Every figure, defined.

Reading the figures

Amounts are in big blinds (bb). A figure the analysis estimates is shown with its 95% range: +6.6 bb ±28 means the estimate is +6.6 bb and, at the runouts it rests on, the true value lies within 28 bb of it 95 times in 100. The range measures how many runouts an estimate rests on: it is not a judgement of the model, and it narrows as more runouts are drawn (four times the runouts halve it).

Where a range is too wide to tell two things apart, the screen says so (Too close to call, within noise) instead of picking one.

Your stats (the tracker figures)

Read straight off the hands, with no model, as any tracker reads them. Each frequency is a share of the hands where it could happen (its chances, shown beside it).

• VPIP: put money in before the flop by choice (a call or a raise; posting a blind and checking is not).
• PFR: raised before the flop.
• 3-bet: re-raised when facing exactly one raise before the flop, at the first such decision.
• Fold to 3-bet: opened, was 3-bet, and folded at the next decision.
• C-bet: as the last raiser before the flop, bet the flop at the first decision there, nobody having bet before.
• Fold to c-bet: faced a c-bet and folded at the next decision, nobody having raised in between.
• Steal: raised first in from the cutoff, button or small blind, everyone before having folded.
• Fold to steal: in a blind, facing a steal with nobody else in, folded.
• WTSD (went to showdown): of the hands where the flop was seen, the share that reached a showdown.
• W$SD (won at showdown): of the showdowns reached, the share that won money.
• AF (aggression factor): after the flop, bets and raises per call.
• AFq (aggression frequency): after the flop, bets and raises as a share of all decisions but checks.
• Win rate (bb/100): big blinds won per 100 hands.
• All-in EV: the same result with every all-in before the river, where everyone's cards are known, paid at the cards' exact equity over every board that could still come, instead of the board that came. The gap between the two is how you ran.

The hand list

• Pos: your seat: UTG, MP, LJ, HJ, CO, BTN, SB, BB.
• Pot type: by the raises before the flop: limped (none), single-raised (one), 3-bet pot (two), 4-bet+ pot (three or more).
• Pot: the final pot. Street: the last street you saw (Folded pre, Flop, Turn, River, Showdown).
• Result: what you ended the hand with less what you began it with.
• EV lost: the sum of your analysed decisions' EV lost in the hand (below), with its range.
• Filters, besides those columns: before the flop, raiser if you made the last raise there, caller if you put money in by choice without it, neither otherwise; and for hands where you saw the flop, against (the positions of the others still in) and in position (you act last among them after the flop). Hands where you did not see the flop pass neither of those two.

Reports and sessions

Read straight off your hands, with no model, as a tracker reads them. Each rate is a count over its chances, and the count is shown beside it; a rate with no chances is a dash, never zero.

• Results by position, pot type, stakes and month: the hands, what they won, and per 100 hands. Stakes are named by the big blind (NL50 is $0.25/$0.50). Month is the month the hand was played: the day the site's session names, else the day it was recorded in Parallax. With all-in EV, each all-in before the river is paid at its equity, as in your stats.
• Starting hands: the 169 classes (AA, AKs, AKo...), each with how many times you had it, what those hands won, and Played: how often you put money in before the flop by choice. Only hands where your cards are known count.
• Streets: every decision you made on each street, as a fold, check, call, bet or raise. A bet is the first bet on a street after the flop; before the flop the blinds are the bet, so every raise there is a raise.
• Bet sizes: your bets after the flop as a share of the pot before the bet, in buckets (up to 33%, 33–50%, 50–67%, 67–80%, 80–100%, overbets), each upper bound included. Won in them is the whole result of the hands you bet that size in, not the bet's own: a hand counts once in each bucket it bet in.
• River: the hands you saw the river in, by the final pot in big blinds, with how often you went to showdown, won there, and bet or raised the river.
• Showdowns: the hands you showed with a full board, by the best five-card hand of your two cards and the board, and what each won on average.
• Ranges: a range's combinations are the two-card hands it holds (6 for a pair, 4 suited, 12 offsuit, 1,326 in all), with every card on the board taken out of the deck first, so its share is of the hands still possible. On a board of three or more cards, each combination is counted under the best five-card hand it makes there.
• Sessions: your hands grouped by the site's session, or by day where a hand has none, with the result and the all-in EV: the gap between them is how the session ran.

Runouts

An analysis plays each action you could have taken on to the end of the hand many times: a runout. In a runout, the other players act as Parallax models them (each one's own style once they are profiled, the pool's otherwise), and the board is dealt at random. A figure about an action is a figure over its runouts.

A runout says who won and the pot, never the cards another player showed in it: those cards are not shown anywhere a runout or a line you simulate reaches a showdown. The showdown recorded in your own hand is real and shows as recorded.

