Private beta — analyser already running

Your hand history already knows why you're losing.

Pokerframe replays every hand you have ever played, ranks the leaks costing you the most, and — unusually — shows you the decisions you got right, so a session reads as more than a list of faults.

Want to say more than that, or ask something? hello@pokerframe.com

opening range — button · example player

in line71

missed opens12

loose opens12

raises should raise, didn't raised a fold

leak foundweak aces foldedA4o · A3o · A2o

played well+24.5 bb98s · three streets of value

124

fields recorded per hand

tracked statistics

per-hand review checks

169

starting hands profiled

What it does

A tracker tells you what happened. This tells you what to fix.

Every hand is re-simulated on import — pot sizes, equities, positions and lines — so the analysis is already done before you open a page.

01Measure5

Where you stand, against what it should be.

◎Leak detection#dashboard ▦Range comparison#dashboard ◈Opponent profiles#field ◉Your field#field ♠Tournament & cash leaks#formats

02Review4

Every hand replayed, decisions judged apart from results.

⚑Hand-by-hand review#review ⚖Decision scoring#after ▲All-in EV & luck#review ◷Volume & sessions#deeper

03Plan3

One thing to carry into the next game.

⌖Fix this first#after ▤Session reports#after ≈Recent form#after

04Practise2

Your own mistakes, handed back until they stick.

⟳Drill mode#drill ▨Postflop sets#drill

↻ Practise feeds the next import, and the loop measures whether anything moved. Fourteen features, each explained in its own section below.

The first screen

A percentage means nothing until you know what it should be.

Every tracker can tell you that you 3-bet 1.8%. The screen you land on tells you that number is wrong, by how much, and how many hands stand behind the verdict.

▤

Headline

Fourteen tiles: volume, result and the core frequencies — and the ones that are out of line say so themselves. Anything sitting outside its reference band flags itself red, so you don't need to remember what a healthy number looks like.

hands5,01079 per hour

net+227 bb4.8 bb/100

VPIP25.3%played preflop

PFR10.3%raised preflop

VPIP−PFR gap15.0too passive

3-bet1.8%5/277

WTSD54.7%132/241

Filter by position, stake depth, site or date and every tile recomputes against the same sample.

◎

Review these hands

A short list chosen for you, and every pick states why it was chosen — a detected mistake, a shove that went in far behind, or an example of the leak costing you most right now. Newest first, so it reads like a session debrief.

A5o UTG −47.9 bb

◎All the money went in on the turn with 0% equity

QQ CO +67.4 bb

↯Example of your top leak: WTSD is too high

Scope it to your current filters, or to just the session you imported a minute ago.

▭

Against the reference bands

Each meter fills to your value with the reference band drawn straight onto the track. Inside the bracket needs no thought. Outside it gets named in plain words, next to the sample size that earned the verdict.

VPIP25.3%

ref 20–28% fine5010 hands

3-bet1.8%

ref 7–12% too low277 spots

WTSD54.7%

ref 24–32% too high241 hands

Bands come from solver-derived and field-tested ranges. The ones that move with stack depth are measured only where they mean something — see formats.

◹

Cumulative result

The running total in big blinds, hand by hand. Hover any point to open the hand that made it — the graph is a way into your history, not just a shape to look at.

Chips and cash are both normalised to bb, so the tool tells you when mixing formats makes the line meaningless rather than drawing it anyway.

The difference

It credits the hands you played well.

Every mistake it looks for has a counterpart, so the same situation is judged on the same axis whichever way it went. Miss a river value bet and it says so; take it and get paid, that counts too.

leak Missed river value with a full house −12.4

played well Three streets of value with a set +67.4

leak Called the river with two overcards −41.4

played well Correct river fold, saved 5bb +5.0

played well Thin value a worse hand called +36.4

Credit for the decision, not the result. A correct call that missed still counts; a lucky punt does not.

Hindsight is labelled. Findings that rely on cards only revealed at showdown are marked as patterns to study, never as proven errors.

Every claim carries its sample. Denominators travel with every percentage, everywhere.

Luck is counted on its own line. All-in equity at the moment the chips went in is set against what actually landed, so a cooler reads as variance, not a leak.

Open the replay from anything. Every finding, stat and period drills through to the hand it came from.

After the session

Your biggest loss is rarely your worst decision.

Every tracker sorts your hands by what they cost. That answers where the money went, which is not the same question as where you played badly — and the two lists barely overlap. So there is a second sort that never looks at the result at all, a per-session report that refuses to say more than one session can support, and a form check that calls most movement what it is: noise.

worst played · decision only

20 K9o limped at 9.5bb, then in behind −8.2

36 ATo limped at 12.3bb, missed river value +7.3

biggest losses · the usual sort

93 JJ all-in at 66% equity −30.9

88 AA good flop c-bet, in ahead −27.6

Across the ten worst of each, the overlap is zero. Jacks in at 66% is a thirty-blind loss and a fine decision. Limping at 12.3bb and winning is still a mistake — and the only list that will ever show it to you is the left one.

