Interpreting 1000 Baccarat Shoes: Data and Limits

These files show the outcomes, shoe lengths and streaks produced by two simulations of 1000 baccarat shoes—one with six decks and one with eight. They let you check recorded hands and calculate payouts under stated rules, but they cannot predict the next hand or promise a profit.

They are original Wizard of Odds simulations, not live casino records. We retrieved and independently analysed both complete files on October 11, 2026. That analysis is separate from the new sequences generated by the ToolsGambling simulator.

The Wizard of Odds baccarat basics page, updated August 3, 2026, provides the original simulations. To interpret them properly, you need their field layouts, dealing rules and shoe-ending procedures—not just their outcome counts.

What is a shoe, and how are hands dealt?

A shoe is the device that holds cards for dealing; in this analysis, it also means the sequence of hands dealt from a prepared set of cards before replacement. Six decks contain 312 cards; eight contain 416. The blackjack shoe explanation provides a brief physical analogy, but baccarat uses different dealing rules.

In Punto Banco, neither side chooses to hit or stand. The archived official Loto-Québec rules specify: A counts as 1, 2–9 retain their values, and 10, J, Q, K count as 0. Totals are modulo 10: 7 + 8 totals 5.

A natural is an initial two-card 8 or 9. Either side's natural stops both sides from drawing. Otherwise, Player draws on 0–5 and stands on 6–7. If Player stands, Banker draws on 0–5. If Player draws, use this table:

The PDF's metadata date is August 12, 2020; it is not a statement of every casino's current rules.

More naturals use fewer cards per hand. With a fixed cut card position, that may permit more hands, not fewer.

Where can you download the files, and how do you read them?

Both original TXT files are available directly, but their field order differs. Do not apply the eight-deck layout to the six-deck file.

• Download the original six-deck, 1000-shoe file.
• Download the original eight-deck, 1000-shoe file.

The literal outcome codes are B for Banker wins, P for Player wins and T for Tie. The placeholders x and - indicate that no third card was drawn.

The first eight-deck record is:

3,2,-,3,5,-,5,8,B

Player has 3 + 2 = 5, while Banker has 3 + 5 = 8. Banker's natural 8 stops both sides from drawing, so Player's usual draw-on-5 rule does not apply. The final B means Banker wins.

The first six-deck record is:

B,9,1,0,9,x,1,0,x

Here the outcome comes first, followed by Banker and Player totals, then their cards.

To count the outcomes, save the downloaded TXT file locally and run this Python code:

from collections import Counter
from pathlib import Path

decks = 8
filename = "1000-baccarat-shoes-8-deck.txt"

if decks not in (6, 8):
    raise ValueError("decks must be 6 or 8")

outcome_index = 8 if decks == 8 else 0
counts = Counter()

for line in Path(filename).read_text(encoding="utf-8-sig").splitlines():
    fields = [field.strip() for field in line.split(",")]
    if len(fields) != 9:
        continue
    outcome = fields[outcome_index]
    if outcome in {"P", "B", "T"}:
        counts[outcome] += 1

print(sum(counts.values()))
print(counts)

For six decks, change decks to 6 and filename to the downloaded six-deck filename. The code accepts only nine-field rows with a valid outcome at zero-based index 8 for eight decks or 0 for six decks. Headers and footers must not be imported as hands.

This is an outcome counter, not a prediction model or a complete rules validator. Streak analysis also requires preserving shoe boundaries.

What results did the independent checks find?

The files contain 143503 hands: 62624 in the six-deck file and 80879 in the eight-deck file. We independently checked every record's stored totals, outcome and third-card rules; no invalid records were found.

That confirms internal arithmetic and rule consistency. It does not validate any casino's random number generator.

Outcome percentages use all hands in the corresponding file, including ties, as the denominator. Percentages are displayed to six decimal places.

For Banker and Player streaks, ties were ignored: they neither added a win nor broke a same-side sequence. The streak counter reset at every shoe boundary. These observed maxima describe this sample, not the longest possible streak.

How do observed frequencies compare with theory?

Observed frequencies differ from theoretical probabilities because these are finite samples, not new permanent probabilities for baccarat.

