Game model and selectable rules
The trainer builds a shuffled shoe from 4, 6, or 8 standard 52-card decks. The user can choose whether the dealer stands on soft 17 (S17) or hits soft 17 (H17). The displayed base bet changes the simulated wager; all chips are fictional.
The current settlement assumptions are:
- dealer receives an up-card and hidden hole card and peeks for blackjack;
- blackjack pays 3:2;
- doubling is available on a two-card hand when the simulated bankroll covers the added wager;
- doubling after a split is supported;
- equal-value pairs can be split to a maximum of four hands;
- split aces receive one additional card per hand and then stand;
- a push returns the original wager.
Strategy recommendations
Each manual decision is classified as a pair, soft total, or hard total. The recommendation function receives the player cards, dealer up-card, whether doubling and splitting are currently available, and the selected S17/H17 rule.
The compact table covers hit, stand, double, and split. If a preferred action is unavailable, the trainer uses a legal fallback. Surrender and insurance are not offered. Because table rules affect optimal decisions, recommendations should be read only within the assumptions listed above.
Important limitation: this is a total-dependent practice table, not a composition-dependent solver or expected-value calculator. It does not model every regional rule, payout, side bet, promotion, or shuffle condition.
Hi-Lo running and true count
Exposed cards update the running count using the standard Hi-Lo tags: cards 2–6 add +1, 7–9 add 0, and tens, face cards, and aces add −1. The dealer hole card is excluded until revealed.
The true count divides the running count by estimated decks remaining, using the exact number of cards left in the simulated shoe and a floor of one quarter-deck to avoid unstable division near the end. This is more precise than the deck-estimation skill used at a physical table, so learners should practise estimation separately.
Feedback, scoring, and leaderboard
A strategy decision is counted when the player manually chooses hit, stand, double, or split. The trainer compares that action with the current recommendation and reports whether it matched. Autoplay is a demonstration feature and does not represent an unaided learner decision.
The streak pauses while the visible strategy hint is enabled. A broken unaided streak may be submitted to the public top-10 leaderboard with a short display name. Scores are for motivation only and have no cash or prize value.
Automated verification
The project test suite covers hand evaluation with multiple aces, S17/H17 dealer draws, pushes, blackjack and ordinary payouts, doubling, splits, split aces, long hands, exposed-card counting, pair recognition, hard and soft recommendations, legal fallbacks, and visit-API privacy controls.
Tests are run against the modular source used to build the browser bundle. A passing suite helps prevent regressions, but errors can still exist. Reproducible reports are welcome through the Contact page.
Inspecting the implementation
The game engine, recommendation table, browser controller, and tests are available in the public GitHub repository. The Editorial Policy explains how factual corrections and source quality are handled.