Reliability run

What the run demonstrates that nothing else does

Every test before it is bounded. The factory test exercises equipment against its specification, the site acceptance test checks the installation against codes and installation instructions, the capacity test integrates one discharge, the P-Q test walks a matrix of operating points. None of them runs the plant for a week.

The reliability run does, and what it catches is the class of fault that only appears with time in it: a thermal loop that trips on the fourth consecutive hot afternoon, a communications path that drops once a day, a firmware watchdog that reboots a converter on a schedule nobody documented, a spares process that takes eleven hours to get a technician through the gate with the right board.

The metric is availability over the window, computed the way the contract says — normally available hours over period hours less excluded hours, the same shape the operating-year guarantee uses. What differs is everything around the formula. The window is short, the plant is new, and the exclusions list is often not the annex negotiated for the operating years. Two documents can both say 97 percent and impose materially different obligations.

What counts as an interruption

The definition has to answer three questions, and money moves on each. What capability was lost: a full trip is unambiguous, a partial derate is not, so the run needs the same threshold rule availability needs — above some fraction of guaranteed capability a derated plant counts as available, below it as unavailable — or capacity-weighted proration that scores the derate pro rata. One power block out of ten, under a time-based metric with a permissive threshold, can score near 100 percent while the plant is visibly short.

Whose fault: grid outage, curtailment instruction, owner-caused delay and force majeure are the usual carve-outs, and a broad phrase about events beyond supplier control, with no enumerated list behind it, will be read generously by the party that drafted it. And how long: a protection trip that auto-recloses in forty seconds and a container down for a day are both interruptions, but a run scored on one-minute SCADA resolution and one scored on hourly averages will not agree about the first.

The fourth question is whether the plant was ever asked to do anything. A window in which no dispatch instruction arrived proves the plant sat there without tripping, which is standby rather than response. A defensible procedure requires exercised charge and discharge, a minimum number of full-power transitions in both directions, and a state-of-charge readiness rule — an asset parked full is perfectly available on paper and physically unable to absorb when the instruction says charge.

Reset, pause, or count

Three treatments exist for any event, and the contract has to assign one to every category. It can pause the clock and resume where it stopped, which is what excluded events normally do. It can count against the threshold and let the run continue, which is what ordinary faults do inside the allowance. Or it can restart the run from zero, the sanction reserved for events the owner will not accept at all — a safety-system actuation, a fire-alarm activation, a trip traced to a defect in a scope the supplier owns.

A loosely written reset trigger is what puts COD out of reach. On a new plant working through infant mortality, a broadly written reset trigger can loop the run indefinitely while the guaranteed COD passes and delay damages accrue daily — which is exactly when both parties stop fixing the plant and start litigating the classification of each event.

The durable drafting is what works for the annual metric too: an enumerated closed list of what resets, one named data source at a stated resolution, and a cap on how many restarts the parties will tolerate before they have to agree on something else.

Why the commissioning threshold is not the annual guarantee

Convert both to hours and the gap is obvious. A 14-day run is 336 period hours, so one point of availability is about 3.4 hours and a 97 percent threshold allows roughly 10 hours of unexcused downtime for the entire window. The operating-year guarantee at the same 97 percent allows about 263 hours; 98 percent allows about 175 and 95 percent about 438. A single eight-hour repair is a rounding error against the annual figure and most of the whole allowance against the two-week one. Short windows punish clustered events, and a commissioning window is the most clustered period in the asset's life.

The populations differ as much as the arithmetic. The run scores a plant in its infant-mortality period, with punch-list work still open and a crew still learning it; the guarantee scores a mature plant over a year, against a negotiated exclusions annex, frequently with a reduced or ramped year-one number precisely because holding a brand-new site to steady-state figures produces a dispute rather than a well-run plant. Carrying an argument from one into the other is a category error. Passing the run is evidence the plant works. It is not evidence the annual guarantee will hold.

Running it on a plant that is not finished

The most common dispute in this test is the denominator. If a block is not yet energised, if the collection system is only partly available, if the ISO has not opened a market-qualification slot, or if the site is still on a temporary auxiliary feed, then available has to be measured against something — full contracted capability, or the capability that physically exists. Say which, in writing, before the window opens. On a phased site the answer is usually per block, matching the partial-COD mechanics the offtake already needs, with the threshold and the reset rules applied block by block.

The second dispute is who may touch the plant. If commissioning work continues through the run, the supplier's own crew generates the interruptions the supplier is being measured on, and each one becomes a classification argument. Freeze the configuration, log every intervention with its authorisation, and decide in advance whether a firmware push during the window is a reset event.

Note where the run sits legally, too: under the FERC pro forma Large Generator Interconnection Agreement, on-site test operations and commissioning before Commercial Operation are Trial Operation, and electricity generated during Trial Operation is expressly excluded from commercial output — so the run is executed on commissioning energy, under whatever arrangement the interconnection agreement and the offtake made for it.