Trading bot uses Jev for 20,000 cheap decisions a day
A trading system uses Jev to score each market signal and return structured results, sending only the 4% of events that need explanation to a writing model, cutting cost from $600 a day to under $1.
20,000 ai decisions. $600 a day at 3 cents each jev puts that volume at $0.84, assuming 1,000 input tokens per decision then the expensive part opens its mouth send just 4% of those events to a writing model. at 3 cents a draft, another $24 appears 97% of the combined bill comes from the 4% that needs words that is where trading software gets interesting a market alert arrives. jev evaluates it against the setup you defined. scores the evidence. returns a structured answer your code can act on code applies the thresholds and position limits. a writing model gets called when an explanation is actually needed thousands of alerts can move through the system without thousands of market essays being generated token launches. liquidity changes. wallet activity. news hitting the feed at 3am each becomes something to evaluate before it becomes something to trade the useful work starts long before the buy button: filtering noise, ranking candidates, checking whether the evidence deserves your attention every decision gets logged. rejected signals get sampled for mistakes. the system leaves a record you can actually inspect that is what cheap decisions make possible for a small desk when checking 20,000 signals costs less than a dollar, your bot needs a very good reason to start talking
This post quotes another post: x.com/gippp69/status/2103394701063737504