Dependency Distribution Concentration¶
Core Idea¶
When a dependent system relies on an upstream network of providers, the shape of how that dependency weight is distributed across the providers is a structural property of the dependent system in its own right. When the weight is concentrated on a small number of upstream nodes — one provider supplying most of a critical input, a single species pollinating most crops, a single counterparty clearing most trades — the downstream system's fragility is bounded by upstream events at those few nodes, not by what the downstream system has invested in its own defenses.
The structural commitment is not the binary fact of dependency, which is the relation; it is the distribution shape across providers, skewed toward a few or spread broadly. A system with the same total dependency volume can be robust or brittle entirely as a function of this shape. Concentrated upstream dependency means a single upstream event reaches downstream as a system-wide event; distributed upstream dependency means the same downstream system absorbs the same upstream event as local noise. The fragility coupling runs from the concentration of the weight distribution and the joint failure probability of the top-weighted nodes, not from the binary presence of dependency.
What changes in a reader's view of a system is that "how robust is our supply?" resolves into a question about distribution shape rather than about whether a dependency exists or whether any single provider is reliable. Diversity at the upstream layer is purchased at one cost — more relationships to manage, higher unit cost, coordination overhead — and concentration is purchased at another — systemic exposure to upstream failure. The choice is structural, and the prime gives it a name and a measure, turning an implicit procurement, ecological, or platform-choice decision into an explicit trade between efficiency and tail exposure.
How would you explain it like I'm…
One Cow or Ten Cows
How Your Needing Is Spread
The Shape of Dependence
Structural Signature¶
a dependent system — a set of upstream providers — a weight on each provider — the distribution shape of those weights (concentrated vs spread) — the correlation structure of provider failures — downstream fragility coupled to top-k concentration and joint failure probability, not to the binary fact of dependency
The pattern is present when each of the following holds:
- A dependent system. A downstream system relies on an upstream network for a critical input or function.
- A provider set. Multiple upstream nodes could supply that dependency — suppliers, vendors, species, counterparties, pathways.
- A weight per provider. Each provider carries some share of the total dependency volume; the weights, not the mere existence of links, are the object of analysis.
- A distribution shape. The weights are distributed across providers either concentrated on a few nodes or spread broadly — and two systems with identical total dependency can be robust or brittle purely as a function of this shape.
- A failure-correlation structure. Provider failures may be independent or share common upstream modes; apparent diversification across providers with a shared upstream is fictitious.
- Fragility coupled to concentration. Downstream fragility scales with the concentration of weight on the top-k nodes and their joint failure probability — distinct from the binary fact of dependency and from any single provider's reliability.
These compose into a measurable structural property: a concentration scalar (Herfindahl, top-k share, Gini) plus a common-mode check determines tail exposure, so the operative question is not "how many providers?" but "how concentrated is the real weight, and do the providers share an upstream?" — with diversity decaying toward concentration by default unless actively maintained.
What It Is Not¶
- Not task interdependence.
task_interdependence(the embedding nearest neighbor) is the degree to which tasks must coordinate to complete work. This prime is about the distribution shape of a dependent system's reliance across upstream providers — a graph-weight property bounding fragility — not about how tightly tasks must mesh. - Not a distributional assumption.
distributional_assumptionis a modeling premise about a probability distribution. This prime concerns the actual concentration of dependency weight across providers (a measurable Herfindahl/top-k property), not an assumed statistical form. - Not risk pooling.
risk_poolingis variance reduction by aggregating independent risks. This prime measures whether dependency is concentrated or spread and whether the spread is genuine (no shared upstream); it names the structural property pooling tries to improve and the common-mode check that pooling ignores. - Not a margin of safety.
margin_of_safetyis headroom against worst-case load. This prime is about where the dependency weight sits, not how much buffer is held; a generous margin against any one provider's failure does not help if weight is concentrated on a node whose failure is system-wide. - Not single point of failure.
single_point_of_failureis the binary presence of one indispensable node. This prime is the continuous distribution-shape generalization — a concentration scalar over weighted providers — of which a single point of failure is the extreme (concentration = 1) case. - Common misclassification. Counting providers and declaring diversity. Twelve suppliers, one carrying ninety percent of volume — or all twelve fed by one upstream plant — looks diversified by count and is concentrated by weight. The tell: compute the concentration scalar over actual weights at the deepest shared layer, not the visible provider tally.
