agriculture computer-science engineering-and-technology

​

Common-mode failure

​

Description

Common-mode failure is the collapse of nominal independence. A system has several replicas, backups, or protective layers and therefore appears able to survive the loss of any one. But those protections share something load-bearing — a power supply, physical location, design assumption, genetic trait, specification, operator, or environmental tolerance. One cause reaches that shared feature and disables several protections together. The diagnostic is not simply “how many backups are there?” but “along which dimensions are their failure domains actually different?” Two servers in one availability zone, three programs derived from one ambiguous specification, and thousands of distinct crop hybrids carrying the same susceptible cytoplasm can each be numerous without being independent. Counting components overstates reliability when the relevant probability is conditioned on their shared exposure. Reliability literature often distinguishes a common cause (the shared initiating condition) from a common mode (the correlated way multiple components become unavailable). The catalog keeps the broader, transferable structure under one concept: the reliability case assumes independent protection, while a shared dependency or exposure couples the failures.

​

Triggers

User-initiated: The user says “we have multiple backups,” “it would take everything failing at once,” or “these were developed independently,” without identifying independent failure domains. Vocabulary cues include “shared dependency,” “same region,” “common cause,” “correlated failures,” and “all the backups failed.” Agent-initiated: The agent sees a reliability argument multiply protection counts while leaving substrate, location, specification, or vulnerability shared. Candidate inference: “What single condition could reach all of these protections, and which proposed backup changes that condition rather than merely adding another copy?” Situation-shape signals: Multiple safety layers colocated below the same flood line; replicas sharing power, identity, deployment, or control planes; organizationally separate reviewers using the same flawed requirement; portfolio or biological diversity whose members share the same underlying exposure.

​

Exclusions

• Only one component fails — a common-mode failure requires one cause to defeat multiple nominally independent protections or alternatives. An ordinary component failure may reveal weak redundancy, but it is not itself the shared-failure shape.
• Failures are genuinely independent — two components can happen to fail near the same time without a shared dependency, exposure, or vulnerability. Temporal coincidence alone does not establish a common mode.
• Failure propagates sequentially — in a cascade, one component’s failure becomes the next component’s cause. Common-mode failure instead has one upstream cause reach several protections through a shared exposure, though the two shapes can co-occur.
• The components were never meant to provide the same protection — several affected systems do not constitute defeated redundancy unless they were alternatives or layers protecting a shared function.

​

Structure

1. Protected function — the property the system must preserve.
2. Nominally independent protections — multiple components, replicas, or layers intended to preserve it.
3. Shared dependency or exposure — a feature present across those protections.
4. Common cause — a condition that reaches the shared feature.
5. Coupled failure — the protections become unavailable together, invalidating the assumed reliability multiplication.

​

Nearest-neighbor review

common-mode-failure is not an alias for redundancy: redundancy names the robustness investment, while this concept names the hidden correlation that nullifies it and projects a different action — diversify failure domains rather than add copies. It is not bulkhead, which is the boundary-forming remedy, or defense-in-depth, which is the layering strategy. It also differs from cascade and contagion: those move failure from node to node, while a common mode lets one cause reach several nodes through a shared exposure. The new entry therefore fills a repeatedly named exclusion in the existing reliability cluster rather than splitting one of its concepts by vocabulary alone.

​

Relationships

• redundancy — common-mode failure is the principal falsifier of a redundancy claim: component count does not buy robustness when failure domains remain coupled.
• bulkhead — bulkheads deliberately separate failure domains; a common mode is a path that crosses or bypasses those boundaries.
• cascade — a cascade passes failure from one element to the next, while a common mode fans one cause into several nominal protections.
• defense-in-depth — depth is real only when successive defensive layers do not inherit the same defeat condition.
• quietly-load-bearing — the shared dependency is commonly invisible during normal operation and becomes legible only under the rare disturbance the protections were built to survive.

​

Examples