Ride Free (Free Bet Blackjack)

1The rules, as data

Geoff Hall's Free Bet Blackjack: the house funds your splits and doubles, and pays itself back with a pushing dealer 22.

Ride Free is a floor name for Free Bet Blackjack. You split any pair except tens for free and double hard 9, 10, or 11 for free, and when a free bet wins the house pays as if your own money were out. In exchange, a dealer total of 22 pushes every live hand. The configuration behind every number here: six decks, dealer hits soft 17, free splits to four hands with aces split once, free re-splits, and free doubles allowed after splits.

The engine tracks player-funded and house-funded wagers as separate ledger entries, because a free wager pays winnings only and vanishes on a push or a loss. That split ledger, built for this game, is the reason the side bets on other pages settled without new machinery.

2The published-figure gate

The engine matches the published edge, the dealer-22 frequency, and a published strategy delta that exercises the whole free-split process.

Dealing off the top, with no cut card, free-plus-basic strategy measures a 1.063% ± 0.043 house edge against the published 1.04%, which is agreement. The dealer 22 frequency lands at 7.354% against a published 7.3536%. A third check exercises more of the game at once: the cost of splitting 5s instead of taking the free double, a published figure that depends on resplit chains, post-split eligibility, and the 22 push together, measured at +3.080% ± 0.185 against the published +3.019% over ten independent runs.

One number here I still can't explain, and it stays on the page for that reason. Dealt to a cut card the same game reads 0.99%, a little below the published figure and below its own off-the-top reading. The obvious suspect is the cut-card effect, the small extra cost of dealing to a fixed depth, and for this game that suspect has an alibi: measured directly it comes to −0.06% ± 0.06, near enough to zero, so the standard game's version of that effect does not carry over to this variant. I checked the strategy explanation too, and playing exact-composition instead of the published chart is worth −0.02% ± 0.04, meaning nothing. Two tidy explanations, both refuted by measurement, and a small gap left standing.

The removal-effect machinery got its own check. Before deriving anything for this game, its standard-blackjack output had to reproduce the classic table in Peter Griffin's The Theory of Blackjack, and it did, in sign, rank order, and magnitude.

3The oracle

A composition-exact calculator prices every decision and every overlay. Each ceiling on this page comes from it.

The oracle prices any decision at any exact shoe composition. Played live against a fixed-chart opponent on paired shoes, perfect composition play is worth +0.12% ± 0.06 per round overall, rising to +0.32% ± 0.06 inside the betting window, where 3.4% of rounds change an action. Composition-aware insurance, taken exactly when the tracked shoe makes it positive, adds +0.15% per played window round.

Those two overlays plus the certified count stack to about +1.06% ± 0.11 per played round. That figure is a ceiling twice over, assuming exact weights and perfect information about every dealt card, so the certified core in step six is the load-bearing number and the stack is the limit it approaches.

4The state dataset

The instinct this game invites, betting when free bets are coming, is backwards, and measuring why produced the count.

Free doubles want small cards and free splits want pairs, so the natural plan is to bet when those events are likely. Measured over the recorded grids, the correlation between free-double probability and the hi-lo true count is −0.937: the shoes where the free stuff flows are small-card shoes, which are bad shoes. EV falls as either event signal rises, and both signals used alone are inverted proxies for the count.

The real structure appeared in the difference against a standard game on the same signals. Free splits add 2 to 3% of relative EV in pair-rich shoes, and pair-richness at a fixed hi-lo count first measured as a real signal, worth about +0.6% per percentage point of pair probability. That thread, pulled, exposed the finding underneath: hi-lo itself is the wrong ruler for this game, and a count derived from the game's own removal effects absorbs the pair signal entirely.

5The fit, and the count

The first removal-effect table for Free Bet Blackjack. The ace is worth three times the ten, and hi-lo's small-card backbone loses half its weight.

The table below is the analytic effect of removing one card of each rank, run through the same machinery that reproduced Griffin. The structure is nothing like standard blackjack's. The ace deepens to −0.64 because it feeds naturals and the free ace split. The ten halves to −0.23 because dealer 22s are made of tens and ten pairs get no free split. The 3, 4, 5, and 7 collapse because the hands they feed now play on house money, and the 8 flips negative outright on the strength of its free split.