Broad Use¶
In supply chains, supplier concentration risk and single-sourcing trade-offs are the canonical case, with documented incidents in which a single facility or route carried a high share of global production. In information technology, single-vendor lock-in, single-CDN dependency, cloud-region concentration, and operating-system monoculture all amplify any single vulnerability or outage. In ecology, keystone-species dependence and single-pollinator reliance in industrial agriculture concentrate ecosystem function on a few nodes. In agriculture, monoculture and single-cultivar reliance — the Cavendish banana, the 1840s potato — concentrate the crop's survival on one genome. In data and information, single-archive storage and single-source data feeds concentrate access. In physiology, single-pathway metabolic dependence and single-receptor signaling concentrate function so that antagonizing one target collapses downstream behavior. In energy, single-fuel reliance and single-pipeline imports concentrate supply. In finance, single-counterparty concentration, single-CCP clearing, and too-big-to-fail banks act as concentrated upstream nodes in financial intermediation. In health systems, single-supplier active ingredients and single-manufacturer vaccine dependence concentrate the supply of care. In each case the dependent system's downstream fragility scales not with whether it depends on something but with how unevenly its dependency is spread — the same structural property under different unit names.
Clarity¶
The prime separates three structurally distinct questions that "how robust is our supply?" usually fuses. Is there a dependency? — the binary fact, handled by the relation of dependency. Is the upstream provider failure-prone? — a property of the provider, not the dependent system. How is dependency weight distributed across providers? — the property this prime names, where two systems with identical per-provider failure rates have different fragility purely as a function of the distribution. Holding these apart prevents conflating a reliable-provider question with a distribution-shape question.
It also separates concentration risk from redundancy. Redundancy is a remediation strategy; concentration is the failure mode the strategy addresses. They are not the same kind of thing: a system can have low concentration without intentional redundancy through organic diversity, and high redundancy can fail to reduce concentration if all redundant providers share a common upstream. And it surfaces the concentration- substitution trade: adding redundancy means accepting higher unit cost or coordination overhead, while accepting concentration means accepting systemic exposure, with the right operating point set by tail-cost asymmetry. The clarifying force is to make this trade visible and measurable rather than leaving it implicit in procurement, ecology, or platform-choice decisions, where the count of providers is routinely mistaken for genuine diversification.
Manages Complexity¶
The prime reduces a wide failure family — single-supplier collapses, monoculture pandemics, vendor lock-in incidents, keystone-species extinction cascades, single-CDN outages, counterparty defaults — to a single structural diagnostic: what fraction of your downstream depends on the top-k upstream nodes, and what is the joint failure probability of those k nodes? The same scalar — a concentration measure such as the Herfindahl-Hirschman index, top-k share, or Gini of provider weights — is portable across substrates, so a supply-chain analyst, an ecologist, a platform architect, and a financial regulator can all use the same number to compare systems and the same diagnostic to identify exposures.
The compression is sharpened by a fixed intervention set the diagnostic makes available: measure concentration as a routine system property; check for common modes, since diversification across providers with shared upstream inputs is fictitious; pre-qualify alternates, since the structural fix is having ready-to-activate alternatives rather than searching after the event; cost the tail by pricing the joint-failure event into the decision; maintain the forcing functions that prevent endogenous drift toward concentration; and stress-test the top-k, since their joint failure dominates the risk profile. Recognizing a fragility question as a distribution-concentration question thus yields one portable scalar and one portable playbook, far more compact than reasoning about each dependency network from first principles.
Abstract Reasoning¶
The clean abstract model has four primitives: a dependent system, a set of upstream providers, a weight on each provider, and a correlation structure of provider failures. From these several reusable inferences follow. Concentration is a system property, not a provider property: diversifying without checking common-mode upstream correlations yields apparent diversification with concentrated true dependency. Tail risk dominates expected loss: when losses are convex in provider unavailability, the concentration measure dominates the average reliability measure in determining tail outcomes. The "two of everything" heuristic is wrong by default: naive duplication does not deconcentrate if both copies share upstream common modes, since effective deconcentration requires independence of failure, a property of the joint distribution rather than the count.
