A Failure After the Previous Pass
At 08:30 Wednesday, E4 began on its existing schedule. At 10:20 it produced the handoff-ready result required by its checklist, releasing Ye Cheng and inspection bench C1. V2 did not delay existing work, so A_D in the joint table can be adjudicated as not occurred.
G2's Wednesday segment still needs an entry review. At 10:35, Ye Cheng first checks C1's current state, then takes the inspection piece left by Tuesday's first segment and reads the internal deviation quantity d under V2 checklist procedure S-3. The rule is pass when d is no greater than 2.0, and fail when d exceeds 2.0.
The first reading is 2.3. G1's archived Monday reading was 1.6 and passed. The results point in opposite classificatory directions, and the new reading crosses the same 2.0 boundary.
d and the threshold are internal scales in Chengwan's fictional checklist and correspond to no actual length, strength, or technical standard. This chapter examines records and action, not external industry conclusions from these numbers.
Preserve the Raw Reading Before the Adjective
Tang Ke records 2.3, its time, inspection-piece identifier, C1 identifier, and S-3 version on a new observation page. She does not first write “material failure,” “equipment drift,” or “weather effect,” because those explain the source rather than state the result.
“Fail” classifies 2.3 under the existing rule. “Inconsistent with G1” states its relation to 1.6 and the earlier pass. Both have recorded grounds, yet neither explains why the difference appeared.
If the record keeps only “abnormal,” a later reader cannot see distance from the boundary, the rule used, or whether the threshold changed. Separate fields for raw quantity, classification, and explanation preserve observation under its own identity.
Do Not Remake the Sample Before Repeating the Reading
Ye Cheng proposes another reading. Gu Ning permits rereading the same inspection piece but not adjusting, repairing, or replacing it first. A changed object would make the second result include an intervention and prevent it from checking the stability of the first reading alone.
At 10:45, they reposition and read under the original S-3 sequence. The same piece yields 2.2. It remains above 2.0 and close to 2.3 without being a mathematical copy.
The repeated observations still share the same piece, C1, executor, environmental time, and procedure. Gu Ning does not call “two failures in a row” two independent failure samples.
Agreement on Rereading Does Not Eliminate Measurement Issues
The proximity of 2.3 and 2.2 weakens some explanations based on a casual transcription error in the first reading, but it does not prove C1 perfectly accurate. A common device offset could affect both; repositioning might also change the object.
C1's entry check satisfies the station's current check page. That supports entering use, not zero measurement error. Coverage of the page, its reference object, and the actual S-3 reading must still correspond.
The measurement candidate therefore changes from “unchecked” to “entry check passed; systematic or object-interaction effects unresolved.” Evidence changes candidates' relative support without selecting a unique cause immediately.
Pause the Wednesday Segment First
At 10:55, Gu Ning compares the stage gate. Wednesday entry requires stable version and material state and uses G1's result as limited input. The new observation conflicts with that relation. Releasing another slot and unit would deepen commitment while the difference remains unclassified.
At 11:00, she marks the Wednesday G2 segment paused, releases no candidate slot, and leaves the remaining conditional unit untouched. The reason is a failed entry result, not a proven environmental shift.
This is the adverse exit of the gate. Tuesday's slot and unit do not lower today's entry standard, and the approaching 14:00 window does not turn 2.2 into a pass.
Pausing Does Not Kill All of V2
G2's first segment produced actual work, and G1 has a pass record. The new observation prevents continuation under the original window but does not prove every material unusable or cooperation permanently over.
The station can examine measurement, procedure, material, and environment and reopen the window with the client, or stop and settle. Change classification and consequence checks determine which path is reasonable.
“Paused” preserves an investigation entrance while limiting new input. If production continued or conditional funds were used, the word would be only a label.
Reopen the External Window Faithfully
At 11:10, Tang Ke tells the client through the original channel that Wednesday's entry check produced a result inconsistent with the prior record, so the station has paused the segment and will not hand over between 14:00 and 16:00; it will provide a retrievable explanation after review.
She does not write “the environment caused this” or promise same-day repair. The message states actual condition and affected window without presenting an uninvestigated cause as explanation.
At 11:25, the client acknowledges this, agrees not to receive today, and waits for the station's explanation. The original Wednesday window will not execute, and no new receipt window has formed.
Window Result and Cause Can Be Separate
By 16:00 Wednesday, V2 has no handoff-ready result within the stored window. B_D therefore occurred under Chapter 11's definition. E4 completed on time, so the joint table lands in “G2 only fails.”
This endpoint does not prove independence. E4 and G2 still share Ye Cheng, C1, and M7; those nodes simply did not prevent E4's result this time. E4's release was itself a condition for G2's check.
