Where does consensus collide?

In distributed computing, consensus is the problem of getting a set of separate processes, some of which may crash or send faulty messages, to agree on a single value even though they run at their own speed and can only communicate by passing messages. Fischer, Lynch, and Paterson's 1985 Impossibility of distributed consensus with one faulty process proved that in a fully asynchronous system no protocol can guarantee agreement if even one process may fail, a result that shaped later designs such as Paxos and Raft. Here consensus is a precisely stated coordination task with defined success conditions, not a state of public agreement among people, a summary of biological trees, or a poll of experts.[1]

Unit

One agreed value chosen by a set of processes despite faults

Recovery means

A run either terminates in agreement or, under asynchrony and a fault, may fail to terminate at all

Measured by

Judged by whether all non-faulty processes decide the same valid value and whether every run eventually terminates

In political theory, consensus is broad agreement reached within a group or a whole society, especially agreement on principles that people holding very different worldviews can each accept for their own reasons. John Rawls's 1987 The Idea of an Overlapping Consensus described how citizens with conflicting moral and religious doctrines might still converge on shared principles of justice, giving a stable footing for a plural society. This consensus is a condition of human agreement about how to live together, argued for in normative terms, and it is unlike a coordination protocol among machines, a computed summary of candidate trees, or a structured survey of specialists.[2]

Unit

A body of shared agreement on principles within a group or society

Recovery means

Not applicable

Measured by

Assessed by argument and by observing whether groups with differing views endorse the same principles

In phylogenetics and systematics, a consensus tree is a single branching diagram that summarizes what many candidate evolutionary trees agree on, keeping the groupings supported across the set and collapsing the ones that are disputed. Margush and McMorris's 1981 Consensus n-trees formalized the majority-rule consensus tree, which retains every grouping present in more than half of the input trees, alongside the stricter rule that keeps only groupings present in all of them. The referent is a computed summary of a collection of diagrams, scored by how much agreement it captures, and it has nothing to do with processes agreeing on a value, citizens sharing principles, or experts converging in a survey.[3]

Unit

One summary tree distilled from a set of candidate trees

Recovery means

Not applicable

Measured by

Built by a rule such as strict or majority-rule agreement, and read by the support behind each retained grouping

In biomedicine and forecasting, expert consensus is formalized agreement drawn from a panel of specialists, often through a structured, multi-round procedure that feeds the group's responses back to it until opinions converge enough to state a shared position. Dalkey and Helmer's 1963 An Experimental Application of the Delphi Method to the Use of Experts introduced the Delphi method, using repeated anonymous questionnaires with controlled feedback to draw a reliable opinion out of a group. This consensus is a managed outcome of a deliberate elicitation among people, reported as guidelines or panel statements, and it is not a machine coordination task, a normative political agreement, or a tree summarizing biological data.[4]

Unit

An agreed panel position produced by a structured elicitation

Recovery means

Not applicable

Measured by

Operationalized by round-to-round agreement thresholds, such as a set share of panellists endorsing a statement