Why Two Loss Rates of Twenty Percent Are Not Enough
At 18:30 Monday, Gu Ning drew a four-cell table beside the exposure map. If the station later knew the separate probabilities of adverse results for E4 and G2, could it review them one by one and say that two projects already provided diversification?
A marginal probability states how much space each event occupies across all scenarios. It does not say how the two occur together. Two twenty-percent losses may mostly alternate, occasionally overlap, or always be triggered together by one condition. Equal margins can expose a subject to very different concentrations in one realization.
This chapter first uses numbers wholly independent of Chengwan to demonstrate three joint structures, then returns to D17 with a blank-valued table. Demonstration probabilities, loss units, and repetition conditions do not enter E4 or G2. D17 lacks a sufficient batch and adopts no numerical joint probabilities.
A Four-Cell Table First Preserves Every Combination
Let A mean that the first item incurs a loss and B that the second does. Their joint results have at least four cells: both A and B, A only, B only, and neither.
The cells are mutually exclusive and, if they cover the modeled range, sum to one. A's marginal probability is the first two A cells; B's is the joint cell plus B only.
Given only P(A) and P(B), the joint cell P(A∩B) remains unknown. Without it, the record cannot show how much loss in the two margins appears in the same scenario.
How the Union Formula Exposes the Shared Portion
The probability of at least one loss is:
$$ P(A\cup B)=P(A)+P(B)-P(A\cap B) $$
The intersection is subtracted because adding the margins counts their shared scenarios twice. Formal inclusion-exclusion and independence relations can be checked against MIT 18.600 Probability, Lecture 4 and MIT Introduction to Probability, Lecture 3. These sources support definitions and formulas, not D17 inputs.
Changing the shared portion changes the union. The formula does not determine the intersection; it prevents two margins from concealing that missing position.
Demonstration One: Mathematical Independence
Let P(A)=P(B)=0.2 and expressly assume independence. Then:
$$ P(A\cap B)=P(A)P(B)=0.04 $$
The probabilities of A only and B only are each 0.16, neither is 0.64, and at least one is 0.36. The four cells sum to one.
The 0.04 comes from the independence assumption, not from different project names or separate budgets. Without grounds for the joint relation, multiplication merely fills in a model.
Demonstration Two: Mutually Exclusive Losses
Keep both margins at 0.2 but stipulate that A and B cannot occur together. Their intersection is zero, A only and B only are each 0.2, neither is 0.6, and at least one is 0.4.
Mutual exclusivity is not independence. If A occurs while B has a nonzero margin, B cannot occur conditional on A, unlike its twenty-percent overall rate. “Each happens on its own” in everyday language does not establish mathematical independence.
This structure has no joint loss and at most one loss per realization. Yet the chance of at least one, 0.4, exceeds the independent structure's 0.36. Before calling one “riskier,” state whether the concern is any loss or simultaneous loss.
Demonstration Three: Perfectly Aligned Losses
Again keep both margins at 0.2 but stipulate that A and B always occur together, so they occupy the same scenarios. The intersection is 0.2, both single-event cells are zero, and neither is 0.8.
The probability of at least one is also 0.2, lower than in the other structures. But whenever loss appears, both fall together. This matters more if the subject fears concentrated loss in one realization.
“At least one occurs less often” and “joint loss occurs more often” can both be true. One aggregate percentage cannot answer both questions.
All Three Structures Have Identical Margins
P(A) and P(B) remain 0.2 under independence, mutual exclusivity, and perfect alignment. A report showing only separate loss rates makes them look identical.
All differences lie in allocation between the intersection and single-event cells: 0.04 jointly under independence, zero under exclusivity, and 0.2 under alignment. Nominally there are two items in each case, but joint consequences range from none of the losses occurring together to all losses doing so.
This is where separate assessment is insufficient. Every individual judgment may be correct while the combination omits its most important cell.
