oh hai there https://nutstacker.space/?q=nevent1qqs8qq7lq8475ga4vnn8zt2lkhkent2n8ypglj3c5w9ajdacwqvz2fgppemhxue69uhkummn9ekx7mp0qgsr7acdvhf6we9fch94qwhpy0nza36e3tgrtkpku25ppuu80f69kfqrqsqqqqqp8kd5zr

the perfect commit doesn't exi–

tacking a long overdue task: playing with broadcast channels

I tried vibe coding a hashpool upgrade. I accomplished my development goal after a lot of back and forth. But now there's a bunch of code I don't understand in there. Seems like I will be spending the next week cleaning up the mess generated this week. I'm hitting my development milestones. A win is a win. I still think the best way to use LLMs is with heavy oversight and manual approval for each code change. I can probably extend the leash a bit and let it run on a tightly scoped problem like "get these tests to pass" or "fix these compile errors". Less so with unknown or fuzzy requirements like "redesign the mint quote flow to send a locking key bundled with the mining share instead of the blinded messages". That's how I ended up with all this vestigial code and sub-optimal flows.