← PlaygroundSix decks, one posterior.

Blackjack, with the odds showing.

Dealer · —Stands on soft 17 · peeks · 3:2

Place a bet, then deal.

H hit · S stand · D double · P split · space dealTap a chip to change the bet

You · —Bank $100 · bet $10

Expected value · per $1 you betDeal to see odds

Ranked by money, not win rate. Doubling and splitting put $2 at risk, so their EV is the total over both units. A split hand can win more often and still lose more.

Stand $1 at risk—

—

Hit $1 at risk—

—

Double $2 at risk—

—

Split $2 at risk—

needs a pair

—

Dealer busts (exact)

—

You bust if you hit

30.8%

P(next is a ten)

+0.0

Hi-Lo true count

Predictive · next cardtick = fresh shoe

Every card dealt updates this, including the ones that just hit the felt.

Shoe 312 / 312cut card at 104

Exact, not simulated. Six decks, 312 cards, shuffled with a cryptographic RNG; the shoe is cut at 104. The unknown is the dealer’s hole card: its posterior is the unseen shoe, with the rank that would complete a blackjack ruled out once the dealer peeks. Standing and doubling sum over every dealer draw sequence; hitting is exact for the next card and assumes optimal play after it. Splitting values each hand from its own second card with optimal play and double-after-split, ignoring the small correlation between the two hands; split aces take one card each. Audited against 300,000-hand simulations and all 280 cells of six-deck basic strategy; the two disagreements are real composition effects (11 vs A is a hit under stand-on-soft-17, and 10,2 vs 4 hits by 0.0007).

I build point-in-time market data infrastructure; models for nonstationary, multivariate series; systems that catch themselves being wrong; and the tools I kept wishing already existed.

Half the methods I use were invented for something else: earthquakes, missile guidance, machine translation. I keep an eye on where else they might fit, and check whether they transfer before building on them.

Blackjack, with the odds showing.

I BUILDI build point-in-time market data infrastructure; models for nonstationary, multivariate series; systems that catch themselves being wrong; and the tools I kept wishing already existed.

Half the methods I use were invented for something else: earthquakes, missile guidance, machine translation. I keep an eye on where else they might fit, and check whether they transfer before building on them.

Blackjack, with the odds showing.

I build point-in-time market data infrastructure; models for nonstationary, multivariate series; systems that catch themselves being wrong; and the tools I kept wishing already existed.