UNDER THE HOOD / SYMBIOSIS: CRITICALITY
When 425 test runs all failed in the same place
Two developer reports show why a failed experiment and an unexpected balance result ask different questions.
In Symbiosis: Criticality, breaking things is only half the job. You strike a reactor core to scatter energy, then have to collect it before it decays. Buying more firepower can make the collection problem worse. How do you balance a game built around that tension? Developer Brehmdig built a test rig that tries different opening strategies and draws the results as a growing tree. One generation revealed a problem with the experiment itself.

In a July development log, Brehmdig describes loading saved games to probe the later stages. All 425 runs stalled in one place: they had started already maxed out. The setup could not reveal the progression it was meant to measure. The developer fixed the starting state, archived that generation and ran another. That is the developer's account; the post does not supply the raw trial logs for an independent reproduction.
An August update describes a different question. A player reported that the coupling system caused over 90% of damage. With equal investment, the developer measured shares from 3.5% at level 10 to 17.3% at level 250. Yet coupling still overtook drones and nodes at the final checkpoint, contrary to the intended balance. Its scaling was reduced. Brehmdig also considered the player's figure plausible for a build focused on that low-maintenance system. These are different setups, so the controlled result does not disprove the reported run. It asks what the game encourages people to keep buying.

| Level | Coupling damage |
|---|---|
| 10 | 3.5% |
| 30 | 3.7% |
| 60 | 6.8% |
| 250 | 17.3% |
The AI has a specific role here. In the July account, ordinary local software runs the sweep and calculates the scores; AI interprets the results and suggests a patch. The Steam disclosure separately credits AI coding assistants and generated visuals, audio and text, selected and integrated by the developer. Brehmdig says the design, balance targets, testing and final decisions remain theirs. A test can expose an unexpected winner. Choosing what a fair choice should look like is still a design decision.
For a smaller introduction to the reactor, the free browser demo is currently described as covering levels 1–20. The level 250 measurements belong to a historical development test outside that advertised range. The full Steam game is still listed as coming soon; Steam also offers a separate demo, whose level range is not established here.