Database Review: How to Study Your Poker Sessions Effectively
Having a database is useless without knowing how to study it — learn filtering techniques, leak identification, and structured review methodology.
The Difference Between Tracking and Reviewing
Walk into any serious poker community and ask a show of hands: "Who has a tracker installed?" Nearly everyone. Then ask: "Who actually opened it this week to study, not to check their winnings?" Watch the hands drop.
This is the central paradox of database work. A database is the most powerful study tool in poker, and the most criminally underused. You have, sitting on your hard drive, the complete behavioral fingerprint of every opponent you've ever played, every decision you've ever made, every spot you've ever botched. And most players use it like a fancy bankroll widget.
If you're reading this, you almost certainly already have PT4 or HM3 humming in the background. You know how to import a hand history. You know what VPIP means. So we're skipping the basics. This article is about what to actually do with the database once it has 100k hands in it — the filtering techniques, the leak-finding workflow, the weekly routine, and the brutal sample-size math that separates real progress from confirmation bias.
Why Most Players Have a Database They Never Open
Before techniques, psychology. Three reasons databases gather dust:
- It feels like work, not poker. Studying spreadsheets doesn't trigger the dopamine loop that grinding does. Your brain reads it as homework.
- You don't know where to start. Open the filter window in HM3 and you're staring at 200 dropdowns. Most players click around for ten minutes, find nothing actionable, and close it.
- The truth hurts. A real database review will show you that you're not actually a winner at the stake you've been playing for six months. That's emotionally expensive. Avoidance is cheaper.
The fix is structure. When review is a defined process with defined inputs and defined outputs, the friction drops. You don't have to decide what to do — you just execute the routine.
Principle: Skill development is the compounding interest of poker. The grinder who reviews 3 hours per week beats the grinder who plays 10 extra hours per week within twelve months. The math is not close.
The Five-Stage Review Framework
Every productive database session moves through these stages, in order. Skip a stage and you're guessing.
| Stage | Question Answered | Time |
|---|---|---|
| 1. High-level overview | Am I winning? Where? When? | 10 min |
| 2. Filter to a problem area | What stat or spot is bleeding? | 15 min |
| 3. Individual hand drill-down | What am I actually doing wrong? | 30 min |
| 4. Fix (rule + drill) | What's the new rule, and how do I burn it in? | 30 min |
| 5. Measure progress | Did the fix actually work over a real sample? | Reviewed weekly |
Notice the time allocation. The trap is to spend 90% of your time on Stage 3 ("I'm gonna review some hands tonight!") without ever defining the problem you're solving. Random hand review is a hobby. Structured review is a profession.
Stage 1: The High-Level Overview
Open your database. Before you touch a filter, look at three views.
Winrate by stake. If you've been moving up, are you actually beating the new stake? Many players quietly run at -2 bb/100 over 30k hands at NL100 while telling themselves they're a NL100 reg. The graph tells the truth.
Winrate by position. Healthy 6-max numbers look roughly like:
- BTN: +8 to +14 bb/100
- CO: +4 to +8 bb/100
- HJ: +1 to +5 bb/100
- SB: -8 to -14 bb/100
- BB: -25 to -40 bb/100
- UTG (in 6-max): -1 to +2 bb/100
If your BB winrate is -55 bb/100, you have a defense problem. If your CO is breakeven, you're either too tight preflop or punting postflop. The position chart points the spotlight.
Winrate by month. Are you trending up, flat, or down over the last six months? A flat line for half a year means your study is not converting to results. Either the inputs are wrong or the volume is too low for fixes to register.
This whole stage takes ten minutes. Its only job is to give you a target for Stage 2.
Stage 2: The Eight Filters That Actually Matter
There are hundreds of filterable scenarios. Most are noise. These eight produce the highest signal-to-noise ratio in my experience coaching mid-stakes regs.
1. 3-bet pots, IP vs OOP
Pot Type = 3-bet AND Position = IP
Pot Type = 3-bet AND Position = OOP
What it reveals: Your 3-bet pot postflop strategy. Most players are far too aggressive OOP (over-c-betting wide ranges into capped opponents) and far too passive IP (failing to barrel turns). Look at flop c-bet %, turn barrel %, river bet %.
