An effective solver-approved river call can cost you money.
Why? The solver assumes your opponent sees enough bluffs.
Most humans do not.
That is where MDA Poker begins.
MDA means mass data analysis. You Collect a large number of hands, group similar Players together, and analyze the behaviors of the group when faced with the same decision (the “spot”).
Not:
Players under-bluff rivers.
Something useful:
BTN vs BB. Pot is single raised. Flop small bet. Turn checks all the way. River gets bet big. How many times is that bet really a bluff?
MDA in poker is closely related to data mining, but the two are not exactly the same. Data mining is the collection of large amounts of poker data. MDA is what you do with that data: group players, study population tendencies and find spots where the pool consistently makes the same mistake.
MDA poker → Mass Data Analysis → poker data mining → population tendencies → exploits.
Now you have something to work with.
GTO tells you what should occur.
MDA tells you what people really do.
What is MDA in poker?
HUD’s tell you about individual opponents.
MDA tells you about groups of opponents.
Perhaps you have twenty hands on Villain. His river aggression statistic is meaningless.
But if he appears to be a typical player and you have a large sample of similar typical Players in the same pool, you have at least a starting point for studying.
That is the real value of MDA poker.
You can study:
- How often regs fold to BB vs BTN c-bets
- How often recreational Players bluff the river
- How often the pool probes turns after flop checks through
- How different bet sizes affect the range
- Which boards get over-folded
The filters matter as much as the data.
If you combine a 27/23 reg with a 45/12 recreational player, the results can be worthless.
Same for combining nl25 with NL500, dry boards with wet boards, or 25% bets with 100% bets.
Good MDA gets narrower.
The more precise the spot; the better the answer will be.
How MDA in Poker actually works.
The method is straightforward.
Collect → filter → compare → exploit
This collection stage is where data mining usually enters the MDA process. The important distinction is whether the hands are your own permitted histories or data collected from other players and tables.
Step one: Collect hand histories you are authorized to use.
Next: group similar Players.
Step Three: compare the pool with a solver baseline.
Step four: node lock the error and observe how the solver responds to it.
This step matters.
For instance, say the pool check raises much less on flops than theory dictates.
You may assume that you can simply:
Bet more.
Maybe.
However, the solver may also begin betting thinner for value.
The solver may employ more low-equity bluffs.
The solver may reduce the degree to which it protects its checking range since Villain is not attacking aggressively enough.
In certain instances, the greatest alteration occurs on the subsequent street rather than the present one.
That is why node locking is such a valuable tool.
You instruct the solver opponent to commit the same mistake as the pool and then you solve against it.
PioSOLVER and GTO wizard are both suitable for this type of work.
While software is useful, it is not the advantage.
Poor filters will generate poor statistics regardless of the quality of the software.
Mda vs GTO
You should not pick one.
You require both.
GTO provides you with a base-line.
Mda informs you about where the pool deviates from it.
Take a simple river spot.
Pot is $100.
Opponent wagers $100.
To call, you require approximately 33.3% equity.
If your hand is a pure bluff catcher, Villain requires roughly one third bluffs within his betting range.
Now, picture your database illustrates only 18% bluffs in that exact position.
These numbers are only examples.
Your call now becomes:
.18 x 200 – .82 x 100 = -46
this is not a minor deviation from what a solver predicts.
Solver will call due to equilibrium Villain expected to bluff enough.
Real pool doesn’t.
Therefore, you fold.
This represents the simplest approach to understanding mda.
You are replacing GTO with mda.
You are determining areas where real people do not play GTO well.
An example of simple MDA in Poker
BTN opens.
BB calls.
Board:
k ♠ 7 ♦ 2♣
BB checks.
BTN bets 33%.
BB calls.
Turn bricks and checks through.
River bricks again.
BB bets 33%.
Now consider your reports show:
| Solver | pool | |
|---|---|---|
| BTN folds river | 43% | 58% |
| BTN continues | 57% | 42% |
| Opportunities | — | 4,200 |
again, these are illustrative numbers.
BB is betting 33 into $100.
A zero equity bluff only needs to be called back around 24.8 percent of the time to break-even.
If BTN folds 58% of those bluffs, that bluff prints.
However, this is not where you end up.
Split regs from recreational Players.
Check different river cards.
Check different bet-sizes.
See whether leak remains.
You also wish to determine whether BTN made mistake on river or arrived on river with incorrect range due to mistake elsewhere.
Statistic itself is not an edge.
Understanding why it exists is.
Three leaks worth studying first
Put these in your shortlists of things to investigate,
1. River under-bluffing
Under-bluffing on river is one place i would look First.
Not because every pool under-bluffs each river.
They do not.
But bluff candidates disappear as the line gets aggressive.
Flopping bluffs are easy to find.
Triple-barreling is harder to do than flopping bluffs.
Raising river is harder than barrelling triple.
People generally discover value-hands easier than sufficient bluffs.
If your data indicates significant bluf-deficits in the area of river raises; you can fold bluff-catchers much higher.
Study river raises separately from river bets.
