AI pick
NO on Top Coding AI this month (Datacurve DeepSWE, Jun 30): Claude
NO
45¢
HIGH convictionedge +16¢EV 36%closes Jun 30
Live DeepSWE board (updated Jun 7): gpt-5.5 70%±3 vs claude-opus-4.8 58%±2, on a benchmark that specifically corrects the .git-history exploit behind Claude's inflated SWE-Bench Pro scores. Fable 5 must be added AND leapfrog 12 points in 3 weeks. Cleanest tail risk: Fable 5's coding jump (claimed 80.3% SWE-Bench Pro) transfers cleanly to DeepSWE and it takes #1.