2. Vs RFI from BTN
Hero Position = BB OR SB
Villain Position = BTN AND Villain Action = Raise First In
What it reveals: Defense quality against the position that opens widest. If your BB defense vs BTN is below 38% (3bb open), you're folding profitable hands. If it's above 50%, you're calling junk.
3. River check-raise faced
Street = River AND Action Faced = Check-Raise
What it reveals: The single most exploitable spot in mid-stakes. Most regs underbluff here, so over-folding is fine. But check whether you're calling enough with bluff-catchers vs the specific opponents who do check-raise rivers as bluffs.
4. C-bet flop, checked turn
Hero Action Flop = C-bet AND Hero Action Turn = Checked Back (or Checked OOP)
What it reveals: A massive leak for most players. You're giving up after one bullet on boards where the turn changes nothing. This filter exposes whether you're capable of double-barreling.
5. Single-raised pot, multiway
Pot Type = SRP AND Players to Flop >= 3
What it reveals: Multiway strategy. Most heads-up GTO instincts are wrong here — c-bet frequencies should drop dramatically. If you're c-betting 65% in multiway SRPs, you're lighting money on fire.
6. All-in equity (luck filter)
Sort by All-In Adjusted Winnings vs Actual Winnings
What it reveals: How much of your recent results are variance. If your actual winnings are 30 buy-ins below all-in EV over 50k hands, you've been cold. The opposite is also useful — if you're 40 buy-ins above EV, you're not as good as your graph suggests.
7. Big bets won/lost
Filter by hands where Hero put in > 100bb
What it reveals: Your stack-off discipline. Sort these by net winnings. The five biggest losers will reveal the same patterns: hero-calling river overbets, bluff-catching too thin OOP, stacking off with overpairs on wet boards.
8. Showdown vs non-showdown winnings
This deserves its own section.
The Money Flow Diagnostic: SD vs Non-SD Winnings
This is the most diagnostic single chart in your tracker. Display two lines on your graph: Showdown Winnings (money won in hands that went to showdown) and Non-Showdown Winnings (money won in hands that ended before showdown).
| SD Line | Non-SD Line | Diagnosis |
|---|---|---|
| Up | Up | Healthy winning player. Continue. |
| Up | Down | Too passive postflop. Not bluffing enough, folding to too much aggression. |
| Down | Up | Too aggressive / spewy. Bluffing into calling stations, hero-calling thin. |
| Down | Down | You should not be playing this stake. |
The "Up SD, Down Non-SD" pattern is by far the most common shape for losing-but-thinks-he's-winning regs. The hands they win are the obvious ones (made hands at showdown). The hands they lose are death by a thousand cuts — folded turns to small bets, paid-off river overbets, never bluff-raised a river in their life. Money flows out non-showdown.
Principle: When the two lines diverge, the gap is your leak. Close the gap and your winrate jumps.
Stage 3: The 10-Hand Drill-Down
Once a filter has flagged a leak, do this exact drill:
- Set the filter (e.g. "BB vs BTN open, called preflop, faced flop c-bet").
- Sort by net winnings.
- Open the 10 biggest losers in that filter.
- Replay each hand. Write one sentence per hand: what did you do, what should you have done?
- After 10 hands, look at the sentences. The pattern will be obvious.
- Write a one-line rule: e.g. "Stop folding mid-pairs to small flop c-bets on dry boards in BB vs BTN."
- Drill that exact spot in your solver of choice or with GTO Wizard for 20 minutes.
- Note the rule in a "fixes" doc you re-read before each session for two weeks.
This is the entire workflow. Filter → drill 10 → pattern → rule → solver drill → repeat. It is unglamorous and it works.
