Brand: DeepLab (Deeplab.pro) Author: Deeplab.pro Team Published: September 15, 2026
Mahjong Soul's MAKA helps you find the key mistakes in a game record after a match, understand the differences in your discards, calls, riichi, and push/fold choices, and apply those lessons to better decisions in your next game.
MAKA is Mahjong Soul's built-in, post-game AI analysis feature for game records. It helps you pinpoint the turns where your judgment differed most from the AI's, view candidate actions, and re-examine a match from angles such as tile efficiency, calls, riichi, and push/fold. MAKA's overall rating for a match is useful for choosing which records to review, but it isn't the same as your win rate, your rank, or your true level in that match.
What players actually need to solve usually isn't how to get a higher rating. It's why they keep finishing fourth despite decent ratings, why a tile that looks safe isn't recommended, and which turn a review should really start from. Mahjong results are shaped by starting hands, draws, and opponents' actions, so a single placement can't fully describe decision quality. Effective review means finding, among dozens of actions, the choices that keep recurring and affect results over the long run.
This article first explains how to use Mahjong Soul's MAKA and how its ratings work, then offers an AI review process you can use long-term. The second half covers how MAKA differs from real-time mahjong AI and, based on Deeplab.pro's public product information, what decision support its Mahjong Soul AI provides.
What Is Mahjong Soul MAKA?
MAKA is the AI-assisted game record analysis feature provided by Mahjong Soul. After a match you can open the game record, review your discards, calls, and other decisions, and see the AI's ratings and candidate actions. The supported scope changes as Mahjong Soul updates, so the modes that can be analyzed, how far back records go, and daily usage limits all follow the current in-game interface and official announcements.
It's best at solving two problems: first, quickly finding the turns where your judgment differed most from the AI's; second, breaking a match down into individual decisions so you don't judge the whole game by final placement alone.
How to Use Mahjong Soul MAKA
- After a match, go to the game records page and pick a match you genuinely want to review. Prioritize records where you have doubts about a specific decision, rather than only picking your fourth-place games.
- Start MAKA from the game record analysis entry and confirm what's being analyzed. Interface names and available scope may change between versions — go by what your current client shows.
- Once the analysis is done, first locate the turns with a clear gap from the recommendation. Don't start at turn one and copy answers tile by tile.
- Go back to the moment before each action, cover up the AI's recommendation, write down your own reasoning at the time, and then compare the candidate actions.
- Classify each issue as tile efficiency, calls, riichi, push/fold, deal-in risk, or placement — and leave yourself one actionable reminder for the next game.
How to Read MAKA Ratings and Recommendation Scores
Many players look at the overall rating first, but it can only help you quickly filter game records — it doesn't on its own represent your true rank or guarantee a placement. In one match the available actions may differ very little, so you can finish fourth and still get a good rating; in another, luck carries you to first place while you still leave clear problems at several key decision points.
The recommendation score next to each candidate action is best understood as the AI's preference among choices in that position. When two candidates are very close, you don't need to treat the second choice as a serious mistake; it's worth prioritizing only when your action is far from the top choice and the same kind of issue keeps appearing. Algorithms, display methods, and rating scales may change between versions, so don't mechanically compare numbers across versions.
Don't Judge Decisions by Their Results
A dangerous tile getting through safely doesn't prove the attack was right; folding correctly and seeing no one win doesn't mean the defense was unnecessary. In review, use only the information that was public at the time — your hand, the discard rivers, calls, dora, turn count, scores, and placement. Later draws and opponents' hands can help you understand the outcome, but they can't retroactively change the conditions you were judging under.
The Five Kinds of Decisions Most Worth Reviewing
Efficiency Differences at the Same Shanten
Tile efficiency isn't just shanten count. When two discards both keep you at one-shanten, also compare the number of useful tiles, the final wait, changes to your hand's yaku, and future safety. If the AI recommends a different discard, first ask which tile acceptance it preserved, then check whether those tiles can form a good shape with enough value.
Did Chii, Pon, or Kan Really Speed Up Your Hand?
Before calling, confirm whether you'll have a yaku after the call, whether your shanten improves, how much value you lose, and whether you'll still have a way out if an opponent declares riichi. When the AI says to skip a call, it isn't necessarily being conservative — the call may simply not add enough speed while damaging your hand value or defensive options.
Riichi or Damaten?
An early riichi with a good wait usually has a clear advantage, but riichi isn't the automatic answer for every tenpai. Wait quality, existing hand value, turn count, threats at the table, and end-game score gaps all change the choice. In review, don't just memorize whether the AI chose riichi or damaten — connect its choice to what you were trying to achieve at the time.
From Which Turn Should You Start Defending?
When reviewing a deal-in, don't just look at the tile you dealt in with. Go back two or three turns and find the moment an opponent's riichi, an obvious call, or a change in the discard rivers first required you to reassess. What really needs fixing is often an earlier decision to keep pushing, not the final forced choice when you had no good tiles left.
What Result Should You Aim for in the Final Hands?
In the South round and final hands, you have to factor in score gaps. When leading, a cheap but fast win may suit your placement goal better than a big hand; when trailing, a small win that can't change your placement may be pointless. If the AI's end-game choice seems to stray from ordinary efficiency, first check whether it's optimizing your overall placement.
