Yvette Teo

The week the calendar filled up

17 August to 23 August 2026

Two events moved markets this week, three days apart. On Wednesday 19 August, Trump gathered crypto executives and both regulators at the White House and publicly pressed Congress to pass the CLARITY Act, the bill that would finally settle who regulates crypto in the US. The next day, Treasury announced it would double its long-dated buyback operations after 10-year and 30-year yields touched two-decade highs.

Quick terms, so the rest is clear:

  • CLARITY Act is the US bill that settles who regulates crypto, tokens versus commodities. It needs about ten Senate Democrats; its cloture vote lands 15 September.
  • Cloture ends Senate debate and forces the final vote. Without it, a bill stalls indefinitely.
  • Buybacks here mean Treasury repurchasing its own long bonds from the market.

Trump put his weight on the scale

The 19 August summit was the most direct presidential endorsement of crypto legislation this cycle. Trump is trying to push the CLARITY Act through Congress: the bill would end the SEC-versus-CFTC turf war by drawing a clear line between tokens regulated as securities and crypto treated as commodities, and give exchanges a real licensing path. He gathered the SEC and CFTC chairs plus Coinbase, Ripple, Nasdaq and NYSE at the White House to signal that the administration wants it done.

A presidential push does not pass bills, though. CLARITY still goes to cloture on 15 September sitting near 10% odds, because about ten Democrats have to cross and none have committed. Markets read the intent anyway: BTC jumped to about $69,500, its biggest daily gain since early June, with over $1bn of short positions liquidated within the hour per Coinglass data cited in coverage. The planning input is the date, not the speech: 15 September is when intent meets arithmetic.

Treasury leaned on the long end

The same week, long-end yields hit levels last seen around two decades ago after a buyers’ strike dating to late June, the 30-year sitting near 5.2%. On Wednesday 19 August Treasury announced it would at least double the maximum size of its liquidity-support buybacks, from $2bn to at least $4bn per operation, covering exactly two maturity sectors: 10-year to 20-year and 20-year to 30-year. Yields fell immediately, the 30-year dropping 9 basis points to 5.196% on the day. The next day Bessent went further on CNBC: the buybacks “could be more than the $4 billion per issue,” arguing yields don’t reflect fundamentals and that Treasury is going to “make a market” in long bonds.

The market’s answer came fast. The 30-year rebounded to 5.247% within 24 hours, erasing the entire drop: the announcement alone did not hold the level. And the announcement did something else worth watching. It also fed the debasement trade: a weaker dollar plus a government openly leaning on long-end yields lifts hard assets, and gold and Bitcoin both rallied in its wake. When Treasury starts managing yield levels while AI issuance piles into the same duration, scarce assets start looking like the hedge.

The operations run 9 September through 4 November, landing on election week. Whether yields stay managed or drift back up tells you if this was control or theatre.

The stealth model everyone is trying to unmask

Ox Alpha appeared on OpenRouter on 20 August: free during preview, anonymous provider, one million token context, text, image and video input. Within days it was processing trillions of tokens daily and beating frontier models on agentic coding evals, roughly 80% on a community DeepSWE subset against about 65% for Claude Fable. The internet did what it does: tokenizer fingerprints point hard at Zhipu’s GLM family, with Xiaomi and Tencent as weaker theories, and nobody has confirmed anything. I ran my own stack on it all week. Verdict next brief, but the lane itself is the story: evaluate the work blind, then find out whose model it was.

The heart attack every model switch brings

Once Ox Alpha dropped on OpenRouter, I moved my whole scheduled stack onto it, with DeepSeek kept as fallback. The migration itself is fine. The recurring fear is the check. Every time I switch models, there’s one cron job somewhere that still fires on whichever model I left behind, and when I open OpenRouter to look at my spend, I hold my breath for a second while I scan the list.

It’s the operator version of checking you locked the door. You know it’s almost certainly fine, you check it anyway, because it’s not the migration that bites you, it’s the one job you forgot to repoint. This time I got it properly sorted: every scheduled task now inherits one shared model setting, so there’s no old copy left to surprise me. The unglamorous fix that buys the calm.

One question I am carrying into next week: if the 30-year does not come down even after Treasury starts buying on 9 September, what does the Fed do? Do they step in and buy more, or is there no lever left, and where does the market land then?

Sources