The cost of
another validator.

A larger committee can spread responsibility while consuming the resources needed to protect its members. The useful question is what the next participant contributes after the cost of coordinating and defending the system has been counted.

My research on The Decentralisation Threshold examines that question through a conditional economic model. Its public supplementary appendix, deposited in May 2026, separates an exact integer calculation from a smooth approximation. That distinction matters: an elegant curve can explain a mechanism without selecting the same committee as the underlying calculation.

Give the budget somewhere to go

Consider a fixed budget divided between coordination and protection. In the appendix’s illustrative quadratic coordination model, a budget of 100 units and a coordination coefficient of 0.005 leave 87.5 units after coordinating 50 validators. Dividing that residual equally gives 1.75 units per validator. At 100 validators, coordination consumes 50 units and protection receives 0.50 units per validator.

These are model units, not measured expenditure from a deployed network. The arithmetic nevertheless exposes the question that a headcount conceals. Which part of the budget expands when the committee expands? If the total stays fixed, adding participants changes more than the number of signatures.

The answer depends on the protocol and implementation. Quadratic coordination is a stated cost specification. It cannot be assigned to every system simply because its participants are called validators. A useful application must measure the relevant communication, verification and operational costs.

The threshold moves in steps

The independent-compromise baseline gives each validator the same compromise probability and counts how many are compromised. The model’s adverse event occurs when that count reaches a specified fraction of the committee, rounded up to a whole number.

With a one-third threshold, 84 validators require 28 compromised members to reach the threshold. At 85 validators, the threshold rises to 29. At 86 it remains 29. The additional member therefore changes the two comparisons differently. When the threshold stays fixed, there is another opportunity for a compromise without an increase in the count required. When it rises, an additional compromise is required.

This produces a sawtooth in the exact probability. The appendix derives the two directions by adding one independent Bernoulli trial to the existing count. It is a small calculation with a large interpretive consequence: a smooth trend does not describe every one-member change.

A bound answers a different calculation

The appendix also uses a Chernoff upper bound on the adverse-event probability. Subtracting that bound from one gives a conservative smooth measure of protection. Under the stated curvature and endpoint conditions, subtracting increasing convex costs produces a unique interior optimum.

The exact integer problem has a finite maximiser because its benefit is bounded while its modelled cost grows without bound. That existence argument does not itself establish uniqueness or a smooth decline at every integer beyond the best committee. Those are different conclusions, and the staircase is a reason to keep them separate.

For the appendix’s illustrative fixed-probability parameters, direct enumeration selects 85 validators. The corresponding smooth calculation selects approximately 122.8. I rechecked those two calculations for this article using the stated equations. This is a numerical check of the illustrative model, rather than an estimate of the appropriate size of any operating network.

The difference is useful. The approximation makes the trade-off easy to study; the exact calculation evaluates the choices that can actually be made inside that model. Treating the approximation as an operational prescription would discard information already available at negligible computational cost.

Measure the protection being purchased

Applying the framework requires more than choosing a committee size from a table. The compromise probability, dependence between operators, coordination costs and value assigned to avoided failure all need evidence. Shared software, infrastructure or control can also undermine an independence assumption even when the machines have different labels.

A practical research comparison would hold the service requirement and threat scenario fixed, document those inputs, and report how the preferred committee changes as uncertain inputs vary. It would distinguish a modelled threshold event from demonstrated failure of a particular protocol, whose timing and execution assumptions also matter.

The contribution of this framework is a disciplined question about marginal resources. What extra protection does another participant buy, what does it consume, and how sensitive is the answer to the assumptions? Counting the participants is the beginning of that investigation.