B_D occurring does not explain the new reading. The window can be adjudicated while mechanism remains unknown. The discipline from Part 1—definite about the outcome, limited about the cause—continues in action.
What Counts as a Deviation?
Here deviation means that a new observation conflicts with a stored reference or expected relation. Both 2.2 and 2.3 exceed the threshold and differ from G1's 1.6, so deviation is established.
Deviation carries no built-in cause. It may come from within-condition natural variation, measurement, execution, a material subitem, environmental conditions, version correspondence, or a combination.
Calling it noise selects “no model change needed” too early. Calling it a regime shift selects “the generating mechanism changed.” Gu Ning uses a neutral name for the location requiring explanation.
What Counts as a Shift in Conditions?
On D17's change page, a condition shift means that a set of conditions relevant to result generation changed enough that the old relation no longer transfers directly, with retrievable material supporting that change.
It does not mean the whole world entered a permanent new phase. Stable differences in M7 subitems, operating interval, workspace state, or equipment relation may form a local condition version.
One inconsistency can nominate a shift, but rarely identifies what changed, when it began, or whether it persists. Naming a shift needs structural evidence beyond outcome count.
Within-Condition Variation Is a Candidate, Not a Dismissal
Repeated results may differ even when procedure and environment remain stable. Listing “within-condition variation” preserves this possibility; it does not excuse every difference from investigation.
Determining whether it can explain a move from 1.6 to above 2.2 needs a repeat distribution or mechanism under comparable conditions. Chengwan has too few observations to call the difference normal merely because variation exists.
If every inconsistency becomes randomness, action never reopens. If every inconsistency becomes a shift, models cannot persist. Both require evidence and thresholds.
Separate Measurement Change from Object Change
A reading can change through the measured object or the measurement chain. C1 state, repositioning, reader, record conversion, and decision rule belong to measurement.
Object differences include that the G1 local sample and G2 inspection piece are not the same item, their M7 subidentifiers differ, and their processing stages differ. A shared batch name does not erase these differences.
Without separation, the team may repair equipment for a material difference or replace material to hide a reading problem. Classification arranges the next check that can distinguish candidates.
Execution Deviation Must Return to the Original Steps
Ye Cheng reviews G1 and G2 process pages and confirms the same S-3 version and the same executor. This reduces personnel and version differences without proving every step identical.
Tuesday's first segment included formal processing while G1 was a local sample. Waiting duration, object position, and prior steps may differ. If the checklist requires them to match, compare each rather than only the executor's name.
“Execution problem” is not a personal accusation. It asks whether actual steps correspond to the adopted version. A deviation should revise process or model, not substitute blame for evidence.
A Shared Material Entrance Still Has Subitem Differences
G1 and G2 both connect to M7, but the record shows G1 used M7-a and Tuesday's formal segment used M7-b. They share a registered entrance without being the same material.
This makes the subitem a candidate and requires refining the exposure map. It does not prove M7-b defective or assign every difference to material from one pass and one fail.
A later comparison can vary subitem while holding other conditions as closely as possible, or inspect M7's internal records. The comparison itself needs bounded resources; searching for cause cannot consume the remaining unit without limit.
The Environmental Record Differs
The station's process page includes a fictional environmental quantity h marking workspace conditions. h was 4.2 during G1 and 6.1 at Wednesday's check. Time and h both changed.
This co-change admits environment as a candidate without proving h caused d. Both may be directly related, share a third condition, or merely coexist in a small record.
Gu Ning preserves h's raw values, device, and times without renaming it temperature, humidity, or a real physical unit. Fictional scales support no real causal claim.
A Threshold Gives Small Differences Different Labels
The move from G1's 1.6 to 2.2 is 0.6; 2.2 exceeds the passing line by only 0.2, yet it is classified as failure. A discontinuous classification does not prove an equally abrupt change in the generating process.
The checklist threshold is an action rule mapping a continuous reading into entry or pause. Looking only at pass/fail may mistake the rule's discontinuity for a sudden environmental jump.
Chapter 15 treats this directly. For now, both d and classification remain, with labels unable to replace values.
Two Rereads Cannot Establish a Trend
The value falls slightly from 2.3 to 2.2, which can be narrated as recovery. But two values share the same short interval and are too few to establish a stable direction.
Likewise, a line through 1.6, 2.3, and 2.2 does not automatically show a rise and fall. The points use different objects, times, and stages; the line records visual order only.
A trend needs more continuous material under relatively consistent acquisition rules. Gu Ning does not predict the next point from three.
Rereading the Same Piece and Sampling Another Answer Different Questions
Rereading the same piece from 2.3 to 2.2 primarily tests short-term repeatability. It does not show how other M7-b material behaves or separate object from measurement differences.
