Fleet Availability and Model Propagation

How multiple units and shared learning lift facility-level availability toward - though not to single-unit Tier III - the honest target.

The fleet is the resiliency unit

A single machine cannot reach Tier III uptime; a coordinated fleet can approach the target at the facility level through redundancy across units. The breeder follows a FOAK-NOAK-BOAK sequence from ~2030, and burner housings (Aegis / MetroVolt) are intended to deploy as multiple units. Fleet resiliency has two engines: N+k unit redundancy and cross-unit model propagation.

N+k across units

If one unit's availability is A_unit, a facility that needs k units running out of n installed tolerates n-k simultaneous outages. The facility availability rises well above A_unit, in exchange for installed redundancy. This is how a MetroVolt facility could approach a firm-power obligation despite each unit sitting in the 0.86-0.995 envelope - but the honest statement remains that no single unit is a sole Tier-III source.

from math import comb
def k_of_n(A_unit, n, k):
    # prob at least k of n units are up (independent)
    return sum(comb(n, i) * A_unit**i * (1-A_unit)**(n-i) for i in range(k, n+1))

print(k_of_n(0.95, n=4, k=3))   # facility-level uplift over a single 0.95 unit

Model propagation

Each unit generates operating data that refines RUL models, disruption predictors, and anomaly detectors. Validated models propagate to the rest of the fleet so unit N+1 starts with unit N's learned reliability rather than from scratch. This is why each NOAK/BOAK breeder is expected to be more available than the FOAK, and why fleet MTBF estimates improve over calendar time.

Propagation is governed: a model is promoted across the fleet only after validation on its origin unit, never silently. Independence of unit failures is checked - shared fuel supply, shared grid tie, or a common design flaw can correlate outages and erode the k-of-n benefit. See the availability gate and availability modeling.