A score that cannot see the result. It reads what the replay found, the equity you held when chips went in after the flop, and whether the preflop action matched the chart for that exact seat and stack. The moment a score consults the pot, it is just measuring luck again.

When your worst hands are all one mistake, it says so. A repetitive list is the finding, not a fault in the sorting — so it names the pattern instead of shuffling the rows to look varied.

A session report that knows its own limits. Twenty hands is a bust-out, not a sample, so below that it calls itself a log, drops the colour from the result and says plainly that it cannot assess how you played. Frequencies never appear per session at any length.

One thing to carry into the next game. Diagnosed across your whole database, because a single tournament could never establish it — then counted within the session you just played. Two of ten chances taken from the button.

And you can pin it until it actually moves. Progress is measured only on hands played after you pinned it, because re-measuring across every hand you have ever imported means the number never budges — which is exactly the feedback failure that makes people stop trying. It will not let you mark a leak fixed while the measurement disagrees.

And a form check that mostly says "noise". Your last few hundred hands against everything before them, judged by a proper significance test. Most stats move; almost none of them move enough to mean anything, and win rate is never called a trend at all.

Drill mode

Then it puts the spot back in front of you.

Knowing a leak is not the same as fixing one. Drill takes the decisions you actually got wrong — your seat, your stack, your table — and asks them again, cold. Answer, and you get the whole reference range back, not just a verdict.

postflop set · c-bet flop12 spots

→ Outside the band

You bet 10 of 12 (83.3%) against a 50–70% reference: you barrelled more than the reference does.

Your standing number is 84.3% over 216 hands — this set went the same way, so it is the habit showing up rather than a bad run of cards.

hand	flop	made	you	at the table
Q♦5♦	2♦T♦A♦	a flush, Ace high	Bet	Bet
A♣5♠	2♣A♥5♦	two pair, Aces and Fives	Bet	Bet
K♦9♠	A♥7♠A♣	a pair of Aces	Bet	Check
· · ·
7♠7♦	6♠4♦8♦	a pair of Sevens	Bet	Bet
K♣2♥	9♣7♦4♥	King high	Check	Check
9♠8♠	5♣6♦J♠	Jack high	Check	Bet

strongest first — not a claim that the weakest hands are the checks, only whether your betting tracks your holdings at all

Your hands, not sample hands. Every spot is one you sat in, graded by the same reference the leak report uses — so practice and diagnosis can never disagree with each other.

The shape, not the answer. Feedback shows the full 13×13 range with your hand ringed inside it, because the next spot will be a different hand.

It says what you really did. A limp is called a limp and a 2bb open is called a 2bb open — never rounded into "you folded" to make the quiz tidier.

Right twice, not right once. On two buttons, one correct answer is what a coin looks like. A spot clears after two in a row and comes back a fortnight later, so what you keep is recall rather than the screen you just read.

It tells you why, in a sentence. Not a restatement of the chart: the size of the range, what the hand class is doing in it, what your actual action cost, and how often the players behind you fold.

And then checks whether it stuck. Your deviation rate at the table before your first session against after it — the one measure a quiz score cannot fake.

Postflop, it grades the set rather than the hand. There is no solver here, and a quiz that called one particular flop check wrong would be inventing its verdict. What is established is the frequency band — so you answer twelve real flops and the verdict is your frequency across them, next to your standing number, which is how you tell a habit from a bad run of cards.

Your field

A solver says how poker should be played. This says how your table plays it.

Every reference band was written for a pool that folds. Yours might not. A hand history records every seat's actions and not only yours, so the same file that holds your hands holds thousands of theirs — which makes the pool you actually sit in measurable, for free, from data already on your disk. Where the band and the field disagree, the field is the one holding the money.

the same stat, three references

stat	you	solver band	your field
Fold vs steal	82.7%	35–55%	66.4%
Fold to 3-bet	44.1%	40–55%	19.4%
Went to showdown	48.3%	24–32%	55.6%

They fold to 3-bets one time in five. A band that assumes half will fold is not describing this table. Light 3-bets need fold equity and there is almost none here — so 3-bet for value, and take the pots by stealing instead, because the same players fold to steals two times in three.

Pooled from ~4,000 opponents across ~31,000 of their hands, over the same filters as the number beside it.

Measured the same way, or not shown. The field number comes out of the same code as yours with one flag flipped — same numerators, same denominators, same filters, same depth gates. A field stat defined even slightly differently would look reasonable and mean nothing.

Every opponent gets a profile too. Reads build as you import, and every percentage carries its sample size, so a 100% stat over two hands never reads as a settled tell.

It can argue with the leak report. When the field sits outside the band on the same side you do, the finding says so and steps back. A tool that only ever escalates is not measuring anything.

