Before the Outcome, the Story Can Already Take Shape
At 5 p.m. on Thursday, Tang Ke asked how R17 should be explained to other recorders in the future. From the morning's definition of the problem, through finding the historical table, to seeing the replies, the material was already considerable. She hoped to draft an easy-to-read retrospective account.
Gu Ning agreed to try drafting one first, but titled the sheet "Narrative Draft." R17 has still not received complete first-version confirmation; the outcome window does not close until 5 p.m. on Friday. The 5 p.m. on Thursday is not the endpoint of the original event; the people here do not know the future first.
This chapter continues the thought experiment of the fictional Chengwan. We examine several kinds of narratives that could go wrong and their correction methods, without assigning the main line a confirmed or failed ending. Even when the future outcome becomes clear, it cannot automatically determine what the people at 10 a.m. actually knew.
Chapter 2 preserved time points; this chapter examines why these time points may still disappear in a retrospective review. The original documents still exist, but this does not guarantee that the narrative continues to use them according to their original relationships. Text can avoid deleting any file yet quietly let a later understanding become an earlier judgment.
A Single Sentence Can Borrow Several Time Points
Tang Ke tried drafting one sentence: "We could tell from the start that this commission would likely confirm; the list-received reply also confirmed the judgment." Gu Ning did not judge it good or bad, but asked her to point to exactly which card "from the start" referred to.
At 10 a.m. the two established event and information boundaries without adopting a supported numerical probability. The statistics of the historical comparable eight items were not obtained until noon, and the original reply indicating list receipt with time yet to be verified was not handed to Gu Ning until 11:20 a.m. They cannot all enter the sheet dated 10 a.m.
"Saw" also requires correspondence. Was it a judgment at the time that the material was worth organizing, a predictive inclination toward the positive, or a belief that complete confirmation had already been received? Several meanings joined in the same verb make the story appear continuous, yet the original record does not provide an equally continuous conclusion.
Revising the draft to "In the morning we established the problem; later we obtained historical material and limited replies" sounds less confident but more accurately preserves how the judgment formed. It does not diminish the later work; it simply does not let that work claim its credit prematurely.
An Old Fact Already Existing at the Time May Also Have Become Available Only Later
The note on the station's schedulable reception period was written at 9:40 a.m. on Thursday, before 10 a.m. This gives the story another temptation: since the note already existed, one might say the two knew in the morning that there was no obstacle on the time front.
But Tang Ke did not find the note until 1:20 p.m., and only confirmed at 1:40 p.m. after rechecking that it came from the station's own schedule and stated only that the station could arrange alternative slots. The original reply still required the client to verify the reception time; the two statements have different subjects.
Being written down earlier does not mean entering Gu Ning's information earlier, nor does the original note support every use the story wishes to assign to it. If a retrospective review omits both points simultaneously, it will write a station condition obtained later as a bilateral acceptance known at the time.
A retrospective review may ask how to obtain schedule evidence earlier in the future, or ask why the note was vaguely written. But neither improvement changes the facts of when it was found and when it was verified on this occasion. How things should be organized in the future and how they were actually organized in the past occupy different positions.
Having Computed a Number Is Not the Same as Having Adopted It
Another draft might read: "We estimated a three-quarter chance and continued to proceed along that judgment throughout." Gu Ning asks for the provenance of that number. It is first the frequency of three out of four in the historical B group satisfying A.
Chapter 10 used the same historical fourfold table again, proceeding from the eight items with five confirmations through a conditional calculation on coarse B to arrive at three quarters. That was one algebraically consistent model scenario, not a second independent verification, and it was not directly adopted as the final probability for R17.
The full content of the original reply still contains reception time yet to be verified; the historical item-by-item order of A and B receipts still awaits rechecking. If the narrative strips away these limitations and leaves only a neat number, it rewrites a candidate calculation as a prediction that was already issued.
