用 Jev 打分加蜜蜂记忆层做 memecoin 交易 bot
交易 bot 里 Jev 190ms 给池子打分,但每池从零评分无记忆,于是加了 2000 神经元蜜蜂脑记住踩坑组合。
I BUILT A BEE BRAIN WITH 2,000 NEURONS AND IT MADE 5 SOL ON MEMECOINS WHILE I SLEPT. THE BEST TRADE WAS NOT FOUND BY THE MODEL. IT WAS FOUND BY MEMORY repo: http://github.com/h100envy/beebrain nerve scans 40 pools per hour. 8 nodes in sequence. reflexes kill obvious rugs. jev scores in 190ms. grok reads narrative but every pool is scored from scratch. node 400 has no memory of nodes 1 through 399 a deployer who rugged twice last week walks in with a new token. jev gives it 0.72. grok reads fresh telegram hype and scores 0.81. the pipeline sees a decent pool because it already forgot the last two so i built a bee. 2,000 kenyon cells. 5 brain lobes. one neuron that decides if the trade was worth it > antennal lobes smell 8 pool features: liquidity, holders, snipers, bundles, creator age, narrative, volume, heat > optic lobes read the last 60 narrative trend readings. a pool rising on a flat market looks different from one rising because everything is up > mushroom bodies form a sparse code. 100 of 2,000 cells fire. their weights remember: did this pattern end in sugar or poison > central complex weighs memory, noise, narrative shape and sniper share > motor output: PASS, WATCH or SKIP with the full waggle vector when the trade closes the result travels back. octopamine for profit. dopamine for rug. only the cells that fired for that pool get updated the bee does not learn rules. it learns combinations. a fresh wallet with low liquidity and a copied description is a rug. a fresh wallet with locked LP and a live telegram might be a print. the signal is in the combination not the individual feature first run: 360 pools over 9 days. the bee skipped everything for the first 40 pools. exploration at 60%. it was drunk by pool 120 it found that creator wallet age plus sniper share below 8% hit 79% of the time. nobody told it this > pools scanned: 360 > trades: 28 > wins: 19 > exploration start: 60% > exploration end: 5% > result: 5 SOL pool #247 was the best. mushroom bodies scored 0.91. highest of the entire run. 4 month old creator wallet, locked LP, 3% snipers, telegram growing 40 minutes before listing. the bee entered. +1.8 SOL on one impulse grok would have read the telegram and said yes. jev would have confirmed the metrics. but neither of them remembers that this deployer already printed twice last week. the bee remembers everyone builds a bigger brain. i built a smaller one with memory and it set the high score because the only node in nerve that learns from its own trades is an insect
这条是引用着另一条写的:x.com/h100envy/status/2103900644176462258