The First Rehearsal Selects the Correct File
At 09:00 Wednesday, V3 is archived. Gu Ning creates internal rehearsal T22 to test the handoff-file entrance: Tang Ke enters the current index from the object workspace, and Ye Cheng checks the object, version, entry page, and item list.
The rehearsal sends nothing externally, occupies no production slot, and spends no working-capital unit. Success requires all four fields to match within ten minutes.
At 09:07 Tang Ke provides the correct V3 index. All four match in seven minutes, so the result is successful.
That line alone makes the process look reliable and Tuesday like a chance error. But the screen record shows Tang Ke still opened the first item in recent files.
A Good Result Answers Only What Happened
T22’s first result is truly good: the right file was found, its content was complete, and time met the standard. Criticism of the process cannot turn success into failure.
The result answers whether this attempt met its objective. Judgment quality additionally asks what rule was used, what material supported it, and whether it still serves the objective as conditions change.
The first recent file happened to be the V3 index repaired at 16:35 Tuesday. Rule and result aligned once, but the rule did not identify the current object.
Preserving the good result makes the problem precise: the result is genuine, but it does not supply the process evidence attributed to it.
The Object of Judgment Precedes the Label
If the claim is “the file retrieved this time is V3,” verification supports it. If the claim is “the first recent file reliably represents the current version,” one success is far too little.
The broader claim “the handoff process is repaired” additionally requires a unique entrance, interruption recovery, joint checking, and isolation of old templates. The first rehearsal did not test all these relations.
The same result provides different strengths of material for different claims. The team cannot jump from the narrowest result to the broadest conclusion.
Gu Ning writes the claim before connecting evidence, preventing the word “success” from expanding on its own.
Why the Recent-Files Rule Is Poor
Recent files are ordered by opening time, not object validity. Viewing an old template, handling another project, or recovering from interruption can change the first item.
The rule checks no object identifier, version state, or register link. It depends on an interface position only accidentally related to the objective.
Tuesday already provided a counterexample: after task switching, the first item pointed to V2. On Wednesday it points to V3 only because the repaired file was opened last.
An existing counterexample makes the general rule “the first item is current” poor even when this outcome succeeds.
Chance Is Not a Mysterious Force
Calling the result lucky does not mean it lacks causes. V3 occupied the first position because it was the last file opened during Tuesday’s archive repair.
Chance means that this cause was neither controlled by the rule nor the stable relation the rule claimed to use for identifying the object. A slight change in list order could change the result.
Analysis should identify how the correct file appeared and whether that path belonged to the plan. When an unplanned favorable condition generates success, the judgment cannot claim all the credit.
“Good luck” is as empty as any other label unless unfolded into conditional relations.
Success Does Not Wash Away Process Deviation
T22 requires entry through the object workspace’s unique link. Tang Ke did not follow it. The page must record both “objective succeeded” and “entrance step deviated.”
Recording deviation only after failure creates a selective process: the same action is flexible when it works and noncompliant when it fails. The rule then cannot be tested.
Deviation under success matters especially because it rewards an unreliable shortcut. Executors learn “this is fine” until conditions change.
Gu Ning retains the success label but does not check process compliance.
Observation Quality Makes Poor Judgment Visible
Had the rehearsal recorded only file contents, the result would be entirely green. Recording the screen path reveals the recent-files behavior.
Observation increases knowledge of the process without changing the file. Separating a good result from poor judgment depends on preserving intermediate actions.
Observation also has a cost. This rehearsal records only entrance, file identifier, checks, and duration, not unrelated personal activity.
Observation design should serve a specific claim rather than become unlimited surveillance.
Reward After Success Can Entrench a Bad Rule
Rewarding only the seven-minute completion makes recent files appear faster than the unique link. The next executor may repeat the shortcut voluntarily.
A short-term efficiency measure then turns accidental success into a long-term path. The error reappears when an old file returns to the first position.
Evaluation must include both target result and critical process. Seven minutes and entrance deviation are both valid observations; neither cancels the other.
Rules can change, but a change requires a new version and validation, not automatic revision by one fast result.
The Second Rehearsal Changes Only List Order
At 09:20 Gu Ning creates T22-B. Without changing file contents, she has Tang Ke open and close the V2 archive note, putting V2 first in recent files.