Where a number appears, when it was calculated, whether it was adopted, and what its applicable conditions are should all be traceable independently. Demonstrating that "three quarters" was once written does not prove that "three quarters" was used at the time to answer the question now under evaluation.
Without an Original Probability, Do Not Fabricate an Easily Evaluated Transcript Card
This means that R17's 10 a.m. card had no numerical prediction available for direct scoring. It contained an event definition and an information record at the time—a product of the judgment process, but not a scorecard that can be retrospectively filled in as seventy or eighty percent.
If the story needs an evaluation of the initial judgment, one can evaluate whether the question was clear, whether material was preserved, and which gaps were identified. To evaluate the predictive quality of the probability, one should wait for or seek the actual record that was issued, and must not let the need for evaluation create a number for the past.
The absence of a number does not automatically mean the work failed, nor does it automatically mean the work was correct. It only says that this evaluation does not have its required object. Therefore one cannot claim the morning guess was correct on the strength of a later positive result, nor deduct points from a probability that was never issued based on a later negative result.
The blank position in the record should preserve the meaning of that blank: it was not adopted at the time. If a decision is made later to adopt a number, a new issuance time and conditions must be recorded. This way it can later be seen whether a judgment was formed, or whether a previously missing field was simply filled in.
"I Always Felt That" Needs to Know What It Originally Meant
Tang Ke may indeed have had hope for this engagement. Such a feeling can be recalled honestly, but if no corresponding object, window, and inclination were preserved at the time, the range it can support in evaluation is limited.
Hoping for a deal does not mean predicting that the first version would receive confirmation within the original window. Feeling the other party's attitude was positive does not mean believing the reception period had been accepted. If one later uses narrower, more adjudicable events to fill content for vague feelings, the recollection has changed the original statement.
We need not declare that every unwritten feeling did not exist. The question is whether, in evaluating a specific prediction, one can verify that it actually answered the same question at the time. If it cannot be verified, this can be stated honestly; inability to verify must not be converted into a free choice of whichever favorable explanation one prefers.
Preserving the original wording is sometimes more useful than preserving a later, more precise summary. The original wording's vagueness is precisely what needs to be learned from this occasion; polishing it into something that appears always precise removes the basis for future improvement.
A Positive Ending Will Not Make a Low Probability an Inevitable Error
Leaving the current main line, consider another independent prediction: the recorder explicitly assigned a one-quarter probability to a binary event before the fact, and the event later held. If the retrospective review simply says "it happened, so the original judgment was wrong," it has converted a probabilistic judgment into a definite negation.
One quarter is not zero. It permits the event to hold, merely assigning it a lower probability. A single occurrence may yield a worse score of some probabilistic kind, and it may prompt examination of the model, but a single result cannot prove by itself that the probability was an unsupported judgment.
Conversely, giving three quarters and later having the event hold also cannot by itself prove that the judge accurately mastered this class of events. The result is compatible with a higher probability, but is insufficient to evaluate how similar predictions perform overall. On cross-event calibration and scoring, Part Four will develop this separately.
A story favors a clear win-or-lose word; a probabilistic record needs to preserve degree. A retrospective review can describe the relationship between the actual result and the original judgment, but must not, for the sake of easy storytelling, convert non-zero into never, or not-yet-one into certain.
A Negative Ending Also Will Not Make Every Earlier Doubt a Foresight
Consider another comparison case: a request was ultimately not confirmed, and the recorder's retrospective review lists every worry that existed at the time, then claims to have foreseen the failure. But raising worries and forming a negative inclination are still two things.
The original may have recorded both favorable and unfavorable material, and may have adopted a higher positive probability. If only the unfavorable sentences are later retained and the actually adopted judgment is not mentioned, the story can convert a negative check into a correct prediction, while the original prediction itself disappears from the review.
Examining risks was originally valuable and need not rely on being called foresight each time to acquire value. It may help fill material, arrange future reopening conditions, or may have been insufficient to change the judgment at the time. What needs to be returned to is which of these kinds of work it actually did.
