Inside Shadowfall's first local AI skirmish
Nitrogen Studios ·
Shadowfall's first local player-versus-computer skirmish gives the Phantom Dark strategy prototype a concrete test: can gathering, production and battlefield commands support a match that either side can win?
The prototype places a Human player against a Draconian computer opponent on Nexus Rift. It is available through Battle options → New AI skirmish in the development build. The earlier demo remains available for comparison.
An opponent with its own economy
Each faction has a separate resource bank and production queue. The computer pays the same unit and building prices as the player and follows the same worker, factory and population requirements. Its reinforcements come from gathering and paid production.
The opponent periodically checks whether it needs workers, a replacement production building, additional supply or more units. These checks create a basic loop between maintaining a base and building an army. They also make economic decisions visible enough to inspect during development.
The starting armies are asymmetric, and the opponent uses fixed construction candidates and timed attacks. Competitive balance and more varied strategy remain future work.
Time to prepare, then pressure
The first scheduled assault order occurs after 60 seconds of simulation time. Later assault orders follow at 30-second intervals. Travel time determines when an attacking group reaches the player's base, and defenders can engage intruders before the opening countdown ends.
Pausing the battle also pauses that countdown. Restart creates a fresh, paused match in the same mode, giving development checks a repeatable starting point.
The player can select Army or Workers, move a group, attack while advancing, focus an enemy, hold position or stop current orders. Winning requires destroying the opposing headquarters. Training and construction connect those battlefield decisions to the resources available at home.
What the recorded tests establish
The skirmish implementation's recorded core tests cover separate resource accounting, paid production, legal placement, supply expansion, assault timing, restart and victories for either side. A simulator interaction test also exercised entering the mode, training a worker, selecting and holding the army, playing, pausing and restarting.
Simple deterministic command experiments produced both a player win and player losses. That is useful evidence that the basic match can resolve in either direction. It is an early playability check; finished balance will require a broader range of players and strategies.
The boundary of this prototype
This is a local match. Backgrounding suspends the simulation, and relaunching starts a new one. Persistent bases, travel between worlds, multiple saved bases and hibernation protection remain design work.
The opponent does not yet have adaptive difficulty or scouting through fog of war. Units can overlap, worker hauling is not animated, and construction does not require a worker to walk to the site. The recorded simulator checks also leave physical-device performance and thermal behavior to later validation.
For now, the skirmish provides a foundation for testing the relationship between economy and combat. It makes the next questions concrete: how clearly the player understands the available orders, how different reinforcement choices affect a battle, and how the opponent should respond as the game grows.