Choose your best five-card ace-to-five low from seven personal cards. Aces are low, and suit or straight patterns leave the low ranking unchanged. Every completed low can compete. The highest opening door card brings in; later streets begin with the strongest exposed low.

Take your seat with Fern and familiar AI company. The room offers heads-up or six-seat play, automatic next hands, adjustable pace, practice peeks, a visible WSRF tree, a Monte Carlo playground, and a notebook that keeps your place. Choose Gentle company while learning the rhythm, then explore Seasoned or Sharp whenever you feel ready.

Start by naming the game and the legal five-card construction. Then count the players still active and read the amount needed to continue. Those small observations give the coach’s estimate a useful setting. The worked pages that follow preserve their cards, assumptions, and numerical receipts so you can explore the same question again.

These are the room’s declared house settings. A short stack contributes its available chips, and side pots keep each contribution matched to the players eligible for it. A returned unmatched contribution stays separate from a contested pot. Opening the Coach or Options panel holds the action while you explore.

A bring-in can be completed to the full 20-chip small bet. Completing it reopens action to the players who entered below that amount. Fourth street uses the small increment.

Rank the chosen low from its highest card downward. A wheel begins 5,4,3,2,A. A six-low begins with 6, making the first comparison easy to see. When the highest ranks match, move to the next rank. This is the same careful comparison the evaluator makes.

Try covering the worked answer and pointing to the five chosen cards. Then name one different final card that would invite a new construction. A concrete card change is a useful way to connect imagination with the rules.

The pocket map uses your first two dealt cards. In an Omaha hand these are an explicit pair of anchors inside the four-card pocket; the hand evaluator still explores all legal two-card choices. In a stud hand they are the first two private cards. The moving map uses the latest two community cards or your latest two exposed stud cards once they are available.

The map’s circular ranks run from 2=0 to A=12. The reference expressions use the two anchors, their sum modulo thirteen, and their absolute difference. Near means distance zero or one. Seven or more near ranks gives LIKELY; six or fewer gives UNLIKELY. Both labels receive balanced examples in each newly trained model.

Known-card counts include your own cards and every publicly exposed card, including cards shown by a folded stud opponent. That keeps the remaining physical-card counts connected to what you can actually observe. The game-specific hand ranking and the map each describe their own part of the hand.

Read a low from its highest card

Write a five-card low in descending order. A six-four low begins with 6, then 4; a six-five low begins with 6, then 5. The first ranks tie, so the second rank decides their order. This left-to-right comparison turns close low hands into a clear sequence you can follow.

Distinct ranks open room

Several different low ranks create room for a five-card low to grow. A repeated rank uses another card while adding less variety to that construction. Later cards can still supply the distinct ranks you need. Track the best five-card plan rather than judging the whole seven-card collection by its least useful card.

The table remembers exposed cards

Every door card and later exposed card contributes public information. Even an opponent who folds has shown something about the deck. The room preserves those exposed cards in the adviser’s known-card set. That gives your study a useful memory: the possible remaining cards reflect what everyone has already seen.

Let each new card reshape the plan

After catching a useful card, name the low you currently have and the next rank that would improve it. After catching a larger card, look at the remaining streets and the price before deciding. The pleasure is in following a developing construction one clear step at a time.

Your exposed cards: J♣

This independently generated teaching hand preserves your own cards and every exposed card, then samples the remaining cards uniformly. Fractional ties and split pots contribute the share actually earned. The interval describes sampling uncertainty under that model; the forest’s estimate is a separate learned prediction.

Compare the action scores with the call price and the active field. They express the explicit betting policy’s use of the learned share, rather than a fitted chip-return forecast. Opening a tree lets you follow the card features that helped produce the share estimate.

Study fixture razz-1 · 4,096 samples · Independent namespace mixed-guide-independent. Cards, seeds, and measurements are preserved in the study receipt.

Your exposed cards: A♣ · 8♦

This independently generated teaching hand preserves your own cards and every exposed card, then samples the remaining cards uniformly. Fractional ties and split pots contribute the share actually earned. The interval describes sampling uncertainty under that model; the forest’s estimate is a separate learned prediction.

Compare the action scores with the call price and the active field. They express the explicit betting policy’s use of the learned share, rather than a fitted chip-return forecast. Opening a tree lets you follow the card features that helped produce the share estimate.

Study fixture razz-2 · 4,096 samples · Independent namespace mixed-guide-independent. Cards, seeds, and measurements are preserved in the study receipt.

Your exposed cards: A♥ · 5♣ · 8♣

This independently generated teaching hand preserves your own cards and every exposed card, then samples the remaining cards uniformly. Fractional ties and split pots contribute the share actually earned. The interval describes sampling uncertainty under that model; the forest’s estimate is a separate learned prediction.