Likewise, when the event later holds, excerpting only favorable sentences creates a corresponding achievement in the opposite direction. A retrospective review should preserve the version actually adopted and its basis, rather than selecting from a draft page the line that most resembles the final ending.
The Target Can Quietly Broaden Inside the Story
Suppose a certain independent object did not fully accept the first version within the original window but later achieved cooperation through modification. If the narrative says "it was negotiated in the end; the original prediction was fine," it may have substituted for the original-window event an unlimited-time, unlimited-version cooperation.
The later cooperation is a genuine result and worth separate recording. It cannot erase the original event's non-confirmation. Only when the original prediction was already answering that broader question can the corresponding result serve as its evaluation; the ending must not decide what the earliest sentence should have meant.
In Chengwan, complete confirmation, later payment, repair completion, and whether the commission was worth accepting have been kept separate. A retrospective review must not let the achievement photograph—easiest to narrate—become the evidence of confirmation within the original window. Success at different stages does not share all adjudications.
The title of the story also participates in the substitution. "A successful cooperation" can tell the full process; "the prediction of confirmation before Friday" must return to the original definition. The breadth or narrowness of the title is not just a stylistic matter; it determines which judgment the reader believes is under evaluation.
Finding the Ending in Personal Character Is Not Finding the Evidence at the Time
Another kind of narrative unfolds around the characters: the other party has always been cautious, so verification of the time came first; the station has always been meticulous, so it was ultimately able to advance. Such description may suit a retrospective review for which sufficient material has been obtained, but one must not infer stable character traits of both parties from R17's original reply alone.
The word "cautious" can accommodate acceptance after verification, as well as non-acceptance after verification. If both endings can be placed inside it unchanged, it does not by itself provide prediction support of clearly identified direction. The narrative explains the style of the sentence, not necessarily the difference in how the target occurred.
Character can be investigated as a candidate condition against broader material, showing how it was acquired and in which objects it applies. One must not deduce a personality backward from a single later result and then let that freshly acquired personality become an already-established earlier basis.
Character stories connect multiple time points in a way that is easy to understand; this is an expressive capacity. Judgment must ask whether the connection has corresponding material. Connecting naturally and connecting with support can hold simultaneously and can also come apart.
The Last Message Does Not Own All Causes
If a certain object in the future confirms after a reminder, the retrospective review will easily build its climax around that reminder. Temporal proximity to the result, concrete expression—readers easily see a continuation from action to ending.
But the comparison gaps of the last chapter do not disappear with the appearance of a climax. Preparation, earlier intention, conditions that changed simultaneously, and the possible outcome under the other process still need to be explained. A message near the ending may participate in the process, or it may simply be the node most conveniently excerpted for the story.
A retrospective review need not distribute every message's role equally. It should distinguish connections already obtained from effects not yet identified, and preserve competing explanations. If process material supports a particular step, it can state specifically where the support lies, rather than simply replacing a temporal relation with a causal verb.
Tang Ke's proposal to attach the sentence has not yet been implemented. Any current summary that writes it as "we improved communication and finally facilitated confirmation" has filled in both the action and the ending. A well-read sentence cannot obtain facts for two positions that have not yet occurred.
Cutting Out Detours Deletes the Conditions of Judgment
A retrospective review usually cannot reproduce the entire record. Gu Ning permits cutting details but asks to distinguish what is mere expressive redundancy from what was a condition under which the original judgment held or failed to hold.
Beyond the historical eight items, there are four items whose original records are obtainable but whose stage or window does not correspond, and four items whose original records are missing. If the story only writes "we checked sixteen items, therefore we have a reliable basis," the reduction has altered the sample scope and no longer merely omits numbers.
The 11:20 a.m. reply was handed over and also read by Ye Cheng, but this does not turn it into several independent client messages. If a retrospective review writes each person's account as a new confirmation, it accumulates rereading and cross-checking into an evidentiary increment that does not exist.