Two further inferences concern trade-offs and dynamics. Concentration and efficiency trade: concentrated dependency is typically more cost-efficient per unit through volume discounts and learning curves, with the structural cost paid in tail exposure, so naming the trade lets a designer set the operating point intentionally. Endogenous concentration: networks under returns-to-scale drift toward concentration without active management, so if no forcing function maintains diversity, the structural state evolves toward fragility. These inferences follow from the graph-distribution structure alone, so they apply to a supply chain, an ecosystem, and a clearing system alike, and they tell an analyst that the binding risk is the joint failure of the top-weighted nodes, not the marginal reliability of any one of them, and that diversity decays by default unless actively maintained.
Knowledge Transfer¶
The transferable content is the four-primitive model together with the concentration scalar and the intervention playbook — measure, check common modes, pre-qualify alternates, cost the tail, maintain forcing functions, stress-test the top-k. Because the property is a pure graph-distribution property with substrate-neutral metrics, the moves carry across domains that named it separately. A procurement officer mapping a real dependency graph, a biodiversity scientist assessing pollinator reliance, a site reliability engineer designing a multi-region deployment, and a central-bank regulator stress-testing clearinghouses all recognize each other's playbooks despite the substrate gap, because the diagnostic — find the top-k weighted upstream nodes and their joint failure probability — and the remedy — seek genuinely independent alternatives — are identical.
The canonical transfer is the discovery of hidden common modes: a firm sourcing from twelve suppliers that all buy from one upstream plant has apparent twelve-provider diversity but real single-node concentration, and the identical structure recurs in a finance firm using twelve brokerages all clearing through one CCP, a web service with twelve regional deployments all on one DNS provider, and an agricultural region with twelve crops all pollinated by one managed bee population. The same intervention class applies in each: map the real dependency graph, find the top-k weighted nodes, deconcentrate by seeking genuinely independent alternatives, and price the tail-event cost into the operating decision. The endogenous-drift prediction is itself a load-bearing transfer, explaining why platforms consolidate, ecological diversity is lost under industrialized monocultures, clearing consolidates in finance, and critical-mineral supply chains drift to single-country concentration without active policy. The portable lesson is that the count of providers is not the measure of diversification — the weight distribution and its common modes are — so the right question is never "how many suppliers do we have?" but "how concentrated is the real dependency weight, and do the providers share an upstream?" — a question that travels intact from a procurement graph to a reef to a clearinghouse, and that, once asked, converts apparent diversification into a measured concentration risk.
Examples¶
Formal/abstract¶
The structure is a pure graph-distribution property, and a concentration scalar makes it precise. The dependent system draws a critical input from a provider set of n upstream nodes, each carrying a weight w_i (its share of total dependency volume, Σw_i = 1). The distribution shape is captured by the Herfindahl-Hirschman index, H = Σw_i² — which ranges from 1/n (perfectly spread) to 1 (one provider carries everything). The decisive demonstration is that two systems with identical total dependency and identical per-provider failure rates can have completely different fragility, governed entirely by H. Twelve providers at equal weight give H = 1/12 ≈ 0.083, and a single provider's failure removes only 1/12 of supply — local noise. One provider at weight 0.9 with eleven minor ones gives H ≈ 0.82, and that node's failure is a system-wide event. But the scalar alone is insufficient, which is the prime's sharpest formal point: H computed over the nominal provider list is fictitious if the providers share a common upstream mode. A firm sourcing from twelve suppliers (apparent H ≈ 0.083) that all buy from one upstream plant has a true concentration of H = 1 at the hidden node — the failure-correlation structure, not the surface count, sets fragility. The reusable inferences follow: tail risk dominates expected loss when losses are convex in provider unavailability, so H dominates average reliability in the tail; the "two of everything" heuristic is wrong by default, since naive duplication does not deconcentrate if both copies share a common mode; and concentration drifts upward endogenously under returns to scale unless a forcing function maintains diversity.
Mapped back: the weighted dependency graph instantiates every role — dependent system, weighted provider set, a distribution shape measured by H, and a common-mode correlation structure — making "measure H, then check for a shared upstream" the literal diagnostic, with the count of providers exposed as the wrong measure.
