用 spider 脚本让四模型按任务竞标
一条命令把业务拆成 6 类任务，Opus、Sonnet、Jev、opendots 各自打分竞争，Jev 在决策环节以 0.97 胜出。
ONE COMMAND SPLIT MY BUSINESS INTO 6 JOBS AND LET 4 MODELS FIGHT FOR EACH ONE stopped picking one model for everything. built a spider that merges opus 5.5, sonnet 5.5, jev and opendots into one system and makes them compete for every job before any of them gets to keep it → one command splits the business into 6 jobs - plan, code, decide, run 24/7, review, ship → every job gets 4 candidates scored, the spider locks in whichever one wins and swings to the next → plan goes to opus 5.5 at 0.94 - jev scores 0.52 on the exact same job → decide flips completely - jev wins at 0.97, opus 5.5 drops to 0.74, so the model that lost the plan is the one making the calls → code goes to claude code, no contest → the job nobody wants, the one that runs all night, goes to opendots → opus 5.5 comes back a second time just to check the work, sonnet 5.5 ships it first run, jev got "plan" by default because it scored first and nothing had been compared yet. it was sitting at 0.52 against opus's 0.94 before i caught it. remember that part - rank before you lock, never after everyone keeps one model because switching feels like extra work. it's 4 lines - score, compare, lock, next $0 extra tooling, just one spider script picking per job instead of me picking once for the whole business 6 of 6 modules locked, system's live run the same test on your own stack before you pay one model to do all six jobs
这条是引用着另一条写的：x.com/noisyb0y1/status/2105964235087856080