RIICHI ORBIT Back to the game

ORBITAI

Why this tile?

Orbit AI chooses the creatures’ discards, riichi declarations, and calls. It also recommends tiles to you and explains why. Both use the same information: the hand, discards, open melds, dora, turn number, and score differences.

Follow a decision
A 3D diagram showing hand and discard tiles feeding into Orbit AI, then branching into four outcomes
These diagrams use the app’s tile faces on 3D tile bodies. They illustrate the calculations, not a live game screen.

From the visible tiles
to the next move.

Rivals choose their actions using steps 1–4. On your turn, the same results identify a recommendation and its alternatives; step 5 explains the differences.

  1. 01Read visible informationHand, discards, melds
  2. 02Estimate unseen tilesWhere 34 tile types may be
  3. 03List legal actionsDiscard, riichi, call
  4. 04Evaluate outcomesPoints and placement
  5. 05Explain the optionsFrom the same evaluations
Three candidate tiles compared on matching five-axis charts linked to Orbit AI’s evaluations
Five criteria per candidate. The same results power the explanation.

00 EXPLAIN

Different tiles.
Five ways to compare.

Every candidate is evaluated for hand progress, win chance, scoring potential, immediate safety, and safe tiles for later. Alongside the top recommendation, the app shows speed-, value-, and defense-focused alternatives in the same format.

Explanations are built from the evaluated shanten count (distance from a ready hand), useful draws, win prospects, expected score, and risk. The recommendation and its explanation refer to the same calculated results.

This comparison and explanation system was built for RIICHI ORBIT. Positions recorded during play can be evaluated again, so you can compare the leading options in post-match reviews and practice questions.

Criteria
Progress, win chance, value, safety now, safe tiles later
Reasons
Shanten, useful draws, win prospects, value, risk
Shown
Top pick, speed, value, defense
A diagram of the player’s hand, four discard pools, concealed opponents’ tiles, and a dora indicator
Read the hand, discards, melds, dora, turn, and score gaps.

01 OBSERVE

Start with what
you can see.

The inputs are the player’s own hand, all four discard pools and open melds, dora indicators, turn number, tiles remaining, score gaps, placement, dealer status, and hands left in the match. Opponents’ concealed hands and the order of the wall are not inputs.

Used
Hand, discards, melds, dora, score gaps
Excluded
Concealed opponents’ tiles and wall order
Updates
After discards, calls, and changes in the hand
All 34 tile types—Characters, Circles, Bamboo, and Honors—with probability markers
Update probabilities for the locations of all 34 tile types.

02 INFER

Unseen tiles.
Estimated, not revealed.

The AI tracks unseen tiles by type and updates the probability that each is in the wall or an opponent’s hand. These estimates inform useful draws remaining, likely waits, deal-in risk, and safe tiles that can be kept for a later turn.

Scope
Possible locations of each of the 34 tile types
Updates
After discards, calls, and new dora indicators
Output
Tiles left in the wall, waits, deal-in risk
A 14-tile hand branching to four candidate discards: 2 Characters, red 5 Circles, 8 Bamboo, and Red Dragon
Evaluate discards, riichi, calls, and passes with the same process.

03 COMPARE

Discard, call, or pass.
Compare the options.

Current scores, placement, dealer status, and remaining hands are used to estimate the chance of holding your position, how quickly you can win, the value you need, and the placement you could lose by dealing in.

Ordinary discards, riichi, discards after Chi or Pon, and passing a call all go through the same evaluation. Every candidate starts with 32 simulation paths. Close candidates are evaluated again with up to 96 paths, running through the end of the hand.

Compare
Discard, riichi, call, pass
First pass
32 simulation paths for every candidate
Refinement
Up to 96 paths for close candidates
A red 5 Circles tile branching to four outcomes: your win, dealing in, another player’s win, and an exhaustive draw
Translate four hand outcomes into points and final placement.

04 VALUE

Beyond this hand.
Think final placement.

The AI predicts your win, dealing in, another player’s win, and an exhaustive draw separately. It calculates point transfers for each, then evaluates changes in placement and the points needed to overtake a rival in the remaining hands.

Outcomes
Your win, deal-in, another win, exhaustive draw
Evaluation
Point changes, placement, remaining hands
Shown
The difference between the leading options
The same hand and discard layout rotated through four seats and linked to Orbit AI
Use the same wall, rotating through East, South, West, and North.

05 VERIFY

Same wall. All four seats.
Put changes to the test.

Walls used for tuning are kept separate from previously unused evaluation walls. Each candidate version plays the same wall from all four seats to reduce the influence of seating and starting hands.

The previous AI is kept as a fixed baseline opponent. Old and new versions play under the same walls and seat rotations. Before a change is adopted, final scores, placement, wins, and deal-ins are compared to reject changes that improve isolated positions but hurt overall match results.

Separation
Separate walls for tuning and evaluation
Compare
The same wall across all four seats
Baseline
Results against a fixed, previous-version AI

Three commitments.
Built into Orbit AI.

Meet Orbit AI
in RIICHI ORBIT.

View on the App Store