Compare the action scores with the call price and the active field. They express the explicit betting policy’s use of the learned share, rather than a fitted chip-return forecast. Opening a tree lets you follow the card features that helped produce the share estimate.

Study fixture razz-3 · 4,096 samples · Independent namespace mixed-guide-independent. Cards, seeds, and measurements are preserved in the study receipt.

Your exposed cards: T♣ · A♠ · Q♦ · A♥

This independently generated teaching hand preserves your own cards and every exposed card, then samples the remaining cards uniformly. Fractional ties and split pots contribute the share actually earned. The interval describes sampling uncertainty under that model; the forest’s estimate is a separate learned prediction.

Compare the action scores with the call price and the active field. They express the explicit betting policy’s use of the learned share, rather than a fitted chip-return forecast. Opening a tree lets you follow the card features that helped produce the share estimate.

Study fixture razz-4 · 4,096 samples · Independent namespace mixed-guide-independent. Cards, seeds, and measurements are preserved in the study receipt.

Your exposed cards: T♣ · 7♥ · 2♥ · 7♦

This independently generated teaching hand preserves your own cards and every exposed card, then samples the remaining cards uniformly. Fractional ties and split pots contribute the share actually earned. The interval describes sampling uncertainty under that model; the forest’s estimate is a separate learned prediction.

Compare the action scores with the call price and the active field. They express the explicit betting policy’s use of the learned share, rather than a fitted chip-return forecast. Opening a tree lets you follow the card features that helped produce the share estimate.

Study fixture razz-5 · 4,096 samples · Independent namespace mixed-guide-independent. Cards, seeds, and measurements are preserved in the study receipt.

Your exposed cards: Q♠

This independently generated teaching hand preserves your own cards and every exposed card, then samples the remaining cards uniformly. Fractional ties and split pots contribute the share actually earned. The interval describes sampling uncertainty under that model; the forest’s estimate is a separate learned prediction.

Compare the action scores with the call price and the active field. They express the explicit betting policy’s use of the learned share, rather than a fitted chip-return forecast. Opening a tree lets you follow the card features that helped produce the share estimate.

Study fixture razz-6 · 4,096 samples · Independent namespace mixed-guide-independent. Cards, seeds, and measurements are preserved in the study receipt.

Each game receives 6,144 training positions, with 128 independently sampled showdowns at every position. Six equally represented pools combine opening, middle, and closing play with LIKELY and UNLIKELY moving-map labels. Private and public suit counts, exposed-board strength, and rank geometry give the forest a richer view. The fitted object is pytoolzx.wsrf.WSRFRegressor, and pytoolzx.indexrng.Index supplies independently named seed streams.

Compact and full forests are compared on 768 validation positions. The smaller validation error chooses the serving forest before a fresh 1,536-position test is evaluated. Validation labels use 384 samples; test labels use 512. MAE is the average absolute difference in estimated pot share, expressed here in percentage points.

The complete-hand benchmark rotates the trained policy through every seat of each fresh deck. Its intervals use deck averages to respect the relationship between seat rotations. These named local practice styles define the scope of the measurements. The full receipt preserves every game, table size, and opponent row.

Begin with a comfortable rhythm

Choose Gentle company and a relaxed pace. Leave automatic next hands enabled if you enjoy a continuous session, or choose each new hand yourself. The panel controls hold your exact place whenever you open an explanation.

Give one hand a question

Choose a question you can inspect: can this card complete the construction, can this price be expressed as a pot share, or can the tree show which input it used? The worked answer becomes useful when you can point to its reason on the table.

Explore with a practice peek

A peek makes the opposing cards visible for your exercise. The model continues to use its own cards and public observations, so its estimate stays comparable before and after your peek. The notebook marks that hand as practice and preserves the result.

Let the forest take a turn

The WSRF partner option lets the trained adviser play your seat using the explicit betting policy. Pause on a street that interests you and open the tree. Watch mode is a way to explore the model’s choices; your own chair is ready whenever you switch back.

Keep a story to return to

Download a notebook from Options. It includes the deal and action sequence needed for exact replay, along with stacks, funding, completed hands, and HUD totals. Restore it in the same game and table size. A saved story can make the next visit feel like a continuation.

VPIP counts voluntary entries on the opening street; OR counts opening raises or completions. AF divides bets plus raises by calls. WTSD counts showdowns among hands reaching the first draw street. WSD counts showdowns winning any contested pot, including ties. Read each percentage together with its completed-hand sample count.

Explore the background for this variant in Wikipedia’s seven-card razz overview. The room’s specific stakes, cap, reopening, and odd-chip conventions are declared on page 3. The examples, exercises, UI, and training experiment in this guide are original to Lantern Poker.

Explore the worked-hand receipt, complete-hand match table, and training and reproduction notes. Return to the Razz River table whenever you are ready.