70% of Binance altcoins fall below their 200-day moving average

From mid-August to the end of September, altcoin market cap rose by about $320 billion. The pullback erased about $200 billion, wiping out two-thirds of the gains. Now, 70% of Binance altcoins are below their 200-day moving average, near where they were before the big August rally.

→ The August rally has effectively been wiped out. Wait for a base to form before buying the dip; don’t mistake a bigger drop for a better bargain.

Mining difficulty has gone 374 days without a new high—the longest stretch ever

After difficulty peaked at 155.97T on October 29, 2025, it hasn’t set a new high in 374 days, surpassing the 364-day stretch after the 2011 crash. Difficulty reflects actual online hash rate; when miners shut down, they go offline.

→ Miners shutting down reflects real losses, not a show. Prices can lie; electricity bills can’t.

BTC spot ETFs saw $1.28 billion in net outflows this week

Bitcoin at $75,817. Farside data shows U.S. spot ETFs saw $1.2817 billion in net outflows this week, with BlackRock’s IBIT alone seeing $1.209 billion in outflows, accounting for 94%. The AHR999 indicator is at 0.431, returning to the buy-the-dip zone for the first time since late August.

→ An indicator entering the zone only means prices are cheap. Whether anyone is buying is reflected in ETF flows—and right now, the answer is outflows.

Hands-on test of AI operating a computer in the background: getting work done and staying out of the way are two different things

Cua and Scale AI released Driver Bench, which tests whether agents can operate software in the background without taking over the mouse or keyboard. On macOS, Cua had a 58% end-to-end success rate, Codex 33%, and browser-use just 3%. browser-use completed 32 of 40 tasks, but interrupted users 31 times, so only one task counted as a pass.

→ The bar for desktop agents is whether you’d trust them to use your computer while you’re typing. Completion rate comes second.

Meta’s former CFO jumps to Fireworks AI

Dave Wehner spent 14 years at Meta before leaving to join AI inference platform Fireworks. The company raised $1.505 billion in July at a $17.5 billion valuation, has annualized revenue above $1 billion, and processes 40 trillion tokens a day, 95% of them through custom models for clients. It will launch a new funding round next week, targeting a $30 billion valuation.

→ What supports a $30 billion valuation is $1 billion in annualized revenue and a business where 95% comes from custom models. Someone who managed Meta’s balance sheet joining the open-source inference layer carries more weight than a research report.

In a falling market, pay less attention to buy-the-dip indicators and more to fund flows and electricity bills.