They are very different hand-ranges.
2. Dry board over-folding
Pools typically defend dry boards differently from wet boards.
On A72r, people see 98 and think nothing.
On 987ss, people suddenly see a gut-shot, pair, high-card draws, backdoor combinations or draws.
So a pool may over-fold small bets on dry boards and call too much on connected boards.
C-betting alone is not the exploit here.
It is:
bluff more where people over-fold too much. Value bet more where people hate calling too much.
Much better!
3. Recreational sizing
“sizing big means strong hand” is way too basic.
Recreational sizing is generally messy.
Twenty percent can be an anchor-block.
Fifty percent can be an automatic click.
Eighty percent can indicate protection.
One hundred fifty percent can indicate absolute nuts or complete nonsense.
You must create separate buckets by street action player type for each bucket type otherwise you’re merely creating fake science from table stories.
Statistics pitfalls
Most bad MDA falls into this category.
1. Show-down bias
You only see the cards that result in show down.
You do not see most folded hands.
Therefore, revealed hands are naturally stronger than folded hands.
If you rebuild early-street hand-ranges solely from show-downs; you can significantly misread the pool’s behavior.
Any serious range models need a mechanism to account for non-revealed hands.
2. Sample size
Never view a statistic without viewing Opportunities (i.e., hand-count).
Twenty-five percent can represent either:
- 1 of 4
- Or
- 250 / 1000
Both percentages are equal, yet completely different levels of confidence.
And 10,000 decisions from the same group of regular Players is completely different than 10,000 individual Players (with no overlap).
3. Pool differences
Nl25 gg is not NL200 Stars.
Week-end traffic is not week-day traffic.
six max is not heads up.
Data older than two months may not reflect current trends or data.
Good exploit in one pool could be terrible in another pool.
MDA Poker software
You do not need everything.
| Software tool | best application |
|---|---|
| Hand2note | constructing and filtering your own database |
| PioSOLVER | Custom solves and node-locking |
| GTO wizard | theory comparison and exploitation work |
| Freebetrange | Preflop population hand ranges |
| MDA Demon | Pre-analyzed populations data |
| Mass buster | automatically analyzing populations |
| HRC | Preflop and tournament solving |
If you play NL10 do not develop a six tool arsenal.
Begin by developing your own permitted database using one solver and learning how to develop one clean report for yourself.
Developing one clean report will help you more than developing thousands of statistics that you know nothing about how to interpret.
Is using MDA in Poker considered cheating?
The answer depends on the definition of MDA.
Studying your own hand history off-line is one thing.
Acquiring millions of hand histories you have never played is another.
Scrape table data is yet another.
Using software while playing to tell you what to do is RTA.
Those are not the same thing!
Mining and MDA in poker are also not the same entities.
Data mining is about gathering the data. MDA is about analyzing it. In practice, large-scale MDA has often relied on mined or shared hand histories, which is why the two terms are frequently used together.
But you can also perform MDA on a database made only from your own hands. That distinction matters because poker sites may allow you to analyze your own history while banning third-party data mining.
Pokerstars permits some statistics based on your own hand-history however prohibits mass shared opponent data collection and datamining.
GGPoker is more restrictive and clearly outlines prohibited systematic mass-data collection and certain third-party software tool usage.
Do not use rule:
Desktop site = tracker permitted
Check current rules of exact room you play.
Same applies to private poker apps and club networks.
Do not run random converters, scrapers or automated collectors because someone says everybody does it!
You can still Learn from ideas based on population without turning the game into a harvest project of data!
How to beat an MDA Poker Player
Generally you will never know for sure whether someone uses MDA or not.
Also does not matter much.
What matters is whether he applies population reads too stiffly.
Example: reg believes river raises are grossly under-bluffed (good exploit) but he starts bet-folding way too frequently against you; then you can bluff raise him more or perhaps he attacks turn checks way too frequently because the average range is way weaker; now protect your checks with stronger hands; now his population read has collided with a player who doesn’t match the population; that is the counter.
Not random play.
Not clicking weird buttons just to be different.
Just stop being average in those nodes where he expects you to be average.
Population data should only serve as starting point.
Specific read should then beat specific read.
Your First Thirty Days of MDA
Keep things basic.
- Week 1. Organize your own data base in accordance with your own MDA system.
- Week 2. Study one formation. The comparison of BTN versus BB SRP is all that is needed to get started.
- Week 3. Compare your largest three deviations using a solution tool.
- Week 4. Convert the three deviations into usable “rules” for playing at the table.
Example:
Board has no dry A’s. You are a recreational player. Flop c-bet small. Don’t bluff as wide.
This is good.
A spreadsheet containing 300 statistics isn’t.
The best MDA poker material will provide you with fewer decision-making options. It provides you with better decision-making options.
For some, MDA in poker is an advantage; others see it as one more step towards drying up the games and making them tougher to beat. If you are rooting against it, here are the online poker clubs with security team which monitor against data mining and MDA.
Another option where MDA will not work is Raize, an online poker site with live Dealers and without RNG.