Applied/industry¶
A financial clearing system illustrates the hidden-common-mode case at systemic scale. The dependent system is the set of trading firms; the provider set is the brokerages and central counterparties (CCPs) through which trades clear; the weight on each is the share of clearing volume it handles. A firm that routes through twelve brokerages believes it has twelve-provider diversity (low nominal H) — but if all twelve clear through one CCP, the true concentration is a single node, and that CCP's failure is a system-wide event no amount of brokerage count mitigates. This is the prime's canonical transfer, structurally identical to a web service with twelve regional deployments all on one DNS provider, an agricultural region with twelve crops all pollinated by one managed bee population, and a manufacturer sourcing from twelve suppliers fed by one upstream plant. The fragility is coupled to the top-k concentration and joint failure probability, not to the count or to any single provider's reliability, so the too-big-to-fail CCP is a concentrated upstream node in financial intermediation exactly as a keystone species is in an ecosystem. The intervention playbook transfers intact: map the real dependency graph (trace clearing to the CCP layer, not just the brokerage layer); measure concentration as a routine property; check for common modes before trusting nominal diversification; pre-qualify alternates (ready-to-activate backup clearing); cost the tail by pricing the joint-failure event into the decision; and maintain forcing functions against the endogenous drift by which clearing consolidates without active policy — the same drift that consolidates platforms and erodes ecological diversity.
Mapped back: the clearing system is dependency-distribution concentration — trading firms depending on a weighted set of clearing providers whose nominal diversity hides a single-CCP common mode — so "map the real graph, find the top-k, deconcentrate toward genuinely independent alternatives" is the same structural move as in supply chains and ecosystems.
Structural Tensions¶
T1 — Nominal Count versus True Weight Distribution (measurement). The prime's central insight is that the count of providers is not the measure of diversification — the weight distribution is, captured by a concentration scalar (Herfindahl, top-k share, Gini). The failure mode is counting providers and declaring diversity: twelve suppliers, one of which carries ninety percent of volume, looks diversified by count and is concentrated by weight. Diagnostic: compute the concentration scalar over actual dependency weights, not the provider tally. The prime relocates the question from "how many?" to "how concentrated is the real weight?"; a procurement or architecture decision that reasons about provider count has measured the wrong quantity and may have bought fragility while believing it bought robustness.
T2 — Nominal Independence versus Hidden Common Mode (coupling). Even a genuinely spread weight distribution is fictitious if the providers share a common upstream — twelve suppliers all fed by one plant, twelve brokerages all clearing through one CCP, twelve regions all on one DNS provider. The failure mode is trusting surface diversification while the true concentration sits at a hidden node the analysis never traced. Diagnostic: trace each provider's own upstream dependencies before crediting independence; the concentration scalar must be computed at the deepest shared layer, not the visible one. The prime's sharpest formal point is that failure-correlation structure, not surface count, sets fragility — and the most dangerous concentrations are the ones that look like diversity until the shared mode fails.
T3 — Concentration Cost versus Efficiency Gain (sign/direction). Concentration and efficiency trade against each other: concentrated dependency is usually more cost-efficient per unit (volume discounts, learning curves), with the structural cost paid only in tail exposure. The failure mode is optimizing the visible efficiency and silently accumulating tail risk, because the efficiency gain is realized every day while the concentration cost is paid only at the rare joint-failure event. Diagnostic: price the tail event into the operating decision rather than comparing only unit costs. The prime makes the trade explicit and measurable; a decision that chases per-unit efficiency without costing the tail is optimizing the term that shows up in every quarter against the term that bankrupts the system once.
T4 — Static Concentration versus Endogenous Drift (temporal). Concentration is not a fixed property to measure once — networks under returns-to-scale drift toward concentration by default unless a forcing function actively maintains diversity. The failure mode is a one-time deconcentration that silently re-concentrates: a diversified supplier base consolidates as the cheapest vendor wins more share, a multi-region deployment drifts onto one provider as it grows. Diagnostic: ask what forcing function maintains the diversity, and re-measure concentration periodically rather than trusting the original design. The prime predicts that diversity decays toward fragility unless held open; treating a past deconcentration as permanent ignores the endogenous pull that erodes it.