Expected Number of Failures May Also Be Identical
Let N be the sum of indicators that A and B occur. Under all three structures:
$$ E[N]=P(A)+P(B)=0.4 $$
Expected count does not require independence and is equal across them. A dashboard showing an average 0.4 failures per round cannot reveal concentration.
The average is correct for expected count in the aggregate. Whether a subject can bear two simultaneous failures requires the joint distribution and resource state.
Equal Expected Losses Can Have Different Tails
Now assign a loss of three demonstration units to each occurring event and zero otherwise. Each item has expected loss 0.6 and total expected loss is 1.2 under all three structures.
The mutually exclusive structure loses at most three at once. Independence has a 0.04 chance of losing six. Perfect alignment has a 0.2 chance of losing six. Total expectation is equal while maximum paths and six-unit probabilities differ.
These units do not enter Chengwan's ledger. They show only that average loss cannot answer concentrated consequence.
Put Joint Loss and Any Loss in Separate Fields
If the goal is to reduce any error, the mutually exclusive demonstration's 0.4 exceeds perfect alignment's 0.2. If the goal is to prevent a six-unit loss, alignment's 0.2 exceeds exclusivity's zero.
No goal-free risk ordering preserves both relations. The subject must state whether it protects all-task success, avoids concentration, maintains minimum cash, or pursues another condition.
For Chengwan, existing fulfillment and the V2 window each matter, while decisive questions include simultaneous occupation of Ye Cheng, C1, or recovery margin. The joint table preserves distinct consequences instead of labeling the combination high, medium, or low.
The Intersection Has Bounds Once Margins Are Given
For any two events, the intersection cannot exceed the smaller margin and cannot fall below the greater of zero and the sum of margins minus one:
$$ \max(0,P(A)+P(B)-1)\leq P(A\cap B)\leq \min(P(A),P(B)) $$
When both margins are 0.2, the intersection can range from zero to 0.2. Mutual exclusivity and perfect alignment occupy the endpoints; independence's 0.04 lies between them.
The bounds enforce probability consistency but do not show where the actual intersection lies. A range rules out impossible inputs without replacing dependency evidence.
A Smaller Union Does Not Necessarily Mean a More Bearable Combination
With both margins fixed at 0.2, a growing intersection lowers the probability of any loss from 0.4 toward 0.2. Judged only by whether a round contains loss, perfect alignment seems less often adverse.
But in every adverse round, both losses occur. If each needs one recovery slot, one slot may suffice under exclusivity while alignment needs two. Recovery capacity changes the action meaning of the same joint structure.
Union probability cannot be called combination safety. Report any loss, joint occurrence, total loss, and recovery demand separately to identify the protected line.
Joint Probability and Joint Loss Are Different Numbers
Joint probability expresses how strongly two events occur together. Joint loss expresses what the subject bears in that cell. A joint probability of 0.04 with catastrophic loss may touch survival more than 0.2 with minor loss.
Multiplying probability by loss gives that cell's contribution to an expected quantity; it does not shrink the realized loss. Chapter 5's bearability check still applies to the full path, while Chapter 4's weighting answers another question.
D17 lacks both a joint probability and a complete recovery cost. Gu Ning keeps both blank rather than using “many shared nodes” as a probability or “one margin slot” as a consequence calculation.
Pairwise Independence Does Not Establish Joint Independence Among Many Items
For three or more objects, pairwise independence generally does not imply that the probability of all events equals the product of their margins. A constraint may appear only when all three are considered together.
A portfolio cannot declare complete diversification from a table of pairwise relationships. If action concerns three simultaneous demands on recovery resources, that joint event or a mechanism supporting full independence must be checked directly.
D17 formally defines only E4 and G2 here; other E jobs remain on the exposure map. Expanding the joint table later requires a larger outcome space rather than mechanical reuse of four cells.