How many: every action at your decisions is played out 32 times when a whole hand is analysed, every action at everyone else's 8 times. Analyse spot plays one decision out more (32 by default, up to 256).

EV of an action

The EV of an action is the average result of its runouts for the player taking it: the stack each runout ends with, less the stack the hand began with. Its range comes from how much those runouts differ from each other.

Freq is how often the player is expected to take the action there, as Parallax models them.

Which lines are judged

An action is judged when none of its runouts ends in a called all-in, and none has a bet or raise of at least the pot on the turn or river that gets called. Parallax does not price those lines reliably yet: in them, the other players call big bets more often than real players do, and with hands that do not fit the call, so such a line comes out too profitable.

An action that is not judged shows Not judged yet instead of its EV, and it is never compared: Best, the verdict and EV lost are read over the judged actions only (Best of the lines we can judge). A decision gets a verdict and an EV lost only when the action you played is judged and at least two actions are. Leaks and the dashboard count only those decisions, and say how many of yours were set aside.

Hands analysed before this rule show Analyse again until they are analysed again.

Best, and Too close to call

Best marks an action only when it beats every other action by more than the noise in each difference, with all the comparisons held at 95% together. When it does not, the actions whose ranges overlap the top one are marked Too close to call: the runouts cannot tell them apart, and nothing is called best.

EV lost, and the verdict on each decision

EV lost at a decision is what the action you played gave up against the best judged one (see which lines are judged). Taking the action with the highest average and subtracting yours would overstate it: with noisy averages, the highest is usually one that got lucky. So the runouts are split in two halves: the best action is chosen on one half and measured against yours on the other, then the halves swap and the two are averaged. The half that measures never chose, so the figure does not flatter the best action: on average it can understate a loss, never overstate one. It is not rounded up to zero when it comes out negative.

Each decision you played gets a verdict from that figure's range:

• Clear mistake: the whole range is a loss: some action was really better.
• Clearly fine: even the top of the range is under 2 bb: a gap you could not act on at the table.
• Within noise: neither can be said at these runouts.

Leaks

Your judged decisions grouped by the kind of spot: the street, what you faced (before the flop: unopened, a limp, an open, a 3-bet, a 4-bet or more; after it: no bet yet, a bet, a raise) and, after the flop, your role (the preflop raiser, a caller, a limped pot).

A leak's figure is the sum of its decisions' EV lost; its range combines theirs (the runouts of different decisions are independent, so the ranges add as the square root of the sum of their squares, and a total over many decisions is far tighter than any one). Per 100 is per 100 analysed hands. A leak is clear when its whole range is a loss, within noise otherwise.

Drills

A drill is one of your decisions that has a verdict, from your 40 most recent analysed hands, asked again with everyone else's cards hidden. Only the judged actions are offered, in the table's order (fold, check or call, then bets and raises from the smallest), never best first. Your answer is graded against the judged action with the top EV: best if it is that action; too close to call if the gap is within 1.96 times its standard error (the two EVs' standard errors combined as the square root of the sum of their squares); behind otherwise. The gap is the plain difference of the two EVs: unlike EV lost, it is not corrected for picking the top of several noisy estimates, so read it beside its range.

Key spots

The decisions that moved the hand most for the player acting: the EV of the action played less the spot's value before the decision (every action's EV weighted by how often the player takes it). Each comes with its range. A better action is named only where it is clearly best.

Opponents

• Type, read off VPIP and PFR from 30 hands: Maniac (VPIP over 45% and PFR over 30%), Nit (VPIP under 15%), Loose-passive (raising under 45% of the hands played, VPIP over 28%), Passive (raising under 45% of the hands played), LAG (VPIP over 28%), TAG (the rest). Fewer than 30 hands: no type. The list filters by type and by the hands you have on a player.
• Profiled: from 20 of a player's hands, Parallax models that player's own style, and runouts and practice tables play them as themselves; it learns them again every 20 more. Until then they play as the pool.

What the figures do not cover

• Big bets that get called. Lines where a big bet gets called, on any street, are not judged yet (see which lines are judged), so decisions where every good option leads to one, as many preflop decisions do, get no verdict for now.
• The range is runout noise only. It does not measure how well Parallax models the players: that is the model's accuracy, which we measure separately and do not fold into the range.
• How strong the hands shown at a runout's showdown are: last measured against real showdowns on 21 October 2026, close to real players' after a river bet and a little weaker after a river check or call, so a river decision against a checked-down range can look slightly better than it is.
• Small samples: a win rate settles over thousands of hands, the frequencies over a few hundred. The screens say how many hands each figure rests on.
• Stakes: no-limit hold'em cash from $0.25/$0.50 to $5/$10 big blinds (NL25 to NL1000); hands outside them are not analysed.