Leaks vs Variance: The Sample Size Brutality
Before you "fix" anything, ask: do I actually have enough data to know this is a leak?
| Stat / Filter | Hands Needed for ~Reliable Read |
|---|---|
| Preflop VPIP/PFR | 5,000 |
| 3-bet % | 10,000 |
| Positional winrate | 25,000 per position |
| Overall winrate | 100,000+ |
| C-bet flop % | 3,000 c-bet opportunities |
| WTSD% | 8,000 |
| Specific filter (e.g. "river check-raised vs CO open IP") | Maybe never — most filters never reach significance |
The implication: for narrow filters, you cannot trust the bb/100 number — you can only trust the patterns in the hands themselves. If you filter to "river check-raise faced" and have 80 instances, the winrate column is statistical noise. But if you replay all 80 and notice you fold 70 of them while the villain pool is shoving with one combo of value and three combos of garbage, that is a real read.
This is why the 10-hand drill-down works even on tiny samples. You're extracting qualitative information, not chasing a number.
Measuring Progress: The 10,000-Hand Rule
Here's the math nobody wants to hear. Standard deviation in 6-max NLHE is roughly 100 bb/100. Over 10,000 hands, the 95% confidence interval on your winrate is approximately ±2 bb/100. Over 1,000 hands, it's ±6.3 bb/100. Over 100 hands, it's ±20 bb/100.
Translation: if you "fixed" your BB defense and ran a 2,000-hand sample to verify, you have learned absolutely nothing. The noise drowns the signal. To detect a 2 bb/100 winrate change with any confidence, you need at least 10,000 hands of post-fix data.
This is liberating, not depressing. It means session-to-session results are meaningless for measurement. Stop checking. The metric you care about is the moving 10k-hand winrate by month, and even that wobbles. Trust the process, not the graph.
The Weekly Review Routine
Open your calendar and block these:
Monday, 60 min — Session Review. Open the weekend's biggest losers and biggest pots (regardless of result). Replay them. Don't filter, don't analyze stats — just replay and note one thing per hand.
Wednesday, 90 min — Filter Analysis. Pick one of the eight filters from above. Run the 10-hand drill-down. Produce one rule. This is where the bulk of your improvement lives.
Friday, 90 min — Solver Work. Take Wednesday's rule. Build the spot in PioSolver, GTO Wizard, GTO+, or MonkerSolver. Run the sim. Compare your decision frequencies (from the actual hands you replayed) to the GTO solution. The gap is your leak in numerical form.
That's 4 hours per week. Players who do this consistently for six months move up at least one stake. Players who don't, don't.
Integrating Solver Work Into Review
A solver in isolation is masturbation. A solver tied to your actual hands is study.
The workflow:
- Filter your database to a recurring spot (e.g. "BTN vs BB, single-raised pot, flop = K72r, hero c-bet").
- Note your actual frequency. Say you c-bet 88% of the time.
- Run the spot in your solver. Solver says c-bet 67% (small sizing).
- The gap (88% vs 67%) means you're c-betting hands you should be checking. Which hands? Look at the solver's checking range. Compare to your range.
- You'll typically find you're auto-c-betting hands like KQ and middling pairs that the solver wants to check for protection and range balance.
- The fix: at the table, when you have those exact hands on that exact texture, check.
This loop — table → database → solver → table — is how live equity becomes actual skill. Most players run sims in a vacuum, learn nothing, forget it within a week.
How DEEPFOLD's AI Coach Compresses This Workflow
Here's the honest pitch. Everything above is doable manually. It's also slow, requires discipline, and depends on you knowing what to filter for. The bottleneck for most improving players isn't lack of tools — it's that they don't know which leaks to look for.
This is the gap DEEPFOLD's AI Coach is built to close. You upload your hand histories and the system parses every decision in every hand against solver-derived baselines. Decisions that deviate from GTO output by more than a configurable threshold get flagged with the texture, position, sizing, and the corrected line. You get a ranked list of your leaks — not "your VPIP is high," but "you're folding A5s in BB vs BTN 3bb open 78% of the time, GTO calls 100%, expected loss = X bb/100 across your sample."
