Inside the Denver control room: monitoring 480 plants at 3 a.m.

At 3 a.m. the room is quiet in the way an aircraft cockpit is quiet — attentive, not idle. Six wall screens carry the live state of every plant we operate: 480 sites across 22 states, 1.9 GW of contracted capacity, roughly 3.4 GWh flowing through the system on a good production day. The night shift is three people. On the desk in front of them, a single number sits in the corner of the leftmost screen: fleet availability, 98.2%. Holding it there is the entire job.

Most of what the room does is triage. Alarms arrive constantly, and the skill is not reacting to all of them but sorting the noise from the signal within seconds. A tripped string on a 6 MWp rooftop in Fresno at midday would be urgent. The same alarm at 3 a.m., when the array is dark, is a note for the 7 a.m. handover. We built the alert tiers so the overnight team escalates only what genuinely cannot wait: a combiner-box temperature climbing on a site with battery storage, an inverter that has failed to restart after a grid event, a met-station reading that has quietly flatlined and is about to corrupt the next day’s forecast.

What the night watches for

The real overnight work is the storage fleet and the grid itself. About 140 of our sites pair PV with batteries, and those keep working after sundown — dispatching into evening peaks, absorbing frequency events, charging against tomorrow’s price curve. When PJM or ERCOT sends a signal, the response has to land in under two seconds, and someone has to confirm it did. The team also watches weather moving east: a hailcore over the Texas Panhandle at 2 a.m. is the cue to stow trackers at nine plants before it arrives, not after.

The instrument we trust most is not any single alarm but the drift between measured and expected output. Every site reports a performance ratio, and the overnight report flags any plant that has slipped more than 3% below its modelled yield over the trailing week. That gap is how we catch the slow failures — soiling on a Nevada site that has not seen rain since April, a tracker row frozen at the wrong angle, a transformer running two degrees warmer each night than the one beside it. None of these trip an alarm. All of them cost megawatt-hours.

By 6 a.m. the eastern sites are waking. Inverters in the Carolinas begin reporting, ramping from zero as the sun clears the horizon, and the map turns from grey to green from right to left across the wall. The night team writes the handover: eleven tickets raised, three dispatched to field crews, one inverter swap scheduled in Georgia for Thursday. The 98.2% held through the night.

People assume monitoring 480 plants is about watching everything at once. It is closer to the opposite. It is knowing, out of ten thousand data points, which forty matter tonight — and having the discipline to ignore the rest until morning.

Read More

94% on-time energization: how we report schedule certainty

Every quarter we publish a single number our clients care about more than almost any other: the share of projects we energize on or before the contracted commercial-operation date. Across the 480 projects we have delivered in 18 years, that figure now sits at 94 percent. We report it because a promise about a date is only worth as much as the record behind it, and because the years it drops tell you more than the years it holds.

The number is not our own arithmetic. Since 2019 we have had it verified annually by an independent engineering auditor, who reconciles each project’s signed COD against the utility’s interconnection service agreement and the actual first-megawatt timestamp from the meter. A project counts as on-time only if it delivered power to the grid on or before the date in the contract — not a revised date, not a milestone we renegotiated midway. Partial energization does not count. Extensions we requested do not count in our favor.

What the dips actually taught us

In 2021 the figure fell to 87 percent. Two projects in the Texas Panhandle — a 74 MWac plant near Amarillo and a 51 MWac site outside Lubbock — slipped by nine and fourteen weeks when a transformer supplier’s lead time went from 40 to 68 weeks with no warning. We could have quietly rebaselined and kept the headline at 94. Instead we published 87, wrote up the root cause, and changed how we buy: long-lead switchgear and main power transformers are now ordered against a signed EPC contract rather than at notice-to-proceed, which moved procurement roughly eleven weeks earlier in the schedule.

That single change is why 2022 and 2023 came back to 95 and 96 percent respectively, across a combined 61 projects and just over 2.1 GWp of capacity. The recovery was not luck; it was the direct output of having told the truth about a bad year and then spending money to fix the cause.

The metric also shapes what we decline. When a developer asks us to commit to a date that our schedule model puts below an 80 percent confidence band, we say so in writing and offer the date we can actually hold. We have walked away from roughly 30 MW of work a year on those grounds. It costs us revenue and it protects the 94.

If you are evaluating an EPC, ask for the audited on-time figure, the definition behind it, and the worst year in the record. A partner willing to show you the dip — and what they did next — is telling you how they will behave when your project is the one under pressure. That is the whole reason we keep publishing it, unflattering quarters included.

Read More
Buy Theme - $39 Launch It For Me