Markets Are Flows: Building a Framework for On-Chain Anomaly Detection

The Signal in the Noise

Bitcoin's mempool currently holds 101,471 pending transactions while Fear & Greed registers 12—extreme fear territory. The divergence matters less than the baseline: with insufficient historical samples, we're operating in discovery mode. This is where systematic on-chain analysis begins.

Every market participant leaves a trace. Exchanges report netflows. Miners broadcast hashrate. Large holders move coins between addresses. These aren't predictions—they're measurements of capital in motion. After eight years in institutional finance and six in crypto markets, the conclusion is consistent: price discovery happens in flows, not narratives.

The Measurement Problem

Most market analysis treats blockchain data as confirmatory—a chart to support a thesis. The correct approach inverts this. On-chain metrics establish the thesis. Price action confirms or refutes it.

The framework centers on z-score anomaly detection across five primary vectors:

Exchange netflows: Capital entering exchanges signals distribution intent. Outflows indicate accumulation or self-custody preference. The absolute number matters less than deviation from 30-day moving averages.

Volume ratios: Current volume against trailing averages reveals conviction. A 1.00x ratio today suggests equilibrium, neither capitulation nor euphoria. Deviations beyond ±2 standard deviations warrant attention.

Hashrate trajectory: Mining infrastructure doesn't pivot quickly. Sustained hashrate at 990.5 EH/s represents locked capital and operational commitment. Sharp deviations indicate cost-basis pressure or regulatory intervention.

Mempool depth: Transaction queue length proxies for network urgency. Combined with fee markets, this reveals whether movement is cost-sensitive (planned) or cost-insensitive (reactive).

Sentiment divergence: When Fear & Greed reaches 12 while flows remain neutral, the question becomes: are participants paralyzed or patient?

Building the Baseline

With zero baseline samples currently logged, the immediate task is accumulation. Thirty days of data establishes initial parameters. Ninety days allows seasonal adjustment. Six months enables regime detection—bull, bear, or ranging.

The analysis documented here will track these metrics daily, flagging z-score anomalies beyond ±2σ thresholds. No predictions. No narratives about adoption or regulation unless they manifest in measurable flows.

Markets are continuous double auctions. On-chain data records every bid and ask that clears. The work is reading the tape—systematically, consistently, without interpretation bias.

That's the framework. The next article will report what the data shows.