The count is the first-order expansion over those effects, expressed directly in EV units. Once it exists, the earlier signals fall in line: the pair effect at a fixed RF count measures −0.08% ± 0.19 (null), and hi-lo's residual slope at a fixed RF count is +0.02% ± 0.04 (null). One linear count carries everything measurable in this game, which also answers the escalation question of this step: the concentration signal that looked quadratic was linear information hi-lo happened to mis-price.

6Certified live

Certified on 21 million clean rounds, after an audit destroyed the first certification.

The first pass at this number, +1.04% per played round, died in an audit: the supposedly independent seed shards shared 98% of their shoes, so the threshold had been graded on its own fitting data. I kept the audit, the seed-hygiene fix, and the re-run, and the corrected figure is the one this page carries. Entering at a predicted shift of +1.25% or better selects 6.6% of rounds and earns +0.59% ± 0.09 per played round, pooled over 21 million fresh-seeded rounds. The looser +0.75% trigger selects 14.5% of rounds at +0.21% ± 0.08.

Calibration on the clean data reads 0.93 to 0.97 realized-on-predicted, so the linear model prices its own trigger nearly at face value. One camouflage claim also fell here: 96.8% of trigger rounds coincide with a hi-lo true count of +2 or better, so a Ride Free back-counter looks like any other counter to a floor running a standard count on the same shoe.

7The ledger

Seated play loses with every spread priced, and the winning shape is in or out. The flat bet is the ramp.

Sitting through every round costs about 1.1% a round of waiting on a game where roughly one round in ten is worth betting, and no realistic spread climbs out. The best seated system, the RF count with a 1-to-8 spread, still runs −0.37% on money wagered, and hi-lo seated is worse at −0.51%. The one profitable mode is standing behind the table and sitting down only on the 6.6% trigger.

The scenario bets flat, and that shape earned its place the same way everything else here did. A graduated ramp needs a middle worth betting, and this game doesn't have one. The certified pair of triggers prices it by subtraction on the same pooled rounds: the +0.75% trigger selects 14.5% of rounds at +0.21%, the +1.25% trigger selects 6.6% at +0.59%, so the band between them, 7.9% of rounds, comes out at −0.11% ± 0.17 per played round. Laddering in early adds volume and no measurable edge. What survives is a $0-to-$100 spread with nothing in between, the standard page's crouch taken to its limit. The scenario below is therefore the most credit this game can be given, money on the felt only when the edge is positive and none anywhere else, and the income it produces is still a few dollars per 100 rounds.

At a $100 flat bet the certified core is worth about $3.9 ± 0.6 per 100 observed rounds, before the deviation and insurance overlays that could roughly double it for a player carrying the full composition. Bankroll and N0 are marked open in the scenario card because I never measured the swing on this exact play. This is the one page whose ledger reports a certified edge without a certified bankroll, and the verdict explains why it was left there.

8The verdict

A certified edge that fails to pay for the trouble of collecting it.

The call is not-worth-it because the edge is real and the income isn't. The certified figure assumes exact weights, no integer card was ever certified, and the money at ordinary stakes is a few dollars per 100 observed rounds. A standard game next door offers 19.8% of rounds at +1.09% against this game's 6.6% at +0.59%, so the same back-counting legs earn about five times more a table away. That comparison is why the bankroll work stopped here and moved.

What the game gave me shows up on every other page: the engine, the split money ledger, the removal-effect machinery, and the lesson that a variant needs its own count.

§What this verdict depends on

• Six decks, H17, free splits to four hands with aces once, free re-splits, free doubles on hard 9/10/11 after splits. Validated against the published edge for exactly this configuration.
• Cut-card shoes at 0.75 penetration. A continuous shuffling machine kills the counting mode, and deeper cuts help it.
• Back-counting needs standing room and mid-shoe entry.
• The certified count uses exact weights, and the deviation and insurance overlays are perfect-information ceilings. Every figure that depends on those says so where it appears.
• The tracker assumes all dealt cards become visible, including the dealer's hole card.
• The bankroll for the back-counting mode was never priced. The certified edge is real, and anyone acting on it would need that number first.

Last updated 2026-08-10.