How Historical Batch Selection Changes Observed Co-occurrence
If records include only cases where both items entered formal stages, earlier stops and resource-limited objects are excluded; observed joint results may not represent all proposals. Keeping only weeks with a loss inflates the shared cell's sample proportion.
Batch starting point, entry criteria, and missing records belong to joint estimation. Identical project names do not make batches combinable when stage selection differs.
Chengwan has no such historical joint batch. E1 through E6 are not imagined as six complete observations, and one G1 pass does not enter an E4-G2 joint frequency.
Concurrent Stress Testing Without Probabilities
Gu Ning can postpone the size of the shared cell and ask: if Ye Cheng and C1 are unavailable together, how do E4 and G2 stop, how many recovery slots are needed, how is the first-payment balance handled, and does the eight-unit protection remain intact?
A stress test stipulates a scenario and traces consequences without claiming a probability. It can expose gaps in recovery rules and prioritize node checks, but cannot report expected loss.
D17's most direct test is E4 extending from Wednesday morning into the afternoon. E4 retains priority; G2 pauses and reopens the window instead of taking the margin slot automatically. This exit exists, but V2 may miss its original receipt condition and that loss remains.
“Safer After Diversification” Must Say What Is Safer
Adding a mutually exclusive path may remove joint loss while increasing coverage of any loss. Adding highly aligned projects may barely change the chance of any loss while increasing the number lost at once.
Diversification for survival must state which unacceptable path it reduces, whether independent recovery remains, and how switching costs are treated. More projects are only a surface change.
Multiple paths may also broaden favorable coverage. The demonstrations focus on losses without declaring diversification universally bad. Action assessment returns to the complete result rather than merely counting failures.
Common Cause Is Not the Only Explanation for Perfect Alignment
Perfect alignment may arise from a common cause or simply from defining two identical events. Frequent co-movement alone does not determine the mechanism.
Conversely, a common node does not guarantee identical results. C1 unavailability may affect both E4 and G2, while a Ye Cheng conflict may delay only the later G2. A dependency structure can generate several joint tables.
Statistical co-movement and a common mechanistic entrance remain separate records. They can reinforce one another without allowing a direct leap from one term to the other.
Conditional and Unconditional Independence Must Not Be Mixed
Suppose A and B both depend on environment C. They may be independent within each state of C but move together when those states are mixed. Or they may happen to be independent overall but dependent within a condition group.
Thus “independent after controlling for device state,” when grounded, applies only to that condition and cannot erase device-state probability. Overall multiplication also does not prove independence within every material or time group.
D17 has no corresponding batch and makes no conditional-independence assumption. C1, M7, and time nodes on the map are future grouping and observation entrances.
More Nominal Paths Can Share More Deeply
If the team splits G2 into fabrication, inspection, and receipt task lines, the table grows from one project to three. If all require Ye Cheng and one version entrance, critical sharing does not fall.
Splitting reveals steps but proves no diversification. The joint-loss structure changes only when some paths obtain different and effective necessary conditions.
Likewise, merging E4 and G2 as “Wednesday work” would hide distinct objects. Records preserve task boundaries and cross-task dependencies; naming choices do not decide independence.
A Backup Must Work in the Relevant Adverse Scenario
If C2 shares C1's power entrance, both fail when that entrance fails. If C2 requires Ye Cheng, equipment redundancy does not address his absence.
Backup count therefore cannot directly enter the joint table. The record asks which failure it addresses, whether it shares the original root, and whether switching time and cost are bearable.
Chengwan has not validated C2 and does not call “two benches” equipment diversification. Stopping G2 and protecting E4 is an existing exit, but it protects the subject rather than guaranteeing on-time V2 completion.
One G1 Pass Supplies No Joint Frequency
G1's M7 sample passed S-3 and supplies one limited observation. It does not form a repeated joint batch with E4 and formal G2 outcomes.
“Only success so far” cannot make joint failure zero; three paths sharing M7 cannot make it one. Both expand either structure or one result into a frequency.
