This is a case study in solar commissioning risk: how an 80 MW solar EPC project 19 days from its contractual COD turned a €2 million liquidated damages exposure into a €440,000 outcome by making its real blockers visible.
The project director sent me a message on a Sunday evening. Short. No punctuation, which told me everything.
“we have a problem. call me”
It was the final stretch of an 80 MW solar EPC in southern Spain, with contractual COD 19 days away. The PPA liquidated damages clause: €55,000 per day past COD. And the commissioning schedule, when I saw it, had 218 open punch list items, a protection relay sequencing failure that had already pushed back the utility acceptance test twice, and a fire suppression certification backlog of 6 weeks.
We had 19 days before the LD clock started. We were tracking to hit it at least 37 days late.
That is €2.035 million in potential liquidated damages.
This is the story of what we did.
Week 1: The diagnosis
The first thing I did was stop the weekly status meeting. Not because it was not useful, but because the information in that meeting was already 5 days old by the time it reached the table. We were making decisions based on Tuesday’s reality on the following Monday.
We deployed ClearSpot’s Commissioning Risk Intelligence agent over the project data on day three. Within six hours, it had processed the punch list, the commissioning test records, the protection relay correspondence, and the fire suppression certification status.
The output was not a summary. It was a ranked risk register with three priority categories:
- COD-blocking (14 items): These had to close before the utility acceptance test could be rescheduled. Three were assigned to subcontractors who had demobilised. Four required parts on a 12-day lead time. Seven were documentation items, including missing certificates, incorrect format submissions, and one MV protection relay settings document that had been submitted twice with different values.
- COD-threatening (31 items): Items that were not immediately blocking but would become blocking if not closed within 7 days based on current velocity.
- Trailing (173 items): Everything else, important but not LD-relevant in the immediate term.
This categorisation sounds obvious. But nobody on the project had it. They had a list of 218 items. No one knew which 14 were the ones that would cost €2 million.
Week 2: The protection relay problem
The protection relay sequencing failure was the most technically complex COD-blocker.
The grid operator in this region had updated their relay commissioning requirements in Q4 of the previous year. The project’s relay settings, designed against the previous version, had been rejected twice by the utility’s commissioning acceptance team. The rejection letters were in the project email archive, but nobody had fully decoded what the specific incompatibility was.
The Commissioning Risk Intelligence agent cross-referenced the two rejection letters against the current utility commissioning specification. The incompatibility: the directional earth fault relay function had to be tested in a specific sequence relative to the differential protection relay, a sequencing requirement added in the new specification that the project’s commissioning engineer had not been briefed on.
That finding, which a human expert would have needed 2 to 3 days to reproduce from the documentation, came back in 40 minutes.
The commissioning engineer updated the test sequence. We requested a new utility acceptance date. Got it in 11 days.
Week 3: The fire suppression sprint
The fire suppression certification backlog was a logistics problem, not a technical one. The accredited inspector had a 6-week booking queue. We were already past the point where conventional scheduling could fix it.
The agent identified three other accredited inspectors in the relevant region with shorter availability, and one had a cancellation slot 9 days out. We booked it. The fire suppression test passed first attempt because the Commissioning Risk Intelligence agent had also flagged, two days before the inspection, a valve configuration issue that would have caused a functional test failure.
A first-attempt test failure would have pushed us back another 3 to 4 weeks while the valve was reconfigured and reinspected.
The drone validation sprint
Parallel to the protection relay and fire suppression work, we ran drone thermal surveys over the entire installed area over three days. ClearSpot’s agents and drones work as an integrated multi-agent system: agents see what drones see, and drones execute what agents decide.
The thermal survey generated 47 punch list items that the manual walkthrough programme had not caught.
Of those 47 items, 6 were in the COD-blocking category. Two were hot junction boxes in the MV string wiring that would have caused a test failure under load. They were caught and repaired before the utility acceptance test was rescheduled.
If we had not run those drone surveys and gone straight to the utility acceptance test, we would have failed. That failure would have added another 2 to 3 weeks, which means €770,000 to €1.155 million in additional LD exposure.
The outcome
COD was achieved 8 days after the original contractual date. LDs: 8 × €55,000 = €440,000.
That number is real, and I am not going to pretend otherwise. 8 days of LDs is not a win in absolute terms.
But against the trajectory we were on at that Sunday call, 37+ days late and €2.035 million in projected LDs, it is a €1.595 million recovery.
The Commissioning Risk Intelligence agent cost €18,000 per year. The recovery in a single event: 88 times that cost.
For broader context on commissioning quality and O&M performance standards, the IEA PVPS O&M guidelines and the NLR solar O&M analysis both reinforce the direct link between commissioning rigour and long-term plant performance. SolarPower Europe continues to track how schedule and LD risk is reshaping EPC contracting structures across the region.
What I learned
The problem was not that the project team was incompetent. They were good. The problem was that commissioning risk is information-dense and time-compressed, and no human team can process 218 interdependent punch list items, a protection relay compliance gap, a fire suppression booking chain, and a drone thermal validation programme simultaneously in real time.
That is not a human skill gap. That is a tool gap.
The agentic AI risk agent does not make the decisions. It makes the decisions visible fast enough to act on them. In commissioning, the gap between visible and invisible is measured in millions.
Closing thought
Have you ever been inside an EPC project where the commissioning risk was not visible until it was almost too late? What was the thing nobody saw coming?
Frequently Asked Questions
The biggest COD blocker was the protection relay sequencing failure, followed by the fire suppression certification backlog and the 14 punch list items that were actually COD-blocking.
Because the information was already about five days stale by the time it reached the table, which made it too slow for managing commissioning risk with only 19 days to COD.
ClearSpot’s Commissioning Risk Intelligence agent ranked 218 punch list items into COD-blocking, COD-threatening and trailing categories, decoded the relay sequencing issue, and drone thermal surveys found 47 hidden defects before the utility acceptance test. COD slipped 8 days instead of 37 or more.
The IEA PVPS O&M guidelines, the NLR solar O&M analysis and SolarPower Europe all link commissioning rigour and schedule risk to long-term plant performance and EPC contract structures.
Book a demo with ClearSpot to see how commissioning risk intelligence would map onto your current punch list, test records and COD schedule.