It may be condensed to "one original reply, after cross-checking, still states only limited content," without retelling the entire provenance. But the quantity and use of the originals should be preserved. The aim of compression is to make relations more visible, not to make it appear that judgment never encountered a gap.
Process Records and Display Narratives Have Different Tasks
An achievement display can center on completed objects; readers may simply want to know how things changed before and after the repair. Predictive evaluation, however, must return to the entrance range and to the positions of non-held, unknown, and not-yet-concluded. One passage of display text cannot be required to bear all tasks alone.
Narration is therefore not the enemy of the original record. It can help others find key information, provided it states what range it has selected and preserves the relation back to the original cards. An achievement story still has legitimate uses; there is no need to pretend it is the entire predictive ledger.
The danger arises when the display narrative is taken to demonstrate judgment ability: visible successful objects substitute for all objects, and later explanations substitute for the ex-ante version. At that point, the scope of evaluation should be restored, not the story criticized for being insufficiently pessimistic.
A retrospective review may also tell the story of a non-held object, but must equally avoid letting it substitute for the whole. Choosing success or choosing failure may both have specific narrative purposes, and a stated purpose is not the same as representativeness having been achieved. The scope of judgment requires separate preservation.
When the Original Card Is Unclear, the Retrospective Review Admits the Unclearness First
If an earlier record lacks explicit version or time points, complete recovery may not be possible later. Sometimes the original message was preserved, but one does not know whether it had already been adopted at the time; sometimes an inclination was recorded, but there is no corresponding window.
This kind of gap can be tried against other material, or it can be left in an unverified state. Because a retrospective review hopes to reach a conclusion, it must not write the restoration plan that best fits the ending as the definite original intent. Insufficient material is not a license for arbitrary reconstruction.
The conclusion of a retrospective review can be: "This occasion cannot accurately evaluate this original prediction; the window should be preserved at the time of issuance in the future." This has one less achievement than a beautiful success or failure story, but gains one improvement that genuinely corresponds to a gap.
Likewise, preserving the original card does not mean every field in it is necessarily accurate. When a later discovery shows an error, it should be corrected with the basis stated; but correcting a record error and changing a judgment later are separate things. Being traceable does not mean never corrected.
Accountability Discussion Must Not Borrow Only One Ending
If a later outcome is unfavorable, people may ask why Gu Ning did not see it at the time. A retrospective review needs to check what she could obtain then, what she actually adopted, and which inquiries had reason to proceed, rather than treating every detail now known as an omission in the past.
This does not mean that limited conditions at the time exempt all checking. If the original card clearly omitted material that was already available and directly related to the target at the time, that is a verifiable problem. If material was obtained only later, one must ask separately whether there was a reasonable earlier acquisition path.
The two questions are different: failing to use available material, and later discovering that certain material was important. The first can be checked against the original record; the second requires discussion of the search scope and basis at the time, and cannot be completed by the mere fact that it is now known to be important.
A favorable ending also cannot cover process deficiencies. If a result happens to hold, duplicated counting of original records and version confusion still warrant correction. The work of a retrospective review is to examine how the judgment formed, not to let the ending sign off on every step.
New Explanations Can Become Future Assumptions
Material obtained later may genuinely change our understanding of the process. For example, in a clearly stated comparison case, restoring communication records may reveal that a certain acceptance condition was never explicit. Such a discovery can drive the next batch of judgments to reselect their inputs.
But when a new explanation enters the future, it must still state how to identify similar conditions, when it will be available, and how it will change the judgment. If it can only be identified after the outcome is known, it cannot, without transformation, be called a reusable ex-ante clue.
A retrospective review therefore has two layers of product: where this record needs correction, and where the future model needs adjustment. The first is grounded in this occasion's material; the second must also accept examination by subsequent material. The two layers must be separated so that learning does not become the reinvention, after each ending, of a past that was always correct.
