Hundred Masks
Move, merge, summon. What’s your next turn?
Combine chess-like movement, merging and summoning in a paper-theatre campaign with readable tactical consequences.
- 208
- tests
- 12
- story battles
- 6
- characters, each with three memories
- 6×6
- board, one projection
As of 2026-09-15, from the repository and releases





Captured from the development build
How it works
- 01
Read the board
The player moves, merges or summons through a consistent board projection.
- 02
Commit a turn
The core checks legal actions, board state, resources and outcomes.
- 03
Meet the reply
A bounded opponent search chooses a legal response under its defined model.
- 04
Undo or continue
The theatre shows the consequence while turn history supports undo and continuation.
The hard parts, and what I did
Several ways to act. One readable turn.
Combine moving, merging and summoning while keeping each tactical consequence readable on a paper-theatre board.
A painted stage that behaves like a board
- The problem
- The stage is hand-painted, but taps, highlights and where a character’s feet land all need exact squares. Separate maths for each drifts apart.
- What I did
- The squares are generated in SceneKit and fitted to the painting. Hit-testing, highlights and footing all read from that one projection; the art is only the material.
- Result
- A test drags onto every visible square and checks the hit lands on the same cell, on every phone size.
Undo a whole turn, not a single move
- The problem
- One turn can include a summon, a move, a merge and the enemy’s reply. Undoing only the last step leaves coins, masks and hand out of sync.
- What I did
- The core records complete turn history, so undo restores the hand, masks and every piece’s position together.
- Result
- Covered by tests for cancellation, summon-then-move and reward de-duplication.
Development buildDevelopment build with local progress and turn undo; opponent search uses a limited action model.
Stack
Swift · MergeChessCoreSwiftUI · SceneKitDeterministic opponent searchNegamax · alpha-beta pruningTurn snapshots / undoLocal campaign persistence
Implementation details
The opponent searches a limited action model; daily challenges rotate authored setups.
- Player choice
- Select a supported board action.
- Paper board
- Present the board with a spatial paper-art treatment.
- Turn rules
- Apply actions and retain whole-turn state.
- Opponent search
- Choose among modelled actions within a bounded depth.
- Progress & undo
- Save progress and support whole-turn undo.
- Daily challenge
- Rotate through a fixed authored collection.