The comparison below uses the eight-deck file and the Wizard of Odds full-shoe probability table, updated August 3, 2026.

The theoretical reference starts with a full set of cards. It is not the exact probability for every later hand in a partially depleted shoe.

Why isn't shoe length a clean six-versus-eight-deck comparison?

The files use different shoe-ending procedures, so their length difference is not a controlled test of deck count alone.

The six-deck file deals until fewer than six cards remain. Its burn procedure is not described; we should not invent one.

The eight-deck header describes this protocol:

• Reveal the first card and burn 1–10 cards according to its value.
• Place the cut card between the 400th and 401st cards.
• If the cut card appears mid-hand, finish that hand and deal one additional full final hand; if it appears instead of the first Player card, deal only one full final hand.

The theoretical differences between main-bet house edges are much smaller than the differences in these files' average shoe lengths.

Values are rounded to six decimal places. The six-versus-eight-deck Banker difference is about 0.002057 percentage points.

Payout variants can matter much more. An eight-deck Tie bet paying 9:1 net profit has a theoretical house edge of 4.844032%, rather than 14.359629%.

Check the actual paytable; commission-free variants do not all use identical rules. Under the standard rules used here, a Tie returns Banker and Player stakes—a push.

Main baccarat payout rows on page 3 of the archived Loto-Québec rulesPage 3 of the archived Loto-Québec rules illustrates the payout variant, not universal current casino terms. The PDF metadata lists August 12, 2020 as the file’s creation and modification date, not a verified publication date.

What do roads and streaks tell you?

Roads show the order of past outcomes, not the complete inventory of remaining cards. A recognisable pattern alone therefore does not give an exact prediction for the next hand.

The Bead Plate records outcomes sequentially. The Big Road groups same-side decisions into columns, with a green mark for a Tie. Red and blue in derived roads describe repetition or choppiness in the structure; they do not directly mean Banker and Player.

For B, B, T, B, P, ignoring the Tie leaves three Banker decisions followed by Player. The Tie is a push on the main bets, not another Banker win.

Cards are drawn without replacement, so exact finite-shoe probabilities depend on the remaining composition. It would be wrong to claim blanket strict independence or that card composition has zero effect.

However, reading side results is not card counting: it does not reveal which cards were dealt. The Wizard of Odds card-counting analysis examines composition effects and limited main-bet counting opportunities, not a universal profitable method. Side bets are outside this guide's scope.

Why are there three different probabilities for five Banker wins?

The probability depends on whether you mean the next five non-tie decisions, the next five literal hands, or a run appearing somewhere in a longer sequence.

For the eight-deck theoretical reference:

This denominator excludes ties. Banker's probability across all hands is 45.8597%.

The first uses $q^5$; the second uses $P(B)^5$. The third accounts for many possible starting positions, including overlapping runs.

These percentages, rounded to four decimal places, use an independent, constant-probability approximation. They are not exact depleted-shoe mathematics. The gambler's fallacy experiment lets you explore the constant-probability model separately.

A long streak alone does not establish bias or fraud. Equally, this analysis does not prove that every conceivable strategy fails.

How do you turn the outcomes into money?

You need payouts, commission and stake sizes—not just a win percentage—to calculate money results. Here we assume 5% Banker commission, Player paying 1:1 net profit and Tie paying 8:1 net profit.

A winning unit on Banker earns 0.95 units of profit; Player earns 1; Tie earns 8. The returned original stake is not profit. Ties return Banker and Player stakes.

Expected net result per unit staked is:

For 1000 hands—not 1000 shoes—at a fixed $25 per hand, turnover is $25000. Using eight-deck theoretical probabilities gives:

Amounts are rounded to cents. Turnover includes every stake, including pushes. Expected loss is an average implied by the model, not the loss you must experience in one sequence.

Theoretical RTP is the complement of the house edge, not the win rate. A returned stake on a Tie contributes to the payout calculation without becoming a Banker win.

What happened with $25 on every hand in the original files?

