Long Narde Strategy: From Opening to Race

Long narde looks like a «child's» game next to short backgammon — no hits, no bar, checkers move in a single direction. In reality it is a deep positional game in which the winner is the player who understands the four phases of the match: opening (first 3 moves), build (home construction), mid-game (the blocking fight), and race endgame. This article walks you through all four phases with real positions from our expert-game database and evaluations from the Nardex neural network.

If you are still learning the rules, start with the rules of long narde. If you want to understand how long differs from short, see the «long vs short» comparison. And if you are interested in scoring (mars, koks) — the article on mars and koks explains how a single koks can flip a short match.

1. Opening — the first 3 moves

In long narde both players start with 15 checkers on one point (point 24 for White, point 12 for Black under the Russian Ministry of Sports numbering). The first move is called the «head» — only one checker may be taken off it (with the exception of special opening rolls 6-6, 4-4, 3-3, when the Ministry rules permit two checkers off the head to speed up the game).

This «head-1» rule creates the central conflict of the opening phase: you want to push checkers out of the starting stack quickly to gain control of key points, but only one checker may leave the head per turn. Every roll therefore becomes a compromise between developing the head and distributing the checkers already in play.

What are the standard priorities in the opening?

• Take point 18 (display 19) — the key «bar point», six pips from the head. It is a great anchor for building a block.
• Spread checkers across different points rather than stacking them on one: diversity makes any roll easier to use.
• Don't keep singles in the opponent's attack zone (points 7–12 and 19–24) — there are no hits in long narde, but a single slows down your structure.
• Roll 6-6 is the best: it lets you occupy two strong points at once. Roll 1-1 is the worst, because you can't move from the head 1 pip (you would land on your own stack).

A real position from our expert-games DB: each player has made 1–2 head moves and pushed scouts forward.

Here White has no advantage — the fight is even. The Nardex network rates the position at 55% to win, which in the opening phase means «still anyone's game». At this point what decides is not the layout itself but the next roll and the quality of the response to it.

2. Build phase — building the home

The long-narde home is points 1–6 (for White) or 13–18 (for Black). All checkers must enter the home before the player may begin to bear off. The build phase is the transition from «checkers scattered across the field» to «a packed structure inside the home».

Tactically the build phase is about point priorities:

1. Point 6 (for White) — the last point before bear-off. Control of it is critical in the endgame.
2. Point 5 — second in importance, because bearing off from it requires exact dice.
3. Points 4 and 3 — deep home, checkers here are «parked».
4. Points 2 and 1 — the deepest home; loading them too early wastes tempo.

The main build mistake of beginners is dragging checkers into the deep home (points 1–2) too quickly. This leaves the middle and far points vulnerable, and the opponent gets a chance to build a counter-block. The right strategy: first reinforce the upper part of the home (5–6), then descend gradually.

3. Mid-game — blocking and the fight for tempo

The mid-game is the phase that decides the match. Here the key strategic instrument of long narde appears: the prime — a wall of consecutive occupied points. A prime of length 6 is completely impassable: an opposing checker physically cannot pass through it, because the maximum single move in narde = 6 pips. A prime of length 5 is almost impassable (you must roll a «6», which happens on average 11/36 rolls, and the die must not already be needed elsewhere).

A real position from the expert-games DB where White built a 5-point prime and effectively won the game:

White controls five consecutive occupied points in the home (2, 3, 4, 5, 6). Black has amassed 6 checkers on point 13 (their own home head) and one single on point 7 — it is stuck in White's outer field and cannot pass the 5-prime. The Nardex net gives 75% for the win: in practice Black can win only if White makes a sequence of catastrophic errors.

How do you play against an opponent's prime?

The counter-instrument against a prime is your own counter-prime or back-game: keep several checkers on the head and avoid building the home too quickly, hoping for an opponent's error. This is a risky strategy (the probability of losing a koks grows), but sometimes it is the only chance.

Which tactical heuristics work in the mid-game?

• Don't break your own prime without urgent need. Pulling a checker from a prime for the sake of bearing off is usually an equity-loss > 0.05 mistake.
• Take key points first, then transit ones. Point 6 is more important than point 7.
• Count pips every 5–10 moves. If you are leading by 15+ pips and the match is in the race phase — that is a cube territory (see below).

4. Race endgame — pip count and exact rolls

The race endgame begins when the contact between the sides' checkers ends — neither side can block the other any more, and both simply race to finish bearing off. In this phase the winner is the one with the lower total pip-count and the more efficient use of dice.

