Because computers are so predictable and rule-driven, programmers tend to overlook that the real world is very much adversarial and unpredictable. The "normal" thing to do here is to simply play with your small solar projects and not pay the fee, because the chance that you'll ever get charged the solar tie-in fee for a solar-powered garden light is basically zero. It's not precisely zero, but it's comparable to things like a drunk driver crashing into your house and killing you - one of those little tiny risks you just have to ignore because dealing with the risk costs more than the expectation cost of the risk.

Consider that they have a finite number of people and it costs more for them to go around inspecting everyone's garden lights than their expected revenue from that, so they won't.

Also, real-world risk and reward are dynamic. When the Deutschlandticket came out in Germany, almost everyone in Berlin bought one, so now they very rarely check tickets on trains - the expected gain is near-zero because almost everyone has a ticket for almost every train now. But if people start exploiting that by riding without a ticket, the gain will go up and after a time lag, they'll notice and start checking tickets more often.

Target, the store, is known to not arrest shoplifters until they exceed the felony threshold, at which time they arrest and prosecute the shoplifters.

If a solar tie in fee is $100/month, and a “customer” is using their garden lights for 1 year, it’s only $1200 to find them and issue a fine (assuming the fine is simply equal to the avoided tie in fees). Probably not worth it for the power company. But… after 10 years, the tie in fees would be $12k - certainly profitable for an intern to do some Google image searches, if they can find just 100 violators, that’s $1.2M. In 10 years, it’s not going to be an intern doing this, it’s going to be some AI agent tasked with optimizing revenue generation, and their costs approach zero.