One may also preserve a new explanation as a candidate without rushing to adopt it. Obtaining a question that can be asked is already progress; immediately promoting every candidate to a rule lets the model be led by a few appealing stories.
Do Not Write Possible Stories into the Existing Main Line
Tang Ke's trial retrospective can simultaneously preserve several possible continuations: complete confirmation within the original window, no confirmation within the original window, and material needing supplementary checks. These are written to examine what original information should still be preserved unchanged under different endings, not to predict that each ending is equally probable.
Under every continuation, the facts that there was no numerical prediction at 10 a.m., that historical statistics were obtained only at 11 a.m., and that the original reply still has content awaiting verification do not change with the ending. If a sentence is considered something that was only explicitly said under a favorable ending, one must return to the material and recheck.
Possible continuations also cannot serve as outcome adjudications. Gu Ning places them on the comparison page, not in R17's result column. Otherwise a record analyzed under the cover of a story will first use the story it tells to overwrite the main line during the analysis.
The boundary of the thought experiment must therefore also be visible. It can help identify how theft occurs, but it cannot obtain new genuine messages for Chengwan. The clarity of the example depends on its position being kept separate from current facts.
The Retrospective-Review Table Lets the Narrative Find the Original Cards Again
Gu Ning split the narrative draft into several question-worthy sentences. Beside each she left four positions: what judgment it states, where the original card is, when the basis entered, and what the later outcome or explanation has added.
"There was a three-quarter chance from the start" cannot find a corresponding adoption record; "the station's time was available" can find the note, but its subject and verification time must be filled in; "multiple staff confirmed" can return to the same original message and must not be written as independent evidence.
One then checks whether each sentence's target is maintained. If the original card answers the first version's original window but the narrative answers eventual cooperation, that is marked as another question. A separate story may be written, but the result of the new question must not be placed in the achievement slot of the old question.
This table does not require publicizing all private discussion, nor does it require the text to become a field checklist. It offers the writer a way back: the narrative can be read through, and every important judgment can also be checked in isolation. Narration and checking can thus work together.
Preserving the Unkemptness of Process
Under RC's position of finite representation, a retrospective narrative is also one act of organization. It excerpts certain relations, making the process easier to express, but may let omitted unknowns and divergences lose their place. A clear story is not thereby invalid, but its clarity is not proof that reality has been exhausted.
The original card preserves the unkemptness of the moment: definition explicit but probability not yet adopted, message obtained but time still unverified, model calculable but applicability still limited. These are not flaws that writing must fix, but the actual positions that judgment occupied at different stages.
If feedback leads us to reexamine these positions, it can promote calibration; if it only helps select the sentences that fit the smoothest ending, it may further rigidify existing explanations. A reliable retrospective review should allow the original judgment to be negated by specific material, while also allowing its limited range to continue to be preserved.
The past does not need to be written as always confident in order for the future to have room for improvement. Acknowledging what was unknown at the time and what was added later does not cancel the organization already completed; it makes the formation, adoption, and reopening of judgment processes that can each be checked separately.
What 5:20 p.m. Preserves Is the Narrative-to-Card Comparison
At 5:20 p.m., Gu Ning preserved the comparison between the narrative and the original cards. Tang Ke's trial draft remains marked as a draft; the retrospective review is revised to state material according to the time it was acquired, without claiming that three quarters was adopted in the morning, and without writing the station's schedule as client acceptance.
The historical comparable eight items and B grouping, the limited use of the original reply, and the various gaps continue to be preserved. R17 has no new confirmation; the window still runs to 5 p.m. on Friday; Thursday's preservation does not adjudicate the ending. Any later explanation entering the record should note its basis and the level of its effect separately.
What this chapter delivers is a retrospective review that can return to the original predictive position. The next chapter will examine a broader form of overwriting: when many people begin to use the same phrasing, how it affects material that enters later, and who will still ask different questions.