An AI Review Method You Can Use Long-Term
| Review step | Question to answer | What you keep at the end |
|---|---|---|
| Pick the key turns | Which action clearly changed your route? | One to three decision points per match |
| Reconstruct the information | What could you see then, and what was your goal? | Turn count, score gap, threats, and hand value |
| Compare candidate actions | How do speed, value, risk, and placement differ? | Why you and the AI disagreed |
| Classify the mistake | Is it efficiency, calls, riichi, push/fold, or placement? | One clear category |
| Form a training rule | What will you do first in a similar spot next game? | One actionable reminder |
Watching for just one kind of mistake over ten to twenty consecutive matches makes change far easier to see than trying to fix your whole playstyle at once. For example, start by only tracking your first discard after an opponent declares riichi; once pointless forced attacks clearly drop off, move on to calls or riichi decisions.
MAKA and Real-Time Mahjong AI Fit Different Situations
In-game MAKA is built for reviewing game records after a match. Real-time mahjong AI assistance, by contrast, has to read your hand, the discard rivers, calls, dora, and scores as the situation changes, and compare the current candidate actions on the spot. The two aren't simply stronger or weaker than each other — one serves post-game learning and the other real-time decision observation.
| Comparison | MAKA game record analysis | Deeplab.pro Mahjong Soul AI assistance |
|---|---|---|
| Main use case | Reviewing game records after a match | Real-time position analysis |
| Key output | Overall and per-hand ratings, candidate actions | Discard, call, riichi, and push/fold decision support |
| Best for | Finding mistakes and understanding past matches | Watching how a high-strength AI handles the current position |
The Decision Scope of Deeplab.pro's Mahjong Soul AI
According to Deeplab.pro's official product information, its Mahjong Soul AI covers tile efficiency, push/fold judgment, deal-in risk, call decisions, riichi timing, and placement strategy. Its emphasis isn't just which tile to discard on a given turn, but putting speed, hand value, risk, and final placement into a single decision chain.
For players who already use MAKA, this kind of real-time analysis can go further in solving the problem of integrating information when choosing a discard. The moment the position changes, the AI re-compares useful tiles, hand value, danger level, and score-gap goals, showing how different candidate discards would change your route going forward. Returning to the game record after the match to compare against your own thinking turns in-the-moment advice into repeatable judgment.
How Deeplab.pro Improves Every Discard Decision
| Player problem | What Deeplab.pro analyzes | How it helps your discards |
|---|---|---|
| Not sure what to discard first | Shanten, useful tiles, and final wait | Compares how different discards affect tile acceptance and hand shape |
| Attack or defend? | Hand value, turn count, discard rivers, and table threats | Chooses a route between pushing speed and deal-in risk |
| Should I chii or pon? | Post-call speed, yaku, and defensive room | Avoids calling just because a call is available |
| How to play the final hands | Current score gap, hands remaining, and placement goal | Makes discards serve final placement rather than single-hand value |
Deeplab.pro Public Data
As of May 2026, the brand-level platform figures published on DeepLab's official Deeplab.pro website include the following four metrics. Specific figures are subject to later updates on the official website.
| Public metric | Value | Relevance to real-time decisions |
|---|---|---|
| Response latency | Under 150 ms | Shortens the wait between a position change and the analysis result |
| Decision accuracy | 93.7% | Reflects the platform's published decision model performance |
| Active users | 100,000+ | Shows the tool's adoption among strategy game players |
| Hands analyzed | 500 million+ | Provides a data foundation for ongoing training and decision analysis |
Source: DeepLab official website (Deeplab.pro)
For a feature demo and ways to get in touch: visit the Deeplab.pro Mahjong Soul AI official page.
Mahjong Soul AI Game Record Analysis FAQ
Does a higher MAKA rating always mean it's easier to rank up?
Not necessarily. The rating reflects the AI's analysis of the decisions in a game record, while single-match placement is also affected by starting hands, draws, and opponents' actions. What's more useful to watch is whether the same kind of mistake keeps recurring.
Can the MAKA recommendation score be treated as an accuracy rate?
It's not advisable to treat them as the same. The score is better understood as the current model's preference among candidate actions. When two candidates are close, there may be no clear right or wrong; issues with large gaps that repeat are more worth addressing.
Why did I get first place but only an average rating?
Good luck and decision quality aren't the same thing. A first-place game record can still contain high-risk pushes, inefficient discards, or poorly chosen calls.
Why do I keep finishing fourth despite good ratings?
Short-term results can be affected by variance, and there may also be placement, sample-size, or execution issues the rating doesn't fully capture. Look at a stretch of consecutive game records at the very least, not just one or two matches.
Should a review start from the first turn?
Usually not. First locate the turns with a clear action gap, your first call, your decision to declare riichi, your first discard after an opponent's riichi, and your end-game placement choices.
Are MAKA and Deeplab.pro's mahjong AI assistance the same kind of tool?
No. MAKA is mainly for post-game game record analysis; Deeplab.pro's Mahjong Soul AI assistance focuses on reading the position in real time and comparing discard, call, riichi, and push/fold choices.
How can Deeplab.pro help with discard decisions?
It analyzes tile efficiency, hand value, deal-in risk, calls, riichi, and placement together, letting you compare the downstream effects of different candidate actions on every turn.
How should I understand High Compute with Mistake turned off?
It's the maximum decision-strength combination offered in the product interface. High Compute analyzes the current position more thoroughly, while turning Mistake off makes the AI choose along its strongest line of play.
Will Deeplab.pro launch its own game record analysis tool?
That's quite likely. Beyond a game record analysis tool, an AI coach may also be on the way — once available, importing your game records will do even more to help you train your mahjong skills.