My optimizer finds the cheapest way to run the battery: $455.72 a year for a real home in Golden, Colorado, on top of the $971 its panels save. I used real NREL sunlight, real Xcel time-of-use rates and real ResStock household usage, and solved all 365 days exactly.
I found a way to ask the same question on 52 qubits instead of 117, with none of the money lost, and proved the battery cannot leave its range between checks.
I ran it on IBM hardware four times, 153 seconds of quantum time, writing down my prediction before each run. The smaller version degraded less, 0.30 against 0.37, and I repeated that in a separate job with a control.
I reproduced the community’s best circuit and validated it on all 9,024 test points.
I said it could not be improved. 25 improvements later I went back through my own eleven claims: seven stood, four failed. The one I got wrong is still up, marked wrong.
I measured the search cost everyone pays and nobody was tracking. My tool for it runs 9.2 times faster than the official benchmark, and I checked it against that benchmark on 199 of 199 attempts before I used it.
High School Research Program, six weeks, summer 2026
I am measuring how hot an ellipsoidal reflector furnace gets at its focal point, and how that changes with the lamp’s power setting, at a U.S. Department of Energy national laboratory.
The output is a calibration curve other people in the group will use after I leave, so most of my time goes to alignment, repeatability and uncertainty rather than to taking readings.
It runs through August 14, 2026. I have not published results yet.