Tang Ke does not know which result the test targets, but knows she should still use the object-workspace link. At 09:23 she again opens the first recent file; Ye Cheng finds V2 during checking.
The second result fails. The shortcut is identical; only interface order differs. Together, the attempts show direct dependence on recency order.
This is not a randomized experiment and yields no failure rate. It is a counterexample test of a claimed stable relation.
What the Counterexample Reveals
T22-B proves that a small order change makes the recent-files rule select the wrong object. The rule cannot guarantee current-object identification.
It does not prove that the unique-link plan can never fail or estimate how often V2 is first in real work. Rehearsal conditions also differ from production pressure.
The counterexample defeats an absolute reliability claim without providing a complete performance model. Gu Ning keeps the conclusion within that scope.
One strong counterexample differs from many repetitions without a defined object.
One Win and One Loss Cannot Simply Be Averaged
Reporting “50% success” from one successful and one failed rehearsal would create false precision. The team deliberately changed list order rather than sampling from a defined set of opportunities.
The informative relation is conditional: the shortcut succeeds when V3 is first and fails when V2 is first. Sorting state determines the outcome.
A future frequency would require defined real handoff opportunities, an observation period, recorded file positions, and a consistent execution rule. Those do not yet exist.
Classification precedes proportion, continuing Part I’s treatment of denominators without repeating its scoring model.
A Good Result Cannot Prove the Plan Sufficient
A plan contains many relations. The first rehearsal traverses one actual path but does not cover task interruption, an old template being opened, link failure, or multi-person handoff.
Success establishes that at least one path reached the objective, not that every important path is controlled. If the plan claims correctness when list order changes, that condition must be tested.
V3’s incident already supplies failure under a changed list; T22-A cannot overwrite it with one success. New and old material must be read together.
Plan sufficiency must face relevant failure conditions rather than accumulate undifferentiated successes.
Poor Judgment Can Encounter a Favorable State
Imagine a separate case in which an actor ignores versions and picks randomly from two similar files, choosing the correct one. The result succeeds, but the rule used no object information.
With well-defined repeated opportunities, long-run performance might be estimated. A single success says only that the favorable item was selected. D17 contains no claim of random choice, so Tang Ke’s act is not assigned one-half.
The example demonstrates logical possibility, not a probability for the main case. Recent-file sorting is not random.
Uncertainty cannot become an arbitrary number merely because two options resemble fifty-fifty.
Later Rescue Can Hide Poor Judgment
Suppose the wrong file is sent, the other party detects it and asks for a replacement, and the team corrects it within the window. Final success would include the other party’s rescue rather than the original judgment succeeding alone.
No such rescue occurred in V3. The independent example reminds the team to record who rescued the process, when, and at what cost.
Looking only at the endpoint can give the same success label to a bad entrance followed by rescue and a smooth execution. Process pages distinguish these paths.
Recovery capability has value but does not justify leaving the original error unrepaired.
Result Attribution Lists Necessary Relations
T22-A required at least the correct file to be first, Tang Ke to select it, complete file contents, and Ye Cheng to check four fields. Changing any relation might change the outcome.
The correct first position came from initial test state, not the recent-files rule’s identification ability. Complete content came from Tuesday’s archive repair; verification came from rehearsal design.
Once contributions are separated, success can be attributed to favorable initial state, existing repair, and checking rather than wholly to the shortcut.
Attribution maps operative relations; it does not distribute one total score.
Good Process Does Not Guarantee a Good Result Either
Chapter 21 showed several compliant entry, resource, and exit processes ending in failure. This chapter shows a process deviation ending in first-attempt success.
Both directions are needed to prevent outcomes from deciding everything. Good process does not promise a good single outcome; one good outcome does not certify process.
This does not sever the relation between them. Reliable process aims to change the distribution and recoverability of outcomes across comparable opportunities. Establishing that requires repeated, bounded, comparable material.
A single-attempt evaluation and a long-term relation are different objects.
Judgment Quality Uses Relations Available at the Time
Before T22-A, Tang Ke knew the unique-link rule and knew that recent files had selected the wrong object Tuesday. Her continued shortcut cannot be explained by absence of a counterexample.
