The edge is positive. Your hand arrives, and every available action has negative EV. You choose the best one and lose. What happened?
Follow this constructed example, calculated with the Advantage Count v1 engine. We compare expected value (EV) at each decision stage with one chosen final result. The shoe and dealer reveal are teaching inputs, not a recorded session.
Before the deal, many different hands are still possible.
Our example starts halfway through an eight-deck shoe: 208 cards remain, running count is +20 and raw true count is +5. The remaining rank counts and rules are listed below.
The completed Player Edge simulation returns +2.22%, with a 95% simulation interval from +1.72% to +2.71%, using 200,000 simulated rounds.
That percentage estimates average net return per original wager across possible deals. Each simulated round starts from the same remaining cards and follows the selected playing policy.
If the units are unfamiliar, start with EV vs win probability. Expected value combines possible net outcomes with their probabilities.
02 / THE INFORMATION CHANGES
Now you have one particular hand.
Enter Your hand → 10, Dealer → 10, then Your hand → 8. You have hard 18 against a dealer 10. These three visible cards leave the unseen composition; the dealer’s hidden card stays unknown.
Under this example’s Ace peek only rule, the dealer does not check a ten upcard for blackjack. That possibility remains in the calculation. Surrender and Cash Out are disabled.
| Legal action | EV | How to read it |
|---|---|---|
| Stand | −0.2845 | Highest EV of the available actions |
| Hit | −0.7009 | Lower expected return |
| Double | −1.4018 | Lower expected return; two units at risk |
The favourable pre-deal average included other possible deals. You now know which hand you received, and all its available actions have negative EV.
These values belong to this remaining shoe and dealer procedure. A general 18-versus-10 table may assume different cards or a completed peek. Published EV tables also distinguish dealer-blackjack assumptions.
03 / ONE OUTCOME ARRIVES
You stand. The dealer reveals a 10.
Record Stand, then add a 10 to Dealer cards. The dealer has 20 and stands. Your 18 loses, for an actual net result of −1 unit.
- Pre-deal estimated return+0.0222 units
- EV of standing after the deal−0.2845 units
- Result after the reveal−1 unit
These are three readings at different stages, not three amounts to add together. Only the final −1 unit is the settled outcome of this round.
The later card tells you how the round ended. It was not known when you chose Stand. Judging that earlier decision means using the information available at the time.
Would a dealer bust make Stand a better decision?
It would make the result better: +1 unit instead of −1. The EV at the earlier decision would still be −0.2845 for the same supplied state. A different reveal changes the outcome, not what was known before it.
04 / CONNECT IT TO THE SCREEN
Read the stage and the calculation label.
- Player Edge: before the round.
Read the completed estimate, ROUNDS and 95% RANGE. Once a result has been captured and the deal starts, its details say Locked pre-deal. It describes that earlier shoe state.
- Decision: the current hand.
Open Strategy and EV details and compare legal actions. This example uses the completed FULL result marked EXACT. An Auto result can be provisional while a fuller calculation runs.
- Round result: what actually happened.
Complete the dealer’s cards and inspect the settled result. Use it to record the outcome; evaluate the action against the EV available before the reveal.
FULL does not mean the same calculation everywhere. Player Edge remains a simulation of a playing policy. The supported single-hand FULL route used here evaluates the finite composition exactly. See how strategy policies and action EV differ.
The Player Edge profile here is Extended, marked EXTRAPOLATED: the policy is extended beyond its source’s deck count. Read that qualification with the estimate.
Which policy does this Player Edge example simulate?
Its six-deck source is extended to eight decks, with incompatible no-peek rows filtered. The simulation includes its positive-EV Insurance policy and excludes Cash Out and other-seat information.
05 / THE USEFUL TAKEAWAY
Keep the decision and the result in view.
For a useful review, keep the rules, remaining cards, available actions, calculation label and chosen action together. Then record how the round ended. That lets you distinguish a lower-EV choice from a losing outcome after the highest-EV choice.
One losing hand cannot validate or disprove the model. Neither does a winning hand establish that the action was best. Reproduce a surprising comparison with the same inputs and inspect its assumptions.
Inspect the example’s rules and starting shoe
Eight decks; S17; 3:2 blackjack; Ace peek only; all bets lose to dealer blackjack; double on any two cards; double after split allowed; at most two split hands; split aces allowed, with no resplit, hit or double of split aces; no surrender or Cash Out. Penetration setting: 75%. Extended profile; floor TC; half-deck estimation.
| Rank group | Cards remaining |
|---|---|
| A | 18 |
| 2, 3, 4, 5, 6 | 14 of each rank |
| 7, 8, 9 | 16 of each rank |
| 10, J, Q, K | 72 combined ten-value cards |
To reproduce the numbers, use these remaining rank counts as well as the rules. A fresh shoe with the same player hand is a different input. A matching TC alone is also insufficient.
The Player Edge run uses 200,000 rounds and deterministic seed 3277297416. The hand uses the production FULL single-hand route. Values are rounded for reading. The dealer’s final 10 was selected to illustrate settlement, not predicted by the calculation.
Sources and further reading
The numerical case study was calculated with the Advantage Count v1 engine on 13 September 2026. These sources support the general concepts; they are not the source of our example’s numerical results.