Thanks for this data point. Is there an extract on Californians you can point to? I didn't see state there.

Additional people have some chance of being in the phase-in range on 2026-12-31 (including some currently outside California) who may plan accordingly over the coming 14 months.

California's wealth tax proposal isn't marginal: At $2B wealth, you'll pay 5% on the full $2B ($100M).

To avoid a cliff, they phase in the 5% over $1B to $1.1B wealth. You pay $0 at $1B and 5% at $1.1B. That's $55M, or a 55% tax on tha $100M marginal wealth.

I have some free time Thursday and Friday afternoons - shoot me a message if you'd like to connect in LA. And you can RSVP to the event here:
forms.zohopublic.com/imaginela1/f...

ChatGPT estimates the anti-Israel boycott against Shouk increased their risk of closure by 40% this year.

Links:
instagram.com/p/DHWA8BWJHl6
washingtonsocialist.mdcdsa.org/ws-articles/...
instagram.com/p/DPUllyAjkxb
chatgpt.com/share/68e223...

Alaska's Fairbanks region has the highest second-lowest-cost silver plan (a key PTC ingredient) for older residents.

As a result, a married couple, each 64 years old earning $83k, would gain $57,292 in 2026 if the enhanced subsidies are extended. They'd get some PTC with up to $756k income.

Here I also used LLMs to check my intuition on how readers might interpret articles (all four I tested interpreted the Telegraph's statement the way I thought the public might; that is, incorrectly).

HMRC projected last year that raising the VAT threshold to £90k would increase revenues in 2028-29. Looks weird until you realize that would represent a tax hike by then, since it'd otherwise be £92k!

Yep github.com/PolicyEngine...

We have a suite of subsequent checks (separate agents) after test-creator and rules-engineer complete. If they misalign that gives more to interrogate.

We'd like to test our hypothesis here with a controlled experiment, but my guess is that our approach reduces cascading errors.

Tests can be wrong! This way we get two interpretations of the specifications (statutes, regs, manuals), coming from different perspectives. If the agent misinterprets the spec in constructing the test, it's more likely to make the same error at the encoding stage--less so if they're starting fresh.