That's a moat over manual review for one reason: scope. A human reviewer can drill down on one filter at a time. The AI scans all of them simultaneously, ranks by EV cost, and tells you what to fix first. You go from "I don't know where my leaks are" to "here are my top 5 leaks ranked by money lost" in minutes, not weeks.
Manual review still matters — you need to understand why the fix is the fix. But the AI handles the discovery, which is the part most players skip.
🎯 AI-powered leak detection → DEEPFOLD AI Coach
Five Worked Review Examples
Example 1: BB Defense Too Tight
Symptom: BB winrate -48 bb/100 over 30k hands.
Filter: Position = BB AND Villain Position = BTN AND Villain Action = RFI.
Drill-down finding: Folding 62% to 2.5bb opens. GTO defends ~70%.
Rule: Defend any suited gapper, any suited Ace, any pocket pair vs 2.5bb opens regardless of opponent type.
Drill: GTO Wizard BB vs BTN trainer, 100 hands.
Example 2: River Check-Raise Hero Call
Symptom: SD winnings strong, but five biggest losing hands all involve calling river check-raises.
Filter: Street = River AND Action Faced = Check-Raise AND Hero Action = Call.
Drill-down finding: Calling with one-pair hands 70% of the time on rivers where the only logical bluff is a missed flush draw. Pool is checking back the missed FD 80% of the time at this stake.
Rule: Vs unknown, fold one-pair to river check-raise unless on board where multiple draws missed.
Drill: Run river check-raise spots in PioSolver, study villain bluff combos available.
Example 3: C-Bet Too High in 3-Bet Pots OOP
Symptom: 3-bet pot OOP winrate -180 bb/100. (3-bet pots OOP run negative for nearly everyone, but this is bad even by that standard.)
Filter: Pot Type = 3-bet AND Position = OOP AND Street = Flop.
Drill-down finding: C-betting 78% on all flop textures. GTO at 33% sizing on most boards is closer to 60-65%, and on Ace-high boards drops to 40%.
Rule: On Ace-high flops in 3-bet pots OOP as PFR, check 60% of range. Specifically check medium pairs, weak Aces, and air.
Drill: Solver work on A92r, A75tt, AT2r 3-bet pots.
Example 4: Folding Turn Too Often as PFR
Symptom: Non-SD line trending steadily down over 50k hands.
Filter: Hero Action Flop = C-bet AND Hero Action Turn = Check-Fold OR Bet-Fold.
Drill-down finding: When opponent calls flop and bets turn, hero folds 71% with overcards, second pair, and weak draws.
Rule: Float more turns with backdoor equity and overcards in single-raised pots IP. Use the small block-bet line on rivers that brick.
Drill: Build the c-bet/check-call vs raise lines in solver, study turn defense ranges.
Example 5: Spewing in Multiway Pots
Symptom: Big losing hands clustered in 3+ way pots.
Filter: Players to Flop >= 3 AND Pot Type = SRP.
Drill-down finding: C-betting 55% in multiway as PFR. GTO solutions and ICMIZER-derived studies suggest 25-30% c-bet frequency multiway.
Rule: In multiway SRPs, check unless you have top pair good kicker or better, or a strong combo draw with backdoor equity.
Drill: Run multiway sims in MonkerSolver — note check frequencies are dramatically higher than HU.
Closing: The Quiet Edge
Database review is the single most underleveraged study activity in poker. Not because it's hard, but because it requires admitting what's broken. Most players prefer the comforting fog of "I'm running bad" to the precise knowledge that they fold A5s in the BB 78% of the time.
The structured framework — overview, filter, drill, fix, measure — turns review from a chore into a process. Add the eight filters as a recurring rotation, hold yourself to the 10-hand drill-down, respect the sample-size math, and run the weekly routine. Layer in solver work for the why. Layer in DEEPFOLD's AI Coach for the what to look at first.
The grinder who does this for a year doesn't get lucky. He gets quietly, mechanically, undeniably better — and the graph eventually catches up.
🎯 AI-powered leak detection → DEEPFOLD AI Coach
Word count: approximately 2,420 words.