With no previous version problem, quality after one success would be harder to assess. Here, history makes the defect clearer.
Review may not treat the later T22-B result as information she already possessed, but it may use Tuesday’s incident and the frozen card.
The time boundary prevents hindsight severity and prevents ignoring warnings already available.
“I Knew It Would Work” Requires a Prior Record
After results appear, people often feel they knew all along. Before T22, nobody recorded the claim that the first recent file must be V3 or defined supporting evidence.
Tang Ke merely used a familiar entrance. Success cannot upgrade that action into a grounded forecast.
A future judgment must state its object, conditions, and falsifying result before the outcome. Otherwise later certainty is a memory and cannot enter calibration.
Without a prior claim, the result can evaluate execution but cannot generate a forecasting score from nothing.
Good Fortune Must Not Enter the Next Resource Budget
The seven-minute first rehearsal does not justify reducing future handoff time from forty minutes to seven. The time includes the favorable state in which the correct file was first and no external sending.
Compressing the budget around luck makes a later ordering change cause both wrong selection and insufficient repair time. Chance success then increases future fragility.
Budget updates require complete steps and multiple comparable records. Seven minutes may be recorded now without changing a V4 candidate estimate.
The danger is not a good result, but extracting a guarantee larger than its material supports.
Repair the Identification Relation
After T22-B, the team makes the object register the sole entrance. The link displays full object and version, while archived templates disappear from the current workspace. The header and object number must still be checked after opening.
After an interruption, the executor returns to the object workspace rather than resuming from recent files. Joint checking remains because the unique entrance can also be misconfigured.
The change does not delete V2’s archive; it isolates searchable history from execution.
These improvements address the exposed relation without claiming to solve every handoff problem.
The Third Rehearsal Tests the New Rule
At 10:00 Gu Ning again places V2 first in recent files. Following the new entrance, Tang Ke opens the full V3 index from its archive workspace. Ye Cheng’s four checks pass in nine minutes.
T22-C succeeds with compliant entry. It supports the claim “V3 can still be found when recent-list order changes” more strongly than T22-A because the counterexample condition was deliberately present.
One pass still does not prove permanent reliability. Link configuration, object switching, and multi-person permissions need later observation.
The team records “passed one rehearsal under the counterexample condition,” not “problem completely solved.”
Ask What a Good Result Cost
T22-C takes nine minutes, plus management time to create the counterexample, check it, and modify the entrance. Keeping only the fastest run would omit construction cost.
The process change consumes no production slot or working-capital unit, but uses three people’s attention. The book records scattered management time directly rather than forcing it into half-day units.
If the unique entrance needs ongoing maintenance, that belongs in a future capability budget. One configuration is not permanently free.
Result, process, and cost together determine whether an improvement is sustainable.
A Single Success Is One Provisional Convergence
From RC’s standpoint, T22-A’s green result is a definite snapshot locked by a particular interface state, action, and check. It is real but cannot establish cross-situation observational consensus that the first recent file is reliable, because the rule does not control the key success-generating condition.
T22-B and T22-C change the observational path so the shortcut and new entrance each face contrary material. Consensus grows through comparable reobservation, not through stronger language around a good result. Success enters epistemic update without becoming a closed guarantee.
Result-Attribution Table A22-1
T22-A: successful result, entrance deviation; favorable order and a complete archive package contribute to success but do not certify the shortcut. T22-B: failed result, same deviation; changing only order reveals instability.
T22-C: successful result, compliant entrance; finding V3 while V2 is first supports the unique link against this counterexample. All three are internal rehearsals and supply no production success rate.
The judgment column preserves prior grounds, execution preserves the actual entrance, observation preserves file and duration, and result preserves the four checks. No column is overwritten by the final green light.
The table separates luck, rescue, and rule relations inside a success label.
Showing Only Success Deletes the Opportunity Set
A report containing only T22-A and T22-C shows two fast correct selections and may suggest stability. T22-B is the counterexample to recency order; removing it changes the structure of the material.
Selective display need not fabricate data. A team may call the failed rehearsal irrelevant after repair, but it explains the change and bounds what the new plan solves.
The complete opportunity set includes all three attempts launched under the rehearsal plan, their conditions, and which rule each used. Only final-version successes cannot remain.
