Hegemon

Hegemon is a deterministic abstract strategy board game I designed and built from scratch — a chess-alternative with a novel core mechanic: directional commitment. Every piece has a facing (N/E/S/W), and turning costs your entire turn. This single constraint reshapes the entire strategic landscape: positional decisions become load-bearing, and over-extension is punished differently than in any game I know of.

The project spans game design theory, algorithmic AI, and a clean web implementation purpose-built for playtesting.

---

Why I Built It

Classical strategy games like chess have three structural weaknesses I wanted to address:

1. Over-stabilized openings — theory crowds out creativity
2. Omnidirectional threat projection — pieces threaten too freely in all directions
3. Draw-prone equilibrium — high-level play collapses into stalemate

Directional commitment attacks all three: pieces must be oriented to threaten, rotation is expensive, and the asymmetric cost of facing changes creates constant pressure to act.

---

Game Design

Hegemon is played on a 9×9 board with 13 pieces per side across 4 types:

Each turn, a player either moves a piece or refacing it 90°. Reface moves cannot be used to escape check — this prevents the mechanic from becoming purely defensive.

Win conditions:

• Royal Immobilization — opponent's Archon is attacked with no legal escape
• Command Paralysis — opponent has zero legal actions of any kind

The central design question I'm testing empirically: Does facing create legible strategic commitment, or mostly bookkeeping cost and accidental paralysis?

---

Technical Implementation

Stack: React 18 + Vite 6, no external game libraries, pure CSS dark theme.

Architecture: The engine is fully decoupled from the UI. Three independent modules:

• engine.js — Pure game logic: move generation, validation, check detection, win conditions. No React imports, no side effects. Could power a server, a bot, or a CLI.
• ai.js — Standalone AI engine with negamax search.
• notation.js — Algebraic notation system for game records.

AI Opponent

The AI uses negamax with alpha-beta pruning at three selectable depths (Recruit / Officer / General). The evaluation function is handcrafted across 8 factors:

• Material balance (Archon: 10,000; Strategos: 600; Hippeis: 350; Hoplite: 100)
• Hoplite advancement with exponential promotion-proximity bonus
• Strategos forward reach and center control
• Hippeis activity and advancement
• Archon safety (penalties for exposure and proximity to enemy pieces)
• Check state (+45 pts for giving check, −45 for being in check)
• Piece facing quality (bonus for correctly-oriented Hoplites)

Move ordering follows MVV-LVA (Most Valuable Victim, Least Valuable Attacker) to maximize alpha-beta cutoffs, with captures sorted before quiet moves and refacings deprioritized. The Hippeis free-reface mechanic is handled correctly: after a capture, the AI evaluates all 4 possible post-capture orientations and picks the best one.

Typical search visits ~500–5,000 nodes at depth 3, with node count surfaced in the UI.

UX Details

• Piece tokens show type + directional arrow (N/E/S/W)
• Legal moves rendered as blue dots; capture squares as red overlays
• Last-move highlighting; auto-scrolling game log in algebraic notation
• R to toggle reface mode, Esc to cancel selection
• Fully responsive: board scales from 65vw to 450px max

---

Playtest Protocol

v0.2 is a structured playtest with a packet distributed to testers:

1. 3-minute quickstart
2. Piece movement reference card
3. Setup + notation guide
4. Results form

Metrics tracked per session:

Data from playtests will drive v0.3 (phalanx formations + ranged Toxotai unit) and v0.4 (Epistates command-aura piece + throne win condition).

---

What I Learned

Designing around a single novel mechanic is harder than layering features — the facing rule touches everything: how pieces move, how attacks work, what a "tempo" means, how the AI should weight positions. Getting it right required treating the engine as a formal system first, and the game second.

The minimax evaluator taught me how sensitive hand-tuned heuristics are: a 20-point tweak to Archon exposure dramatically changed whether the AI played aggressively or passively. Calibrating it felt more like training a model than writing an algorithm.

---

Status: Active playtest (v0.2). Game design, engine, AI, and UI are all original work.
Tech: React · Vite · Vanilla CSS · Custom negamax AI