The G1 result still supports identification of material and local procedure. Refusing to invent joint probability does not delete favorable evidence.
Return to D17 and Define Two Loss Events
Gu Ning defines A_D as E4 failing to obtain the checklist's handoff-ready result within its stored deadline. B_D is G2, if released, failing to obtain a handoff-ready result in the jointly stored Wednesday window.
Both need records for adjudication. B_D is action-conditional on G2 release; because G2 remains unreleased, it is not yet a live outcome-window event.
The joint question asks how A_D and B_D occur together when both enter their processes. Defining events neither adopts probabilities nor predicts failure.
How D17's Four Cells Are Filled Now
The cells are: neither E4 nor G2 obtains its result; E4 alone fails; G2 alone fails; both obtain results. Each lists candidate paths, affected goals, resource consequences, and adjudication material before probability.
The shared cell connects candidates such as Ye Cheng unavailability, C1 entrance failure, or an M7 problem. The G2-only cell includes an E4 extension compressing the afternoon, V2 execution deviation, or a changed client window. The E4-only cell retains E4-specific paths.
Candidate paths are neither established causes nor a mutually exclusive complete list. Every probability remains unknown, and no joint total is calculated. Progress consists in no longer setting the shared cell to the product by default.
A G2-Only Loss Can Still Arise from a Connection Between Both Tasks
If E4 extends but still completes by its own deadline, A_D may be false while its occupation of Ye Cheng and C1 causes G2 to miss the afternoon. The final table lands in “G2 only,” though the mechanism is a cascade between both tasks.
Shared dependencies therefore appear beyond joint failures. A connection can change the probability or consequence of a one-sided cell.
The joint table needs both process arrows and final cells. Endpoints alone may miss E4's effect on G2; the exposure map alone cannot say which event ultimately occurred.
Why the Joint Cell Reopens the Survival Check
If E4 and G2 need recovery together, one retained slot may not serve both. If the common problem needs added material, use of the four unallocated units requires new authorization.
This does not prove the joint cell will occur or breach the protected eight. It identifies a concrete stress: simultaneous recovery needs can change survival even when each is separately bearable.
Gu Ning marks the shared cell “maximum concurrent recovery must be checked before G2 release” instead of treating separate project budgets as joint protection.
Action Can Improve Without a Joint Probability
Under an unknown intersection, the station can limit commitment: retain E4 priority; let G2 Wednesday continue only after E4 releases Ye Cheng and C1; do not consume the margin slot automatically; and check M7 and version identifiers before starting.
These measures target known dependencies without invented percentages. They may reduce certain misuse or cascade consequences but lack batch validation and cannot claim a quantified reduction in common failure.
If mitigation costs are too high, the station can decline G2 and settle the first-payment balance. Joint uncertainty does not automatically demand the most elaborate safeguard, but it belongs in the choice.
What Would Support an Independence Assumption?
Using marginal multiplication later requires the events and condition range, data or mechanism supporting independence, and treatment of shared nodes. Different objects and separate owners are clues, not sufficient proof.
Comparable joint records may support estimation, and a clear random mechanism may support a model; each has a scope. Environmental, resource, or schedule changes may reopen the relation.
Without support, retain an intersection range, competing scenarios, or stress test. Unknown means neither worst case by default nor automatic multiplication.
Correlation Coefficients Cannot Replace Four-Cell Meaning
Binary loss indicators permit a correlation coefficient, but its value still depends on margins and the shared cell and does not directly state loss amount, cause, or recovery path. Equal correlation under different margins produces different joint probabilities.
A coefficient can support some comparisons but is not a portfolio action license. It needs defined data and cannot be scored subjectively from edge count. D17 receives no coefficient here.
Gu Ning retains cells and causal nodes because they return directly to how the two events occur together. If enough batches later exist, a coefficient may summarize them without deleting the underlying table.