Failures are preserved to make the denominator and conditions traceable, not to lower a score.
Better Metrics May Merely Replace the Objective
If the metric becomes “open any handoff file within ten minutes,” both T22-A and B may pass. But the objective was to find current V3 and match four fields. Lowering the requirement manufactures a good result.
Another substitution records only sending time, not whether the other party can verify the version item by item. Forcing V3’s wrong package out could improve punctuality while defeating receipt.
Metrics derive from objective and object and freeze before results. A change after an unfavorable result becomes a new evaluation version and cannot overwrite the original.
Good numbers sometimes come from easier questions. Judgment quality checks whether the question changed midstream.
Success After Rescue Preserves the Full Cost
Suppose Ye Cheng finds a wrong file within two minutes and repair still finishes on time. The endpoint succeeds, but checking and rework capacity were used.
Reporting only final duration understates dependence on joint checking and says nothing about recoverability in a tighter window. Rework inside a successful path is not noise.
T22-C has no file rework but does use entrance modification and counterexample setup. Construction and operating costs remain separate rather than disappearing when the goal is reached.
Recovery capability is worth retaining, and the original deviation is worth repairing. The evaluations do not conflict.
Good Luck Most Easily Erodes Stopping Rules
After one successful process violation, the team may deliberately skip checking next time under time pressure. Success becomes alleged evidence that the rule is too strict.
Testing that claim requires the protected object, exposure created by skipping, and comparable repeated results. One favorable state cannot change X17-1.
Nor can any rule be frozen forever by one success. If later material shows the unique entrance and double check cost out of proportion, the team may revise them before a new object and preserve the basis.
The distinction is between change from structure and continuous evidence, and change because nothing happened this time.
External Delivery Requires Another Transfer Layer
T22-C shows that the unique link finds V3’s archive package under an internally arranged list counterexample. Real delivery also includes a new object, current entry, external version, sending authority, and window.
It cannot prove V4 will succeed or supply a delivery probability. If a new object forms, stages and exit cards must be established again.
The rehearsal reduces one unknown: resistance of the entrance to recent-list order. It does not remove other execution and technical unknowns. A transfer claim must name the supported relation.
A good result should update its corresponding link rather than endorse the whole action.
Evaluate Consecutive Successes the Same Way
Ten consecutive correct uses of the unique entrance would be stronger than one, but still require comparable objects, recorded deviations, and conditions that include difficulty rather than only smooth cases.
Repeated success may support process stability, not an absolute guarantee. New permissions, people, or interruptions can remain outside scope.
Repeated failure likewise does not prove every judgment poor. Chapter 23 decides recovery from failure types and resource consequences rather than raw counts.
The evaluation rule remains the same on both sides so that outcomes do not acquire interpretive authority.
Gu Ning also records the adjudicator and time. If executors excuse the same deviation after success while reviewers condemn it after failure, the standard still moves with the result. Frozen rules determine process state first; review then considers amendment, preserving both judgments. Process records cannot shrink merely because the endpoint went well, or future comparison will lose its common fields.
The Rehearsal Does Not Rewrite D17’s State
V3 remains incomplete at 16:00 Tuesday. The three-unit final payment has not arrived, and cash remains eleven: eight safeguarded, three originally unallocated. T22 uses no half-day slot or working-capital unit.
The correct archive package and new entrance improve future reuse and handoff capability but are not late delivery. The other party has not proposed V4.
The week remains seven existing slots, two used by V3, and one recovery slot. The rehearsal uses management time without consuming that slot.
A good rehearsal cannot flow backward into a commercial result, just as a bad commercial result cannot erase a locally correct judgment.
From Result Back to Sustainable Judgment
By 10:30 Wednesday, the team has missed two successive windows under V2 and V3. One successful internal repair does not remove their effect on resources, relationships, and people.
Looking only at T22-C may prompt an immediate V4 to validate the fix; looking only at two nondeliveries may make every capability investment seem pointless. Both reactions let the latest result control the next step.
Chapter 23 puts the definition of consecutive failure, remaining resources, recovery actions, and reentry conditions on one page. Recovery is not proving all is well through one good result. It rebuilds action capacity without continuing to expand exposure.