A real race position from the DB: contact has ended, both sides are bearing off:

White leads by ~3 checkers and ~10 pips — a decisive race advantage. The network rates the position at 75% for White; in money play this position is an obvious double, in a match it depends on the score.

How does checker distribution affect the bear-off?

Bear-off efficiency depends on how evenly the checkers are distributed in the home. If all 15 are on point 6, any dice except a «6» bear off only one checker. If the checkers are distributed 6:2, 5:3, 4:3, 3:3, 2:2, 1:2 — almost any roll bears off two checkers. Understanding distribution is the key to a skilled endgame.

How do you count pips and when do they justify a double?

Pip-count is calculated like this: for each checker add the number of its point (for White) or the corresponding number (for Black) to the sum. Checkers already borne off count as 0. The pip-count difference is the key race indicator. Empirical thresholds for the cube:

• Advantage of 0–7 pips — not yet a double, neutral zone.
• Advantage of 8–12 pips — double (if match length > 1).
• Advantage of 13+ pips — double, and «take» becomes questionable.
• Advantage of 20+ pips — double, «pass» (drop).

5. How does the doubling cube shape strategy?

The doubling cube in long narde appears only in match play (matches > 1 game). The Russian Ministry of Sports rules (Article 23, 2024 edition) officially included the cube in tournament narde. The cube multiplies the stake of the game by 2 after each double; the cube owner = the side that accepted the last double, and only they may double again.

Strategic moments every match player should know:

• Crawford rule: the game in which one side reaches «match point minus one» is played without the cube. This is an additional rule for tournament matches.
• Holland rule: the cube becomes available only after two complete moves by both sides. A defense against «cube vulturing» in very short games.
• Post-Crawford: the trailing side should double on any advantage, because otherwise they simply cannot win.
• Take point (the threshold for accepting a double): on average ~25% of winning equity. If your win-probability is below 25% — drop; above — take.

The Nardex AI makes cube decisions based on the same neural-net evaluator as move decisions. The key logic lives in the game/bot_doubling.rs file; equity-thresholds were calibrated over 100K+ self-play games.

6. What mistakes do beginners make most often?

1. Move «second checker off the head». If the first roll is not 6-6/4-4/3-3, the second move off the head is forbidden. Beginners often try and get an «illegal move».
2. Premature deep-house build. Bringing checkers to 1–2 in the opening severely limits flexibility. The deep home is needed only in the bear-off phase.
3. Greed for points. You want to grab any free point, but if it doesn't help blocking — you only scatter your checkers.
4. Ignoring the counter-prime. If the opponent is building a 4-prime and you see your checker may get stuck, don't move it there. Better to sacrifice tempo than to land behind an impassable wall.
5. Bear-off off-plan. In the race phase what matters is not the speed of removal but the distribution of the remainder. Bearing off from point 6 while leaving 5 checkers on 6 is essentially sacrificing the tempo of the next move.
6. Ignoring the cube. Many amateurs are afraid to double «what if the opponent accepts». In matches from 5 points, refusing to double at 70%+ equity is statistically worse than doubling and accepting the drop.

7. How does the AI evaluate a position?

Nardex trains a neural network to predict the probabilities of six outcomes: P(win normal), P(win mars), P(win koks), P(lose normal), P(lose mars), P(lose koks). These probabilities are encoded non-standardly — as nested ordinals: win includes all win types (normal + mars + koks), win_m = P(win ≥ mars), win_k = P(win ≥ koks). The «probabilities» do not sum to 1 in the usual sense; the real difference is in the bonus-win nuances.

Equity (the mathematical expectation of the outcome in points) is calculated as:

equity = (win + win_m + win_k) − (lose + lose_m + lose_k)
       = (2·win − 1) + win_m + win_k − lose_m − lose_k

where lose = 1 − win. Equity exactly 0 = a drawn position, +1.0 = a guaranteed one-point win, +2.0 = a guaranteed mars, +3.0 = a guaranteed koks. On the roll any opening position is in the ±0.05 range — a draw. By the race phase equity usually spreads to ±0.5 and beyond.

All examples in this article are real positions from the narde_positions DB with real eval numbers from our network. If you want to integrate the same analyzer into your project, see Nardex Analysis API — it returns the same probability distribution + ranked alternatives + grade-label for each position.

Frequently asked questions