How to analyze a long-narde game: PR, equity loss, and mistake patterns

After any game in Nardex you can open a full review: a timeline of the quality of every move, equity-loss values, best alternatives, and best/ok/doubtful/error/blunder labels. This guide explains what those metrics mean and how to use them — no jargon mystique, just three real PR-chart patterns.

Why review a game at all?

Long narde is a sequence of decisions under partial information. Most mistakes stay invisible to the player: you picked a move, it produced an outcome, and without comparing against the best move you have no idea whether you lost equity. Engine analysis provides that comparison: for every position it evaluates 5–20 alternatives and ranks them by equity. Your choice is scored against the best one found.

This lets you:

• Find moves where you genuinely lost a lot of position (blunders).
• Spot systematic patterns (e.g. always taking off the head with the second checker when advancing one would have been better).
• Quantify the cost of specific strategic decisions (offer the cube vs wait; push for the sixth point vs commit to the race).

What does the PR (Performance Rating) scale show?

PR is the cumulative equity-loss over the game, normalized by the number of non-forced moves (positions where you actually had a meaningful choice). Less loss = lower PR = more accurate play.

PR	Level	What it means
< 4	Master	Game close to the engine. Achievable for strong club players on standard positions.
4–7	Strong amateur	One or two doubtfuls + an occasional error. No blunders.
7–12	Average	Regular doubtfuls, one error, possibly one blunder.
12–18	Beginner	Several errors and blunders. Systematic losses.
> 18	Learning	Many blunders. It makes sense to first read the long-narde rules and strategy.

What do the move-quality labels mean: best / ok / doubtful / error / blunder?

Every move on the timeline is tagged with a label and a colour. The boundaries between labels are defined by equity-loss relative to the best move:

• bestyour move matches the engine's suggestion.
• okequity-loss up to ~0.02. Not best, but statistical noise.
• doubtfulequity-loss ~0.02–0.08. Noticeable loss, not critical.
• errorequity-loss ~0.08–0.20. A serious mistake; if repeated it changes the result.
• blunderequity-loss > ~0.20. A gross mistake, usually game-losing.

Exact thresholds are configurable and may differ between gnubg/XG/Nardex; the concept is the same.

Pattern 1: what does a clean game look like?

Here's what a well-played game looks like on the equity timeline. The line oscillates around the central midline, no spikes:

When you see a chart like this — there's almost nothing to review. Worth a glance at the isolated doubtfuls (which phase they happened in), but that's already fine-tuning.

Pattern 2: what does a single blunder in the middle look like?

The most common form of partial loss is a single mistake in a key position. Here, the position fell apart on move 8:

With a game like this, the review starts at the blunder position. Open it in the editor, check the phase, compare your move against the best move (variation arrows). It's often: a premature 6th point of a block (the no-6-in-a-row rule), an extra checker off the head on a non-double roll, or abandoning a defensive structure under tempo pressure.

Pattern 3: what does cumulative drift look like?

No obvious blunder, yet the game is lost. Example — many doubtfuls plus a couple of errors, evenly spread:

This pattern is the trickiest. The player feels they played «fine», but the engine sees that almost every move dropped 0.04–0.08 equity. The remedy is general training on tempo, the head rule, and proper checker distribution.

The doubling cube in analysis: what does the engine flag?

On every move where a cube action was possible, the engine separately flags:

• Missed double — you should have offered, you didn't (equity loss).
• Bad double — you offered too early, the opponent took to their advantage.
• Missed take — you should have taken, you passed (gave up partial equity).
• Bad take — you took a losing position, losing more than you would have by passing.

Cube blunders often hide behind a «normally played» game — so looking only at the positional timeline is not enough. More on this in «How the Doubling Cube Works».

Which mistake patterns are most common in long narde?

• Premature 6-pt prime. The player adds the sixth block point without checking that the opponent is fully behind it — and breaks the «6-in-a-row» rule.
• Two off the head outside of move one. You can take two off the head only on the very first move of the game with rolls 3-3, 4-4, or 6-6. Any other situation — not allowed.
• Ignoring a leading race. When you have a clear pip-count lead, the logical play is to commit to the race. Players often keep building a block «for beauty», losing tempo.
• Late bear-off positions. If the dice allow movement inside the home but one checker is still outside — bear-off is forbidden. Many people miss this check.

Where to go next

• Pillar — long-narde rules (the basics).
• Long-narde strategy — phase-specific priorities.
• Mars and koks — when to play for the triple.
• Developers — API for analyzing your own games programmatically.

FAQ

What is PR (Performance Rating) in long narde game analysis?

In long narde game analysis, PR is the total equity-loss in a game, normalized by the number of non-forced moves. Lower PR means more accurate play. The long-narde scale: < 4 — master, 4–7 — strong amateur, 7–12 — average, > 15 — many mistakes.

How is a blunder different from an error or doubtful in long narde?

In long narde game analysis these are equity-loss bands relative to the best move. Doubtful = small loss (~0.04–0.08), error = noticeable (~0.10–0.20), blunder = very large (≥ 0.20). The exact thresholds depend on engine calibration.

In long narde game analysis, which is worse — one blunder or many doubtfuls?

In long narde, a single blunder can cost the game, but cumulative mistakes are often more dangerous: they are harder to notice and fix. A drift chart (many doubtfuls + a couple of errors) signals that you need to work on overall level, not just one position.

How does long narde game analysis help with the cube?

In long narde, the engine flags missed doubles and doubtful takes — moves where the cube action was suboptimal. This is a specific type of blunder that beginners often miss.

Can long narde game analysis be automated via API?

In long narde — yes, through our analyze-position endpoint. See developers/quickstart for details; access via the lead form on developers.