in different temporal directions.
These distinctions do not require that one may judge only before all outcomes occur. Assessing unknown past facts can also be meaningful, and this book's later evidence methods may still apply to it. But when comparing predictive ability, one must know at what temporal position a judgment was formed; different tasks cannot be mingled without explanation.
The time card likewise needs to preserve whether, at the time the prediction was formed, outcome material sufficient for confirmation of the original event already existed. Outcome unknown, outcome not yet occurred, and outcome already obtained cannot be treated as the same state merely by a future-tense sentence. A reliable record does not pursue calling every row a prediction; it lets every row know what it is actually answering.
Time Records Also Require Limited Cost
Saving every detail of every message could swallow the time Tang Ke has for communication. If Gu Ning insists on infinite traces, the record itself may alter the work process, degrading the very judgment it was meant to help.
One can first preserve the key time points: the original judgment's formation, the arrival of new material sufficient to change the object or evidence state, the window's end, and the completion of the outcome check. Irrelevant routine content need not all be collected, and material involving private arrangements should be limited according to use.
A record as coarse as morning and afternoon is sometimes already sufficient; if the outcome sits near the 17:00 boundary, finer material is needed. Precision should connect to the question at hand; one need not write unverifiable seconds onto every step for the sake of looking professional.
Unverifiable fine-grained times may instead manufacture false certainty. If Tang Ke remembers only "early evening," she should not be required to fill in 16:58 from memory. A rough range can be kept, with a statement of why it does not suffice for this adjudication.
Therefore an adequate time record is not the most exhaustive record. It makes the important relations identifiable, lets insufficiency exist honestly, and leaves later steps a way to obtain the necessary material. Limited representation accepts updating without needing to masquerade as a flawless copy of the entire process.
Keeping Before-and-After Positions for the Same Question
Gu Ning added a time note beside the event card. The original question's prediction start is Thursday 10:00, and its deadline range runs to Friday 17:00; the information list contains only what has currently been obtained. If new material arrives, its acquisition time and the new judgment version are recorded separately, without overwriting the original card.
The outcome record separately lists the time of arrival on the agreed channel, Gu Ning's viewing time, and the time the adjudication was completed. If material for any of the three is missing, the gap is stated; the message's sending time serves only as auxiliary evidence when relevant to the current event and checkable, never as the answer to the receipt condition.
These fields protect different relations; they do not force every prediction to fill in the same amount of form. When the object is simple and the material direct, abbreviation is fine; supplement when boundaries are disputed. The user should know what each time point answers; otherwise even complete fields may continue to confuse.
The main line here still assigns R17 no early confirmation outcome and gives no sample-unsupported probability. What Gu Ning completed is an original record capable of admitting future messages. The timely arrival, late viewing, and window shifts in the examples are comparison branches and are not written into the commission process as having occurred.
The judgment obtained in this chapter is that prediction requires preserving both the event window and the position of the information at the time. Updating can occur while the original judgment remains returnable; outcomes can be checked later, but later material cannot rewrite what was known earlier. The next chapter turns to historical records: Gu Ning wants to find similar requests for comparison, and must first know which objects entered the sample, and where those records actually came from.