NXS-Go

NXS-Go is a local-first abstract strategy game about connection, pressure, and isolation.

Two players build living networks from their Sources. They route pressure through edges, capture weak points, and try to trigger isolation. A branch that loses connection to its Source goes dark.

The board uses a 2.5D depth-field view: the rules remain graph-based, while shadows, perspective lines, and visual layering make network depth easier to read.

Current status: v0.2 AI Arena experiments.

Human playtest status: controlled local hotseat feedback wanted.

v0.1 Scope

The first playable prototype focuses on the core loop:

1. SYNCH: place a node near your living network.
2. ROUTE: direct flow along an edge.
3. PULSE: resolve route pressure and capture overloaded opponent nodes.
4. Isolation check: disconnected branches go dark.

Advanced asymmetry such as Signal Amplify and Noise Ghost Node is intentionally deferred until the base loop is stable.

Quick Start

Clone the repo:

git clone https://github.com/creator35lwb-web/NXS-Go
cd NXS-Go

Install dependencies:

python -m pip install -r requirements.txt

Start the prototype:

python nxs_go.py

Controls

• H: open/close in-game help
• E: open/close event history
• U: undo last completed action
• S: save session history to history/
• 1: select SYNCH
• 2: select ROUTE
• 3: select PULSE
• Space: execute PULSE
• [ / ]: lower/raise 2.5D tilt view
• R: reset
• Esc: quit

Use the mouse to place nodes or select edges. Hover over an edge to inspect the route bridge before clicking.

First Play

1. Start the game. The help panel opens automatically.
2. Press H to close the help after reading it.
3. Signal starts on the blue Source ring.
4. Press 1 for SYNCH, then click near the blue Source to place a connected node.
5. Noise takes the next turn. Repeat near the red Source.
6. Press 2 for ROUTE, then click an edge touching one of your nodes to create an arrow.
7. Press Space to PULSE when you want to resolve pressure.

If a branch loses connection to its Source, it goes dark.

Press S any time to save the session history as a Markdown file. The saved history is useful for reviewing how a move created pressure, capture, or isolation.

For a more detailed guide, see docs/PLAYBOOK.md.

Human Playtest

NXS-Go needs human competition feedback before broad AI-agent review.

Run a local hotseat best-of-3 set, then open a GitHub issue with the Human Playtest Feedback template.

See docs/HUMAN_PLAYTEST.md.

Best feedback includes a saved history file from history/, a screenshot of a confusing board state, and whether both players wanted a rematch.

Winning Pattern

The basic pattern is:

Build from Source -> route pressure into weak nodes -> capture bridges -> trigger blackout

The game does not use points yet. For v0.1-A, advantage is shown as network structure:

• live nodes
• routed edges
• whether the Source is still connected

You win by keeping your Source alive while making the opponent's Source/network go dark.

If neither Source is isolated after 60 completed turns, Horizon Scoring awards the game to the structurally stronger living network.

Test

Run the core logic tests:

python -m unittest discover -s tests

AI Arena

NXS-Go includes an early agent-facing environment in nxs_go_ai.py.

It exposes legal actions, observations, rewards, and baseline bots so the game can be tested by simple agents before future self-play work.

See docs/AI_ARENA.md.

Run a small baseline benchmark:

python scripts\benchmark_agents.py --games 5 --max-turns 30

Run a focused defense-vs-greedy geometry probe:

python scripts\benchmark_agents.py --suite defense --map contested_lanes --games 1 --max-turns 60

The current AI hypothesis is whether bridge-aware defense can contest greedy isolation pressure.

Turn-limit benchmark games include structural score leaders so experiments still produce evidence when no Source is isolated.

Current follow-up: CounterRouteAgent tests whether active defensive routing can contest greedy isolation better than passive bridge holding.

Latest hypothesis: TargetedCounterPressureAgent tests whether defense improves by routing into opponent pressure sources.

Current geometry finding: contested_lanes sharply reduces GreedyIsolation's horizon margin, suggesting defense is partly geometry/search dependent.

Current search finding: TacticalDefenseAgent improves some map margins but does not yet convert defensive positions into wins.

The AI Arena is designed for CPU-first development. Focused benchmarks do not require a GPU.

Project Docs

• ROADMAP.md: milestone direction from v0.1-A to v1.0
• CONTRIBUTING.md: contribution guide and design rule
• SECURITY.md: local-first security and privacy policy
• docs/ARCHITECTURE.md: current code structure and rule boundary
• docs/AI_ARENA.md: agent interface and baseline bot direction
• docs/AI_BENCHMARKS.md: early baseline-agent results
• docs/HUMAN_PLAYTEST.md: local human playtest guide and feedback questions
• docs/PLAYBOOK.md: first-game guidance and winning pattern
• LICENSE: MIT license

Current Design Rule

Connection is life.

Isolation is defeat.

Next Milestone

Playtest v0.1-A before adding asymmetric abilities.

Answer these questions first:

• Does SYNCH feel clear?
• Does ROUTE feel like intention?
• Does PULSE explain why nodes change owner?
• Does isolation feel immediate and meaningful?
• Can both players understand why a branch went dark?

Only after those answers are clear should v0.1-B add:

• Signal Amplify
• Noise Ghost Node
• stronger visual feedback