But it never excuses one. Open-limping is beaten by raising or folding whoever is sitting behind you. A field that limps as much as you do is an invitation to isolate them, not a reason to keep doing it — and those two cases are told apart rather than blurred.

Turned into something to do. A tendency large enough to change your play is stated as the adjustment, not the statistic: attack the blinds more, bluff the river less, raise over the limpers.

Tournaments and cash

Tournaments are not cash with a clock. Both get read on their own terms.

Most trackers analyse cash and go quiet on tournaments — not because tournament players don't leak, but because a tournament stat is a blend. Your "went to showdown" number mixes 100bb play with 12bb play, and averaging them describes no situation you were ever actually in. So the depth gets read before the hand gets judged — and cash runs through the same engine, without the gates it has no need for.

the same stat, four different games

≤ 10bb 70%

10 – 15bb 55%

↑ shove-or-fold — not a showdown you chose

25 – 40bb 48%

40bb + 46%

Only the lit rows are compared to the reference band. The faded ones are a different game.

Bands that know how deep you were. Showdown and c-bet numbers are measured only where those are real decisions. Below that the money goes in preflop and the stat stops describing a choice you made.

Push/fold against Nash, not vibes. Short-stacked, every open is compared to an ante-adjusted shove range for your exact seat and stack — the highest-EV decision in tournament poker, and the most memorisable.

Eight rules a cash-shaped model cannot have. Push/fold accuracy against Nash, limping a stack too short to play a flop with, min-raising a fifth of it away, re-shoving over an open, squeezing, isolating limpers, cold-calling on a stack that cannot play back — and failing to open wider as the table empties, which only exists because tournament tables break and cash tables do not.

Referenced against the pool where no chart exists. Nobody publishes a solver frequency for "re-shove at 20bb in a freeroll". So those rules are measured against the players at your own tables, over the identical slice — which is a reference that cannot be waved away as the wrong stake or the wrong format.

Every finding ends in an instruction. Not "your 3-bet is low" but the range to add, the sizing to use and the seat to use it from — next to the hands where it went wrong and what they cost.

Results, not just play. Finish, field size, ROI and how deep you ran sit beside the hands that got you there.

Go deeper

Then keep pulling the thread.

Summaries are where you start, not where the answer is. Every number opens into the hands underneath it.

◐

Behaviour inside a session

Four comparisons that contrast a slice of your play against its complement — after losing a big pot, after winning one, late in a sitting, and short-stacked. The baseline is your own play, not an average player's.

after losing a big pot −29.7

after winning a big pot +9.7

late in a session bb/100

Each comparison carries its own caveat — a tournament big loss also leaves you short, and some of the shift is forced rather than chosen.

▩

Hand inventory

All 169 starting hands, each with how often it was dealt, played, raised, walked and won, plus what it actually made you. Sort by any column to find the hand that quietly costs you money.

Click any hand to jump straight to every time you were dealt it.

⌗

Hand browser

Filter thousands of hands down to the spot you mean — position, stake depth, site, date, board texture, showdowns only — then search by the shape of the action itself.

R|B|B|*

B betR raiseC call X checkF fold| street* any

That line finds every hand you raised pre, barrelled flop and turn, however the river went.

▶

Replay any hand

Watch it back at the table — seats, blinds, every action in order, pot growing street by street, with your equity at each decision. Mark it as a favourite or flag it for review, and attach a note to your future self.

A♠K♦7♣9♥2♠

▶ pot 1,258

Export a hand as a shareable clip when you want a second opinion.

◆

Tournament results

Hand histories know how you played but not how you finished, so you type that in once: place, field size, buy-in, prize. From then on profit, ROI and how deep you ran sit beside the play that got you there.

12th of 180 top 7%

buy-in $1.08 prize $4.20

ROI +289%

A freeroll has no buy-in, so ROI is left undefined rather than reported as zero or infinity.

▥

Every stat, sliced

Thirty statistics across preflop, postflop and showdown — then cut by position and by stack depth, because "too loose" usually means too loose from one seat at one depth, not everywhere at once.

seathandsVPIPbb/100

BTN61231.2%+12.4

CO54824.8%+6.1

SB70127.7%−18.9

BB85922.9%−73.0

Under-15bb play is separated out, where raise-or-fold replaces everything you know about deep stacks.

How it works

Three steps, then it keeps itself current.

Drop in your hands

Point it at your hand-history folder or drag the files in. CoinPoker, GGNetwork and PokerStars-family formats are read natively, and who you are is detected from the hands themselves.

It does the work on import

Every hand is replayed, equities simulated and stats derived once, up front — so nothing recomputes while you're reading.

Read what changed

Leaks ranked by cost, hands worth reviewing picked out for you, and volume tracked by month — all filterable by position, stake depth, site and date.

Worth building?

Find out what your hand history has been trying to tell you.

Tell us you want it. No email, no account, nothing stored about you — just a count, so we know whether to build it.

Want to say more than that, or ask something? hello@pokerframe.com