Reading your analysis and your leaks

On this page

• Find your way around the overview, the breakdowns, the reference comparison and the leaks.
• Express results in big blinds per 100 hands to compare sessions and spots fairly.
• Rank leaks by the big blinds they cost, not by how far a frequency is off.
• Tell a leak built on a thin sample from one you can trust.
• Open the hands behind your biggest leak and replay them.

Concepts used: EV, EV loss and how Rail grades

The overview: one number for how you play

The Analysis overview puts your whole sample on one screen: your score, your EV loss per 100 hands, how your moves split across the five grades, and coverage, meaning how many hands were fully analysed, partly analysed or not at all, and why.

EV loss per 100 hands is the number to watch. It adds up the EV you gave away in every graded decision and scales it to 100 hands, so samples of different sizes compare fairly.

Say you lost 36 bb of EV over 1,200 hands. That is 12 hundreds of hands, so 36 / 12 = 3 bb per 100.

Breakdowns: where it goes

The breakdowns split the same numbers by street, position, pot type and preflop scenario, with the same filters as your stats: dates, stakes, room and game. They answer the first question worth asking: is the EV leaking preflop, on the flop, from one seat?

Leaks, ranked by what they cost

A leak is a situation plus one wrong turn in it: the street, the scenario and the seats, and an action that differs from the reference's best, or the right action at the wrong size. The leak finder ranks them by the total EV they lost, in big blinds.

Total EV is the right order because it counts both how bad a mistake is and how often you make it. A small mistake in a spot you meet every session can cost more than a big one you made once.

Frequency differences on their own are not leaks. If you take a mixed action more often than the reference but every one of those decisions graded Perfect, nothing was lost and the finder leaves it out. Leaks smaller than 0.05 bb in total are left out too.

How sure is the finder?

A leak built on three hands might be three unlucky spots, not a habit. So every leak carries a confidence level based on how many decisions you made in the spot and how many were mistakes.

• High confidence: at least 50 decisions in the spot and 5 mistakes.
• Medium confidence: at least 20 decisions and 2 mistakes.
• Low confidence: fewer than that. It could be a single hand; open the hands before you change anything.

Spots with very few decisions are merged into a broader group, such as the same scenario from other seats, so they can still say something. Trends that compare periods say "too few to tell" when there is not enough play to separate a change from chance.

From report to plan

The study plan takes the three costliest areas the sample can vouch for, of medium or high confidence. Only if there are fewer does it fill in with thinner ones, marked "tentative", with a note to review the hands first.

Each area comes with concepts to read, a trainer session set to the spot, and hands to review. That closes the loop: the report tells you which lesson to take next.

Rules of thumb

• Watch EV loss per 100 hands: it compares samples of any size.
• Work on leaks in order of total EV lost, not of how painful a hand felt.
• A frequency difference matters only when it costs EV.
• Open the hands behind a low-confidence leak before you change your game.

When they break

• Spots the analysis does not cover are not in the totals, so a clean report can hide a leak Rail cannot measure yet.
• A leak ranked low because you rarely reach the spot can still matter once you play more of those spots, for example at a new stake or table size.