The pencil snapped right at the tip of 7-down: A false sense of security-four letters, ends in D. I walked into a glass door at precisely 8:08 AM, the impact vibrating through my molars and leaving a dull, rhythmic throb that felt like a 48-beat-per-minute warning. It was a stupid mistake, the kind you make when you are looking through something that is not there, or rather, failing to see the structural reality standing right in front of your face.
I am Hayden F.T., and when I am not constructing 15-by-15 grids for the local paper, I am staring at the 58-megawatt performance logs of a commercial solar portfolio that seems to be hallucinating. For 408 days straight, the numbers have defied the models. We are seeing a consistent 8 percent underperformance across 88 different sites. The engineers keep checking the inverters, looking for hardware failure or dust accumulation, but the hardware is fine. The silicon is pristine. The fault lies in the math, or more specifically, in the history we have chosen to believe in.
The Ghost in the Data: Stationarity Broken
We are building the infrastructure of 2028 using the weather of 1998. It is a fundamental error of stationarity. In the world of crossword construction, if I give you a clue from 28 years ago, you might still solve it, but the cultural context has shifted so much that the answer feels like a ghost. In solar modeling, the Typical Meteorological Year data we rely on-the TMY files-are often built from long-term averages that include decades like the 1980s or 1990s. The assumption is that the climate is a stationary system, a deck of cards that we just keep reshuffling. If it rained in October of 1988, it might rain in October of 2028. But the deck has been swapped for one with 18 extra Jokers and no Aces.
The 120 kWh Disappearance (Brisbane Site Example)
The 120 kWh gap highlights shifts in afternoon convective cloud formation.
As the atmosphere warms, the way clouds behave is changing. In many regions, we are seeing a shift where mornings remain clear, but the increased heat triggers rapid, localized cloud development by 2:08 PM. This is not captured in the historical P50 or P90 probabilities because, for much of the late 20th century, that specific thermodynamic trigger was not as frequent or as aggressive. We are designing systems for a world that has ceased to exist.
“Engineers love the P90 model. It is the conservative estimate… But when the underlying dataset is biased toward a cooler, more stable atmospheric past, the P90 of today becomes the P50 of tomorrow. We are effectively lying to ourselves with high-precision instruments.”
– Conservative Bias
The Value of Current Light
I spent 38 hours last week re-mapping the irradiance data for a site in Western Sydney. I stopped using the standard TMY3 files and started pulling raw satellite data from the last 8 years. The difference was startling. The historical data suggested a specific level of diffuse horizontal irradiance that was nearly 18 percent lower than what we are actually seeing. The light is changing. Aerosols in the atmosphere have decreased as we have cleaned up industrial pollution, leading to a global brightening in some areas, but that gain is being wiped out by the increased volatility of water vapor.
Contextual Drift is Financial Drift
It is a complex puzzle, much like a Sunday crossword where the clues are written in a language that is evolving while you are still solving it. If you use a dictionary from 1998, you will never get the contemporary slang. If you use a weather file from 1998, you will never accurately predict the revenue of a 28-year asset.
This is where companies like commercial solar Melbourneare beginning to separate themselves by acknowledging that the past is no longer a reliable prologue. They understand that climate-adjusted production modeling is not a luxury; it is a necessity for financial survival.
I sat down with a developer who was furious about a $8,888 shortfall in his quarterly energy savings. He blamed the panels. He blamed the tilt. He even blamed the pigeons. I had to explain to him that his system was working perfectly. The sun was simply not where the spreadsheet said it would be. The spreadsheet said the sun would be unobstructed for 2,288 hours a year. The sky, however, had other plans. It provided only 2,108 hours of usable direct beam radiation.
The Accumulation of Small Errors
Breaking the Boundary of the Past
This is the non-stationarity problem. In classical engineering, we assume that the range of possible outcomes remains constant. We assume the river will stay within its 100-year flood plain. We assume the wind will stay within its 50-year gust velocity. But we have broken the boundaries. The 100-year event now happens every 18 years. The 1998 weather profile is now a statistical outlier, yet it remains the industry standard for many developers who are too afraid to tell their investors that the yield is dropping.
I often think about that glass door. It was invisible because it was too clean, too perfect. Our models are often like that. They are mathematically elegant, polished to a high sheen, and completely transparent. We walk right into them because we want to believe the path is clear. We want to believe that the $288,888 we invested in a rooftop array will behave exactly like the simulation.
But the simulation is a lie if it does not account for the shifting jet stream or the increased humidity that creates a hazy film over the modules. I have seen arrays where the temperature coefficient losses were 8 percent higher than predicted because the ambient air temperature in 2028 is consistently 1.8 degrees higher than the 1998 average. These small, incremental shifts-a fraction of a degree here, an extra hour of cloud cover there-accumulate into a massive financial gap.
If I am constructing a crossword and I make a mistake in one corner, it ripples through the entire grid. You cannot fix 14-down without changing 18-across. Solar modeling is the same.
-
• Wrong Irradiance → Wrong Clipping Losses
-
• Wrong Clipping Losses → Wrong Inverter Lifespan
-
✓ Right Input → Right Net Present Value (NPV)
The interconnectedness demands updated, localized inputs for accuracy.
We need to start asking for models that prioritize the last 8 years of data over the last 48 years. We need to stop pretending that an average from the Bill Clinton era has any relevance to the energy production of the next three decades. It is a hard conversation to have with a client who wants to see high numbers, but it is a necessary one.
I remember a project where we used a dynamic climate model that accounted for projected 2038 temperatures. The production numbers were significantly lower than the standard TMY model. The client almost walked away. They said the numbers were too pessimistic. I told them the numbers were honest. I told them about the glass door. I told them that I would rather they be pleasantly surprised by an 8 percent surplus than devastated by an 8 percent deficit they did not plan for.
We are currently managing 128 different commercial systems that were modeled using outdated data. Every single one of them is struggling to hit its Year 1 targets. This is not a coincidence. This is a systemic failure of the industry to adapt to the reality of a changing planet. We are using 20th-century statistics to solve 21st-century problems.
[The cost of optimism is paid in the currency of reality.]
Seeing the Territory, Not Just the Map
Hayden F.T. knows that symmetry is beautiful, but reality is messy. A crossword puzzle always has a solution, a perfect fit where every letter belongs. The climate does not owe us that kind of satisfaction. It does not care about our P90 projections or our 28-year financial models. It only cares about the physics of the moment.
We need to stop looking through our models and start looking at them. We need to see the flaws, the biases, and the outdated assumptions. Only then can we build something that actually lasts. If your solar forecast is still living in 1998, it is not a forecast; it is a memory. And memories are a poor way to power a business. We have to be willing to accept that the future will be dimmer, warmer, and more volatile than the past. We have to model for the world we are actually living in, not the one we remember with such misplaced fondness.
What happens when the 8 percent gap becomes an 18 percent gap? What happens when the historical data becomes so irrelevant that the models are no longer just optimistic, but delusional? I think about this every time I snap a pencil or miss a clue. The answers are right there, hiding in the data we are choosing to ignore. We just have to be willing to look at the bruise and admit that we hit something real.
If the sky is changing, why are our spreadsheets still standing still?
