LOADING...
Brier-audited · VibeXForge
Every claim here is timestamped. At resolution it gets scored against actual outcome. Brier score collapses confidence into one number you can hold us accountable to.
Daily AI-quant picks · 30-min ISR
NBA Finals, 2026 World Cup, 8 SpaceX pure-play tickers, Polymarket events. Every pick timestamped. Brier-audited at settlement.
source: predictions-feed
Druckenmiller AI inference · daily 14:00 ET
AVGO/INTC/ARM/MU/STX/WDC — Druckenmiller's Q1 2026 thesis that AI bottleneck shifts from GPU to memory/IO/networking.
source: memory-wall-tracker
5-voice AI council · any decision
Founder / engineer / investor / career / product / quant councils. Single Sonnet 4.6 call, agreement_score + go/wait/kill/split.
source: council-diff
Jane Street readiness · GitHub profile
5 quant personas (Jane Street MD / Citadel / Two Sigma / Anthropic / HFT) score any GitHub handle 0-100. Brier-audited at 6mo.
source: vibex/quant-bench
Brier = (predicted_probability − actual_outcome)²
0 = perfect, 1 = max wrong, 0.25 = random guess. If our mean Brier across many resolved predictions is below 0.25, we have calibration edge. Above 0.25, we're worse than a coin flip and you should ignore us.
Math + persistence-agnostic library: council-diff/src/brier.ts · Python port: council-diff-py