PR measures decisions rather than results
You can win a match with a worse PR than your opponent because dice affect results. Equally, you can lose while making the more accurate choices. PR tries to describe the decisions made with the information available, rather than reward the final score.
A low PR is not a percentage chance of winning, and zero does not mean a player will win every game. It means the analysis found no measured loss under its own decision set and evaluation settings. Different engines or analysis depths need not agree exactly.
How BlitzGammon derives its PR
BlitzGammon’s GNUBG statistics parser takes the absolute value of the reported error rate in mEMG (thousandths of an equivalent money game) and divides it by two to obtain its XG-style PR. It uses that error-rate value, not the separate match-winning-chance percentage printed alongside it.
This is a conversion of GNUBG’s reported rate, not a claim that BlitzGammon reproduces another program’s analysis exactly. Evaluations, decision counting and settings can differ between tools. It is also not the percentage of your moves marked as mistakes.
The app assigns descriptive performance labels from the resulting PR. Treat those as summaries of that analysed performance, not permanent qualifications. This guide does not introduce a new set of rating bands or change any existing classifications.
The size of an error matters
PR responds to estimated equity loss, so counting mistakes alone is misleading. One very costly cube decision can outweigh several small checker-play inaccuracies. Two matches with the same number of flagged moves can consequently have very different PRs.
Look inside the report before deciding what to practise. If cube errors dominate, memorising more opening moves will not directly address the problem. If most loss comes from leaving contact too early, work on race assessment and holding-game timing.
Short samples are noisy
A short match may contain very few difficult choices. Another may contain repeated cube decisions and complex contact positions. A single blunder has a larger influence when spread over a small number of assessed decisions.
Do not average a short match and a long match as though they necessarily represent equal evidence. A meaningful combined measure should account for the analysis method and decision counts. Even a longer series needs comparable settings and a reasonably representative mix of positions.
Use a trend, not a target for every match
Study your performance across many sessions and connect changes to specific habits. Are you missing fewer doubles? Are you recognising a race transition sooner? Those observations explain an improving PR more usefully than a single best-ever number.
Avoid protecting a good PR by playing only easy positions or treating every close engine disagreement as a failure. The aim is stronger decisions in the positions you actually face. Review the largest recurring losses, practise them and assess progress over time.