Flat betting on Player lost less than betting on Banker in both files, even though Player has worse theoretical expected value. That is the realised settlement of these particular simulated outcomes.

Banker settlement is $25 × (0.95 × Banker wins − Player wins). Player settlement is $25 × (Player wins − Banker wins). Ties cause no loss on either of those bets.

These historical simulated settlements assume an unlimited bankroll and no table cap. They are not reports of real-money gambling experience.

Why doesn't doubling guarantee recovery?

A Banker Martingale does not even guarantee a break-even cycle after a win: commission reduces the profit on the winning bet.

Seven losses at $10, $20, $40, $80, $160, $320 and $640 lose $1270. The next $1280 Banker bet earns only $1216 net profit if it wins:

The cycle finishes at −$54 despite the final win. You also need sufficient bankroll for the next stake and a table limit that permits it. Betting progressions increase turnover and can accelerate bankroll depletion; they do not remove negative expectation. Use the bankroll management guide to plan spending limits.

How can you repeat the checked simulator example?

Open the exact parameterised simulator link to reproduce Banker flat betting at $10, with a $1000 bankroll and 500 hands. This generates a new simulation; it does not replay a source-file shoe.

1. Open the link.
2. Check that Banker and Flat are selected.
3. Check the $10 base bet, $1000 bankroll and 500 hands.
4. Read the automatically updated results; there is no separate Run button.

The link supplies seed 2026, which is not an editable field. New run adds 7919 to the previous seed, modulo 2000000000. Reopen the original link to repeat this exact example.

Local simulator inputs showing Banker, Flat betting, a 10-dollar base bet, a 1000-dollar bankroll and 500 handsInputs in the local version captured on October 11, 2026. This screenshot is not evidence of production-site testing.

The checked result is a final bankroll of $1259, net profit of $259 and turnover of $5000, with 242 wins, 204 losses and 54 pushes.

This calculation includes commission and treats ties as returned stakes. All 500 bets count towards turnover.

Local simulator result showing a 1259-dollar bankroll, 259-dollar net profit, 242 wins, 204 losses and 54 pushesResults in the local version captured on October 11, 2026. One profitable 500-hand run does not validate a betting system.

The HUD's 5.18% is realised net profit divided by turnover: 259 / 5000. It is not the casino house edge. The displayed peak of $1313 rounds $1312.50 to whole dollars; the low is $924.

Any autogenerated sentence below the HUD describing this positive realised return as negative is incorrect. Use the numeric settlement instead. The theoretical expectation of the Banker bet remains negative.

What can the tool compare—and what can't it do?

The tool can compare newly generated sequences under selected parameters, but it cannot analyse uploaded historical shoes.

It uses eight decks only, in a continuous session with a reshuffle when fewer than six cards remain. It has no:

• CSV upload;
• six-deck selector;
• historical shoe analysis;
• exact remaining-card inventory oracle;
• bulk backtest of the original 1000 shoes.

Use the seed link for repetition. When comparing settings, change only one parameter at a time and inspect turnover and drawdown, not just the final balance.

The settlement above was verified for flat betting only. It is not a guarantee that every betting progression is implemented perfectly, and one positive run does not establish long-term profitability.

How should you test a strategy on the original data?

A useful test starts with predefined rules and separate, unseen shoes. Optimising on every shoe and then treating the same results as validation is historical overfitting.

1. Define the rules before testing. Specify entry conditions, betting side, stake size, stake changes and stopping rules.
2. Fix the settlement assumptions. Include commission, pushes, bankroll and table limits.
3. Split the data by shoe. Keep unseen shoes separate and do not alter the rules after viewing their results.
4. Preserve hand order and shoe boundaries. Do not let the strategy access future information.
5. Report more than win percentage. Include net result, turnover, drawdown and occasions when bankroll or table limits prevented continuation.
6. Separate historical fit from future expectation. A favourable backtest is not a promise of profit.

Even 1000 shoes cannot guarantee future income or certify fairness. Their value is transparent checking of records, rules and settlements—not discovering a guaranteed winning pattern.

The source files let you verify past simulated hands. They do not reveal everything needed to price the next hand or remove the financial limits of betting.