NEXUSFRAMEWORKS

Home / Track record

Backtest

Would our engine have found them first?

A ranking is only useful if it points to places that actually get built. So we score all the land, then check where real projects landed, including the ones we missed.

Test 1 · Megasites

Where chip plants went in New York

We scored 134,000 parcels on big power, buildable land that could be assembled, workforce distance and ag-district limits, grouped them into 151 areas, and checked four real chip sites.

SiteCompanyChosenOur rankResult
White Pine Commerce Park, ClayMicron2022#2 of 151Flagged
Marcy Nanocenter, MarcyWolfspeed2019#4 of 151Flagged
STAMP, Genesee CountyEdwards Vacuum and others2023Outside top 151Missed
Luther Forest, MaltaGlobalFoundries2006–09Outside top 151Missed

What we got right

Both sites that already had big power next door, a large block of flat land and a nearby workforce came out near the top. In an earlier version of the test, Micron's site stayed in the top three across 200 different weightings.

What we missed, and why

STAMP and Luther Forest were "made" sites: the state and utilities brought major power to them after they were chosen. Our engine sees power that exists today. The next version adds planned transmission and state site investments.

Test 2 · Utility solar

Where large solar farms went

We collected 113 solar farms over 5 MW, built and planned. We tuned the model only on farms online by 2022, then tested it on farms built or planned since 2023.

ModelNew farms in its top 10% of landIn its top 20%
Utility solar engine34%51%
Community solar engine (1–5 MW)11%20%
Random choice10%20%
4.5×

Large farms are 4.5 times more likely within about half a mile of a substation. Distance to the substation is the strongest single signal.

27–42%

Across 162 weight settings, the share of new farms caught in the top 10% stayed in this range. The result doesn't hinge on lucky weights.

46%

For farms of 20 to 100 MW, the model catches nearly half in its top 10%. Bigger projects follow the grid more closely.

How to read these results

Honest limits

  • These tests use today's land and grid data, so they are hindsight tests. The real proof is predicting sites before they're announced, and we're tracking that now.
  • 113 large solar farms is a small sample, and each farm is matched to the parcels around its center.
  • Our rankings don't yet check how much spare capacity each substation has. We're adding the state grid queue next.

Want to see how your site or region scores?

We can run any upstate county, town or project type through the engines and share the results privately.