文章
關於 AI 代理、回測與市場預測稽核，以及平台架構的研究。
加密貨幣與市場
Retail sees price and volume. Machine learning reads liquidity — depth, order-flow imbalance, market impact, hidden orders — the short-lived information beneath a single price. And its honest limit.
閱讀時間 9 分鐘7 個章節
加密貨幣與市場
Measured over the full Binance history: the root-t band is 20% too wide at one day, converges by thirty, and crosses over for ETH. Plus the overlapping-window correction almost everyone skips.
閱讀時間 7 分鐘7 個章節
預測紀錄
Our AI agent team's 24h, 7d and 30d ranges for Bitcoin and Ethereum, 21 July 2026. BTC core 50% $64,435-$66,017, ETH $1,873-$1,940. Locked and Bitcoin-timestamped before publication, scored in public.
閱讀時間 4 分鐘5 個章節
加密貨幣與市場
Why Ethereum is a different forecasting problem from Bitcoin — staking queues, fee burn, rollup migration, and the beta trap that makes weak ETH models look skilful.
閱讀時間 9 分鐘7 個章節
加密貨幣與市場
Where AI genuinely changes a crypto exchange: surveillance, risk engines, liquidity, AML, matching-engine ops — and where the label is mostly marketing.
閱讀時間 9 分鐘7 個章節
加密貨幣與市場
What AI can honestly do for Bitcoin price prediction, what is theatre, and why a timestamped, hashed, publicly scored forecast is the only real evidence.
閱讀時間 7 分鐘6 個章節
可證明性
Most AI crypto bots show a curve and no ledger. What an accountable trading agent looks like: reasoning logs, risk-first design, and a public record.
閱讀時間 4 分鐘5 個章節
可證明性
AI stock predictions are everywhere and almost none are verifiable. Why a real market forecast must be timestamped before the outcome, and scored in public.
閱讀時間 5 分鐘6 個章節
預測市場
One model is a single point of failure. How a team of AI agents produces calibrated probabilities on Polymarket and Kalshi, and why every call is scored.
閱讀時間 5 分鐘5 個章節
預測市場
How Polymarket turns money into probability, why market prices are hard to beat, and how an AI forecaster uses them honestly — benchmarked and scored in public.
閱讀時間 3 分鐘4 個章節
預測紀錄
France vs England 2026 World Cup bronze final: NeuPortal's Poisson model call — locked, SHA-256 hashed, and Bitcoin-timestamped before kickoff, scored in public.
閱讀時間 3 分鐘4 個章節
AI 產業
Most "AI trading" claims can't be proven wrong. Here's how to actually evaluate an AI trader: locked forecasts, proper scoring, risk-adjusted returns.
閱讀時間 6 分鐘7 個章節