T5 — Redundancy versus Genuine Independence (scopal). The prime separates redundancy (a remediation strategy) from concentration (the failure mode it addresses), and warns the "two of everything" heuristic is wrong by default — naive duplication does not deconcentrate if both copies share a common mode. The failure mode is buying redundancy that fails together: a hot standby in the same data center, a backup supplier with the same upstream, a second CCP that depends on the first's settlement bank. Diagnostic: verify that redundant providers have independent failure distributions, not merely separate identities. The prime's deconcentration requires independence of failure — a property of the joint distribution — so redundancy counted without an independence check adds cost and coordination overhead while leaving the true concentration untouched.
T6 — Concentration Risk versus Provider Reliability (scopal). The prime carefully separates the distribution-shape question (how is weight spread?) from the provider-reliability question (is any single provider failure-prone?) — two systems with identical per-provider reliability can have completely different fragility. The failure mode is conflating them: vetting each provider's reliability exhaustively while ignoring that the weight is concentrated, so a perfectly reliable top node still makes a single failure a system-wide event. Diagnostic: hold the two questions apart — assess concentration as a system property and reliability as a provider property, separately. The prime's leverage is that fragility couples to top-k concentration and joint failure probability, not to the marginal reliability of any one node; auditing reliability alone can certify a system as robust precisely while its concentration makes it brittle.
Structural–Framed Character¶
Dependency distribution concentration sits at the structural pole of the structural–framed spectrum — aggregate 0.0, every diagnostic structural. It is a pure graph-distribution property: how a system's dependency weight is spread across providers — concentrated or diffuse — measured by a substrate-neutral scalar (Herfindahl, top-k share, Gini) that bounds fragility independent of the system's own defenses. Nothing about it depends on a particular substrate's vocabulary or values.
Every diagnostic points one way. The pattern carries no home vocabulary that must travel: "concentration," "provider," "weight," "top-k" are the same metrics whether the providers are suppliers in a supply chain, pollinators in an ecosystem, cloud platforms under an IT stack, or counterparties in a financial network, each told in its own field's terms. It carries no evaluative weight: concentration is neither good nor bad in itself — it lowers cost and raises fragility, a value-neutral structural fact whose desirability depends entirely on what the system needs. Its origin is formal — a weighted-graph distribution property statable with no institutional content. It is not human-practice-bound: a single keystone species pollinating most of a plant community instantiates the same concentration as a sole-source supplier, with no human present. And to invoke it is to recognize a distribution shape already present in the dependency graph — a fact one measures, not an interpretation one imposes. On every diagnostic it reads structural, matching the all-zero aggregate, and its uniform spread across physical, biological, and social substrates confirms it.
Substrate Independence¶
Dependency-distribution concentration is a maximally substrate-independent prime — composite 5 / 5 on the substrate-independence scale. The signature — downstream fragility set by the shape of the dependency distribution, scaling with concentration on a few upstream nodes rather than with the mere existence of dependence — is recognized, not translated, across substrates that share no other vocabulary: supply chains (single-sourcing and supplier-concentration risk, with documented single-facility chokepoints), information technology (single-vendor lock-in, single-CDN and cloud-region concentration, OS monoculture), ecology (keystone-species and single-pollinator reliance), agriculture (monoculture — the Cavendish banana, the 1840s potato), data systems (single-archive storage), physiology (single-pathway metabolic dependence, single-receptor signaling), energy (single-fuel, single-pipeline reliance), finance (single-counterparty and single-CCP concentration, too-big-to-fail), and health systems (single-supplier active ingredients). That breadth across biological, physical, social, and engineered media earns the full domain score. Structural abstraction is maximal because the load-bearing element is a pure distributional property — the concentration of a dependency distribution — carrying no domain-specific commitments. The transfer evidence sits at 4 rather than 5: the same concentration-drives-fragility structure is documented across all these substrates, but it travels as a shared structural property recognized field by field rather than as one closed-form model applied verbatim, so the abstraction and composite reach 5 while documented transfer stays strong at 4.