Making a new sample can test across objects but adds preparation, material, and time. It is not a free “independent data point”; its shared conditions with the first must be recorded.
Gu Ning authorizes only same-piece rereading because Wednesday entry first needs confirmation of the reading with limited added commitment. A stronger comparison becomes a separately bounded question in the next chapter.
The Start of Change Can Only Be an Interval for Now
G1 recorded 1.6 Monday; Wednesday entry produced 2.3. Even if later evidence supports changed conditions, the initial claim can only place change between observation times, not at an invented Tuesday instant.
Completion of Tuesday's first segment does not mean S-3 was measured then. Without a reading, one cannot know whether crossing occurred before, during, or after processing. Process records can narrow the interval, not add an observation never taken.
An uncertain starting point affects attribution and recovery. Change before material use points toward the entrance; change after processing points toward the process. The page records “after Monday's check and before Wednesday's entry” until narrowed.
Deviation Magnitude Differs from Action Sensitivity
Although 2.2 exceeds the threshold by only 0.2, it blocks stage release. Another large difference far from any decision boundary might be less urgent. Action sensitivity comes from rules and consequences, not necessarily a stronger mechanism.
Nor can a 0.6 difference become causal importance. Measurement scale, ordinary variation, and causation are unknown. Gu Ning neither ignores a small crossing nor calls it dramatic environmental change.
Retaining raw distance permits later review when rules or measurement uncertainty become clearer. A pass label alone would lose it.
The Classification Page Begins with Six Candidate Layers
Layer one is record or version: transcription, threshold, or checklist reference. Layer two is measurement: C1, repositioning, and the reading chain. Layer three is execution: correspondence between actual steps and adopted procedure.
Layer four is material subitem: a result-relevant difference between M7-a and M7-b. Layer five is environment: h and other process conditions. Layer six is within-condition variation: whether results may cross the line even while the other relations hold.
Candidates can coexist and are not a forced choice among six. Investigation seeks material that distinguishes them while permitting a new relation outside the present classification.
What Material Each Candidate Needs
The record layer returns to originals, version hash-like identifiers, and manual transcription. Measurement examines reference pieces, state checks, and independent readings. Execution compares process times item by item. Material compares subidentifiers under similar conditions.
Environment needs continuous records, same-object controls, or an explicit mechanism. Within-condition variation needs comparable repeats and a distributional range. One endpoint cannot fill all fields.
Gu Ning records support, opposition, and unknown for each layer. “Not checked” never becomes “no effect.”
Investigation Order Also Considers Reversible Cost
Returning to originals, checking versions, and reviewing transcription cost little and can remove obvious record error, so they come first. Dismantling completed work or making many samples consumes more and requires separate authorization if earlier checks are insufficient.
This order does not say cheap explanations are more likely. It first uses low-commitment discriminating actions. A found version error may avoid destroying the object; if none is found, other candidates remain.
Investigation also has a stopping line. The remaining conditional unit cannot all be used without the client and stage rules, and four original unallocated units are not an automatic investigation fund.
A Control Must Make Competing Explanations Predict Differently
A new sample that changes material, device, executor, and h all at once cannot reveal which mattered even if it passes. A useful control varies the relation under examination as narrowly as feasible and discloses what could not be held fixed.
Real operations may not achieve strict single-variable control. The book does not discard imperfect controls but limits their conclusions: which explanations they weaken and which remain entangled.
The next chapter designs continuous recording and paired checks so material subitem and h at least have comparable positions. A design does not itself establish an environmental cause.
Present Support and Limits
Original records show an unchanged threshold and S-3 version, with times for 2.3 and 2.2, reducing a simple transcription explanation. C1's state page passes, limiting some obvious unavailability cases without excluding common offset.
The executor is the same, while object and stage differ. The M7 entrance is shared, while subidentifier differs. h moves from 4.2 to 6.1. Every group contains held and changed relations, and none determines cause.
The page therefore chooses no unique mechanism. The strongest conclusion remains that the new observation fails under the same decision rule and the old G1 result cannot directly authorize Wednesday continuation.
How One Deviation Can Change Action
Unknown cause does not prevent a pause. A gate needs only a failed entry condition to block further release. Mechanism requires more material and may be investigated after stopping.
Unknown cause also does not authorize unlimited investigation. The station has one conditional unit and two uncommitted slots, one serving recovery margin. Every new check needs a question, ceiling, and relation to the window.
Required evidence depends on consequence. Pausing new commitment can follow a clear failure; declaring all of M7 permanently unusable or rewriting every model requires stronger support.
