Repository navigation
feat(article): add blackjack README table intro - #14
Conversation
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configuration
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
| A wrong board is visible. A wrong expected value just quietly tells strangers that the worse play was better. So the calculator is checked three ways in `tests/test_bjmath.py`: | ||
|
|
||
| 1. **Against an independent Monte Carlo** that deals a real shuffled shoe with a physical hole card and rejection sampling for the peek. It shares no code with the calculator. | ||
| 2. **Against the published basic strategy chart.** All 260 cells have to fall out of the numbers from a full shoe, without the calculator ever being told what the chart says. |
There was a problem hiding this comment.
"All 260 cells" reads as recalled. A standard six-deck basic strategy chart is usually 10 dealer upcards × (hard + soft + pair rows), which gives a different total depending on which rows you count (e.g. 280 with hard 5–17, soft A2–A9, pairs 22–AA). Please check this against the chart tests/test_bjmath.py actually uses and state which chart it is (source, and S17/DAS/late-surrender variant).
|
|
||
| When that chip went down the running count was +0 with 4.9 decks left, | ||
| so the true count was +0.0. A shoe like that is worth about -0.12% a unit, | ||
| measured over 200 million rounds of these rules, so 1u on it was worth |
There was a problem hiding this comment.
The "-0.12% a unit" edge at true count 0 and the "200 million rounds" figure are presented as measured results, but nothing in the article or repo points to where they come from. A true-count-0 edge for these rules is commonly quoted around -0.4% to -0.5% for six decks, so -0.12% needs a source or a note on how it is defined (per initial bet, conditional on count bucket, etc.). Also, "+0 running count with 4.9 decks left" giving a true count of exactly +0.0 is fine, but the example then prices 1u at -0.001, which is -0.12% of 1u rounded; say so or the sample looks inconsistent.
| A strategy chart answers "what is best on average over a fresh six-deck shoe". That is the wrong question halfway through a shoe. `bjmath.py` works in compositions (how many aces, twos, ..., tens remain) and enumerates every outcome, so each action is priced for this shoe and no other: | ||
|
|
||
| - Cards are drawn without replacement all the way down, including the dealer's. | ||
| - The calculator is never allowed to see the hole card. Enumerate it, take the best action inside each branch, and sixteen against a ten starts "standing when the dealer has sixteen", a quarter of a unit no player can collect. Every decision is averaged over the hole cards the player cannot tell apart. |
There was a problem hiding this comment.
This sentence is garbled: "sixteen against a ten starts 'standing when the dealer has sixteen'" does not parse, and "a quarter of a unit" is an unsourced number. Rewrite to say what actually happens (an optimiser that sees the hole card picks stand/hit per hole card and overstates the EV of 16 vs 10 by about X units) and give the measured X from the calculator.
| - githubactions | ||
| - gamedev | ||
| series: ShowDev | ||
| date: '2026-11-15T00:00:00Z' |
There was a problem hiding this comment.
Compare with the sibling ShowDev articles (e.g. shogi-on-github-profile.md): they carry a cover_image under articles/assets/<slug>/ (see README "Assets", 1000x420 via scripts/gen_cover_image.py). This one has none. Also confirm the future date: (2026-11-15) is intended; with published: false the hourly scheduler will flip it publicly at that time.
|
New article Findings, all factual or housekeeping:
I did not run |
1 participant
Intro article for 0-draft/blackjack, written as the exposure step from the 2026-10-15 archive check (kanywst/bellman#317).
published: falsewithdate: 2026-11-15T00:00:00Z, so merging schedules it:schedule.ymlflips it live at that time. Change or drop the date before merging if you want it held.blackjack.ymlafter the move. The mechanics described (hole card drawn only on reveal, peek roll, composition-based pricing, the three calculator checks) are taken fromscripts/blackjack.py,scripts/bjmath.pyandtests/test_bjmath.py.validate_articles.py: 0 errors, 2 warnings (nocover_image; the single-article series warning, which also fires on feat(article): add identity-deep-dive radar intro #13 although ShowDev has 24 articles).