Diversification Also Checks Whether Recovery Is Aligned
Two losses may have different causes but require the same person or money to recover. Even approximately independent events can compete after occurrence, leaving combination survival undiversified.
Conversely, a shared upstream node that can be replaced quickly may have controlled joint consequences. Cause, outcome jointness, and recovery structure are three levels; none alone is sufficient.
D17's one margin slot and four unallocated units belong to recovery. How they would serve E4 and G2 together remains unresolved, so the event table cannot declare the margin adequate.
A Portfolio Average Cannot Answer for Subject Bearability
Even when each expected loss is known, their sum may not depend on joint structure. The subject may care instead about a one-time demand of six units, two recovery slots, or its only inspection entrance.
A combination page therefore places expected total, joint occurrence, maximum concurrent consequence, and recovery resources side by side. An average supports part of the comparison but not a claim of complete evaluation.
Chapter 5's survival line returns here: if a low-probability joint cell crosses it, full consequences cannot enter only as a small expected weight. Adoption still connects consequence to capacity.
The Joint Table Must Preserve Time Order
Both jobs may fail Wednesday because C1 was unavailable in the morning or because E4 first extended and then removed G2's window. The endpoint matches, but preventive actions differ.
Gu Ning adds a time column: when the common node changes, when E4 releases, when G2 must continue, and when adjudication forms. The result can later distinguish common cause, resource competition, and cascade.
A static correlation table easily loses order. The chapter does not build a dynamic probability model; it requires only that joint records can return to process.
The Joint Table Extends Spatial Margin
Path diversity in space must be realized through joint relations. Two projects with identical marginal outcomes provide less heterogeneous margin when driven by one observational condition than when their conditions differ. Independent, exclusive, and aligned structures therefore create different concurrent consequences.
RC praxis uses heterogeneous information and paths to hedge blind spots in a single consensus. The joint table translates that demand into action: ask not only how each path behaves but whether the same pressure makes them converge together. Mathematics describes structure; the subject's recovery capacity decides which joint consequence is bearable.
The D17 Joint-Failure Table
At 19:00 Monday, Gu Ning saves the table. Marginal fields A_D and B_D are unknown, as are the shared cell, both single cells, and joint-success cell. The cause field links Ye Cheng, C1, M7, Wednesday's window, the version entrance, and task-specific paths.
Consequence fields separately preserve E4 fulfillment, the V2 window, refund or final payment, recovery labor, and cash effects. The shared cell adds a check of whether concurrent recovery exceeds one margin slot and unallocated-resource authority. No candidate cause is recorded as occurred.
The model field retains the three demonstration structures for sensitivity but expressly adopts none as D17's actual model. If margins are later obtained, the intersection still needs grounds or a consistent range.
What the Nominal-Diversification Check Establishes
E4 and G2 are two task objects with separate outcomes and cannot be merged into one event. They also share critical nodes and cannot be inferred independent from object count.
The margin ledger shows point-estimate capacity; the exposure map shows concentrated execution and recovery conditions. The joint table connects the facts: separate budgets do not prevent simultaneous breach, while sharing does not make common failure certain.
G2 remains “joint structure listed, actual probability unknown, maximum concurrent recovery pending” and receives no resource release.
Handing Off to the Waiting Question at 19:20 Monday
At 19:20, there is no new client message. Fourteen-unit cash and three uncommitted slots remain unchanged. E4 has not moved; G2 has no reservation, purchase, or production.
The three demonstrations share margins of 0.2 and total expected loss but yield joint-loss probabilities of 0, 0.04, and 0.2. The difference comes from joint structure rather than arithmetic technique, and none supplies D17 with numbers.
The next chapter adds the capacity to wait. Keeping G2 unreleased can reveal M7, C1, and E4 states but may lose Tuesday preparation or Wednesday receipt. Immediate release gains time while locking resources under unresolved joint relations. Waiting has value only in relation to how long the subject can wait, what can arrive, and which paths waiting itself changes.