Do Not Select a Cause by Severity of Outcome
Missing Wednesday matters to V2 and can make a major-change explanation emotionally attractive. Severe consequences do not make an environmental shift more likely; they increase the value of investigation and recovery.
Nor does a currently bearable loss permit ignoring 2.2 as a small error. Crossing the adopted threshold is a rule fact; causal judgment has different evidence.
Consequences govern how cautiously to act; evidence governs how strongly to state a cause. They connect without substitution.
The Old Result Is Not Invalidated by the New One
G1 obtained 1.6 Monday under its object, procedure, and time with adequate records, so its pass remains. Wednesday's failure cannot rewrite G1 as having failed.
The G1 pass likewise cannot erase Wednesday by calling 2.2 necessarily mistaken. Both observations stand. The conflict concerns whether one relation explains both, not which fact to delete.
Preserving both lets the model reopen. Keeping only the result that supports the current narrative prevents feedback from changing anything.
A Plausible Post-Result Story May Still Lack Evidence
Tang Ke suspects workspace change from Monday to Wednesday; Ye Cheng suspects a difference between M7-b and M7-a. Both narratives connect present observations without excluding measurement, execution, or within-condition variation.
A coherent story is not uniquely supported. Once the result is known, many paths can be narrated smoothly.
Gu Ning records the conjectures as explanations to test but forbids certainty in external communication. The next step seeks material under which competing explanations expect different results.
External Explanation Reports Only Formed Layers
Tang Ke tells the client: “The entry check is inconsistent with the prior record, and we are reviewing it.” This reports observation and state without naming M7-b, C1, or h as the fault.
Calling it environmental change could alter the client's schedule and responsibility; evidence might later point toward measurement. Limited language is not concealment but separation of knowledge from candidate explanation.
The station still acknowledges that the original window cannot execute. Unknown cause does not postpone state notice. Transparency can remain bounded without choosing between facts and a certain cause.
Change Classification Does Not Delay Adjudication
“The cause is unclear” sometimes becomes a reason to deny window failure. Here B_D is definite at 16:00, and investigation cannot turn receipt back into pending.
Conversely, a failed window need not end every check. If the same condition affects later versions, E tasks, or recovery, causal material has action value.
Outcome adjudication and mechanism investigation have different completion points. Neither completes or blocks the other.
Deviation Is the Reobservation Entrance of the Prospective Cycle
RC's prospective focus does not solidify expectation into reality. It requires continued movement through expectation, reality, and renewed expectation. Values 2.2 and 2.3 deviate from the prior pass and become reobservation signals: inspect which relation changed rather than preserving the old narrative or inventing new certainty.
The change-classification page structures a healthy feedback loop. It lets deviation move through record, measurement, execution, material, and environment, preventing old consensus from hardening through self-reinforcement. At the same time, outcome adjudication constrains explanation so an open field of possibility cannot evade a closed window.
The D17 Change-Classification Page
At 15:00 Wednesday, Gu Ning saves the page. Observation: G1 d=1.6 passed; G2 entry d=2.3 and same-piece reread d=2.2 both fail under the rule d≤2.0. All values retain their objects and times.
Held items: S-3 version, executor, and shared M7 entrance. Changed items: inspection piece, material subidentifier, processing stage, time, and h. C1's current-state check passed, while systematic reading issues remain unresolved.
Candidate layers are record, measurement, execution, material, environment, and within-condition variation. The conclusion stops at: “a reviewable deviation exists and suffices to pause Wednesday; whether conditions shifted is unknown.”
Updating Resource and Event State
The Wednesday segment was not released, so no new slot or working-capital expenditure occurs. Cash remains thirteen: eight protected, four originally unallocated, one V2 conditional balance. Both uncommitted slots remain; no final payment arrives.
G2's first-segment slot and unit have occurred and cannot return to uncommitted merely because the window failed. Whether its product can be recovered, reused, or become later material remains to be judged.
At 16:00 the joint result is A_D did not occur and B_D occurred, in the G2-only cell. One result updates no joint probability and does not prove shared structure irrelevant.
Handing Off to Evidence of a Shift at 16:20 Wednesday
At 16:20, the client's no-receipt reply, closed original window, and change-classification page are saved together. E4 is complete, C1 can be checked after release, V2 is paused, and no new window exists.
The chapter neither dismisses deviation as meaningless noise nor promotes it to proven environmental change. It fixes observation, separates layers capable of producing the difference, pauses resources according to consequences, and leaves competing explanations awaiting discriminating material.
The next chapter seeks evidence of environmental change through continuous records, controls, and acquisition times. The goal is not more occurrences of the word “fail,” but comparisons of the same object and procedure under different conditions, testing whether changes in h and d repeat while preserving unsupported parts of causal inference.