- Composite substrate independence — 5 / 5
- Domain breadth — 5 / 5
- Structural abstraction — 5 / 5
- Transfer evidence — 4 / 5
Relationships to Other Primes¶
Parents (1) — more general patterns this builds on
-
Dependency Distribution Concentration presupposes Dependency
How a system's dependency WEIGHT is distributed across providers; it presupposes a dependency structure and characterizes the shape of that reliance (a graph-weight property).
Path to root: Dependency Distribution Concentration → Dependency
Neighborhood in Abstraction Space¶
Dependency Distribution Concentration sits in a moderately populated region (40th percentile for distinctiveness): it has near-neighbors but no dense thicket of synonyms.
Family — Overextension & Load Fragility (18 primes)
Nearest neighbors
- Correlated Capacity Demand — 0.75
- Defense In Depth — 0.73
- Single Point of Failure — 0.73
- Dependency — 0.72
- Partition Dependence of Aggregates — 0.70
Computed from structural-signature embeddings · 2026-06-14
Not to Be Confused With¶
The cleanest contrast is with single_point_of_failure, because the two are genuinely related as a continuous measure to its binary extreme — and the relationship is precisely why they must not be merged. single_point_of_failure is the binary property that some one node is indispensable: if it fails, the system fails, full stop. Dependency distribution concentration is the continuous distribution-shape generalization: it measures, via a concentration scalar (Herfindahl, top-k share, Gini) over weighted providers, how unevenly dependency weight is spread, of which a single point of failure is the limiting case where concentration equals one. The generalization is load-bearing, not cosmetic, because most real fragility lives in the intermediate regime that the binary frame cannot represent. A system with one provider at ninety percent weight and eleven minor ones has no strict single point of failure — losing the big provider leaves ten percent of supply, so the system technically survives — yet it is nearly as brittle as a true single point of failure, and the concentration scalar captures this where the binary predicate reports "no SPOF, we're fine." Conversely, the binary frame can fire on a node that carries trivial weight (indispensable but tiny), overstating its systemic importance. The practical consequence is that single_point_of_failure analysis tends toward a checklist ("is any node strictly indispensable?") that misses graded concentration, while this prime forces a measurement ("how concentrated is the real weight, and what is the joint failure probability of the top-k?"). Treating the two as identical leads a designer to certify a dangerously concentrated system as safe because no node is strictly indispensable — the exact blind spot the continuous measure exists to close.
A second genuine confusion is with risk_pooling, since both concern spreading reliance across many providers and both are invoked in the name of robustness. The difference is that risk_pooling is a strategy (aggregate independent risks to reduce variance) while this prime is the structural property that determines whether that strategy actually worked. Risk pooling assumes the pooled units fail independently; its variance reduction is real only under that assumption. Dependency distribution concentration supplies exactly the check risk pooling omits: the common-mode test that asks whether the nominally-diverse providers share a hidden upstream. The prime's sharpest formal point is that a Herfindahl computed over a nominal provider list is fictitious if those providers share an upstream — twelve suppliers all fed by one plant, twelve brokerages all clearing through one CCP have true concentration of one despite an apparent spread. Risk pooling, applied naively, would credit the twelve-provider count as twelve-fold diversification; this prime exposes that the effective diversification is nil because the failures are perfectly correlated at the shared node. So where risk_pooling says "spread across many to reduce variance," this prime says "spreading reduces variance only if the failures are genuinely independent, and you must measure the concentration at the deepest shared layer to know." The confusion is dangerous because it licenses the "two of everything" heuristic that the prime explicitly flags as wrong by default: duplication that shares a common mode adds cost and coordination overhead while leaving the true concentration untouched.
For a practitioner the distinctions sort the analysis of any dependency-fragility question. Use single_point_of_failure only as the extreme flag it is, and reach for this prime's continuous concentration measure to capture the graded middle the binary predicate misses. Treat risk_pooling as the strategy and this prime as the verification of whether the pooling delivered genuine independence, via the common-mode check at the deepest shared layer. The prime's unique contribution is the insistence that the count of providers is never the measure of diversification — the weight distribution and its hidden common modes are — and that diversity decays endogenously toward concentration unless a forcing function holds it open.
Solution Archetypes¶
No catalogued solution archetypes reference this prime yet.