Correlated Capacity Demand¶
Core Idea¶
When multiple consumers draw on a shared finite resource and their demand peaks are tail-correlated rather than independent, the realized joint peak overruns capacity that was sized — under an implicit independence assumption — for a diversification-discounted sum. The failure is rare but structurally predictable from the correlation.
How would you explain it like I'm…
Everyone Showers at Once
Imagine one water pipe shared by every house on the street. Most of the time people use water at different moments, so the pipe is fine. But on a hot morning everyone turns on their shower at the exact same time, and suddenly the water slows to a trickle. The pipe was big enough only because we forgot everyone might want it at once.
When All the Peaks Line Up
When many users share one limited resource, like power, beds, or bandwidth, planners size it for how much everyone needs together. The trap is that they often assume the busy moments happen at random separate times, so the peaks cancel out. But sometimes the peaks line up: a heatwave makes everyone crank the air conditioning at once, and the shared supply was never built for that combined spike. Worse, sharing the resource was supposed to be the safety trick, smoothing demand across users, and that trick fails exactly when the peaks co-occur. It looks like bad luck, but it was actually predictable from the fact that the demands move together under stress.
Tail-Correlated Peaks
Correlated capacity demand is what happens when a finite shared resource serves several consumers whose demand peaks are tail-correlated rather than independent, so the real joint peak exceeds the capacity that was sized for independent peaks. The shortfall scales with how strongly the peaks are correlated in the tail. The deep problem is a mismatch between two distributions: the planning model treats demands as independent and produces a small capacity envelope, while the realized world makes the peaks co-occur and blows through it. Diversification, the usual resilience move of sharing one resource to smooth load, collapses precisely under the stress conditions that make the demands move together. The pattern is invisible in normal operation because demands genuinely do look uncorrelated, and reveals itself only in the rare stress event, so what looks like bad luck is structural under-provisioning whose timing is rare but whose occurrence is predictable from the correlation.
Correlated capacity demand: when a finite shared resource serves multiple consumer processes whose demand peaks are tail-correlated rather than independent, the realized joint peak exceeds the capacity sized for independent peaks, by an amount that scales with the strength of tail correlation. The structure has five commitments. A finite shared resource with definite capacity (bandwidth, beds, megawatts, liquidity, responders). Multiple consumer processes drawing on it, each with time-varying demand. Demand-correlation in the tail: the peaks are statistically dependent under stress even when they look independent in normal conditions. Capacity sized for non-correlated peaks: the planning assumption, often implicit, is independence, so the design load is a diversification-discounted sum rather than the joint exceedance. And joint-peak realization: when the correlated stress hits, the resource is overwhelmed and the shortfall is rationed with downstream cascades. The driving force is the tail-asymmetry between the planning distribution and the realized distribution; the model's independence assumption yields a capacity envelope far smaller than the joint-exceedance event realizes. The pattern is invisible during normal operation and surfaces only in the rare conditions that drive the correlation, so apparent bad luck is in fact structural under-provisioning that is predictable from the correlation structure.
Broad Use¶
- Healthcare: ICU capacity sized for independent admissions is overwhelmed when a pandemic wave, heatwave, and trauma weekend correlate.
- Electricity grids: a heatwave drives air-conditioning demand up while cutting solar, wind, and hydro, so demand-up and supply-down correlate structurally.
- Banking: correlated default scenarios have many borrowers failing to roll over funding at the same window, overwhelming diversification-sized capacity.
- Reinsurance: hurricane seasons with multiple correlated events overwhelm capacity sized for independent storm losses.
- Incident response: on-call rotations sized for independent incidents saturate when a deploy, a holiday, and a dependency outage coincide.
- Cloud infrastructure: redundant replicas across zones all depending on one upstream storage layer fail together in a regional outage.
Clarity¶
It converts "we got unlucky" into "a known correlation we failed to size for," making the correlating stressor the object of analysis rather than each consumer's marginal demand.
Manages Complexity¶
A wide class of "the resource was overwhelmed" failures collapses into a four-part accounting — enumerate consumers, name the stressor, estimate the joint exceedance, then add capacity, decouple, or ration — that transfers across every substrate.
Abstract Reasoning¶
It exposes the diversification illusion: pooling reduces variance only under independence, and under tail correlation the only structural defense is reserves not exposed to the correlating stressor, since redundancy that shares the common mode is worthless.
Knowledge Transfer¶
- Finance to operations: copula-based correlation modeling ports from credit risk to surge planning, the insight that independence underestimates joint-peak risk carrying intact.
- Grid to hospital: joint-peak planning (correlated demand plus correlated supply failure) ports to surge planning (correlated admissions plus correlated staff illness) unchanged.
- Catastrophe bonds to cloud: correlated-peril-exposure logic ports to multi-tenant capacity planning, the question always being "what correlates the peaks, and is our reserve exposed to the same thing?"
Example¶
A power grid "adequate" by independent-peak planning blacks out during a heatwave because extreme heat couples millions of air-conditioning loads and cuts thermal, hydro, and wind supply at once — fixed only by sizing for the coincident peak and holding interconnection to a grid in a different weather system.
Relationships to Other Primes¶
Foundational — no parent edges in the catalog.
Children (1) — more specific cases that build on this
- Thundering Herd is a kind of Correlated Capacity Demand — The file explicitly names thundering_herd "a subspecies of correlated capacity demand" twice (Clarity + Not-to-be-Confused-With): both are shared-finite-resource + correlated-tail-demand, differing only in what makes the correlation (a shared release event in thundering_herd vs a common-cause stressor in the general prime). Direction verified: the general prime subsumes the timing-artifact special case. thundering_herd is a valid candidate slug. (Distinct from adaptive_capacity, risk_pooling, margin_of_safety per Phase-C — those stay severed.)
Not to Be Confused With¶
- Correlated Capacity Demand is not Risk Pooling because risk pooling reduces variance by aggregating independent risks, whereas this prime is precisely the case where shared tail correlation reverses the diversification discount.
- Correlated Capacity Demand is not a Margin of Safety because it explains why a margin sized under independence is the wrong size, the correlating stressor making the realized joint peak far exceed the marginal peaks the margin assumed.
- Correlated Capacity Demand is not a Thundering Herd because the herd correlates in time via a shared release event (cured by jitter), whereas this prime's correlation is a standing feature of the demand under a common-cause stressor (cured by joint-peak sizing).