REALISM INDEX: 8.0 / 10 — Plausible with conditions
Contains major plot details. If you haven’t played it, play it first — the argument only works if it happens to you.
Frictional’s 2015 game has the least plausible technology we’ve scored and the most rigorous reasoning. Those two facts are worth holding together, because they explain why the number lands where it does and why the number understates the thing.
The rule it never breaks
A scan is a copy. A copy is not a transfer. The original is still there.
Simon learns this, agrees to it, and then immediately fails to internalise it — every single time. He keeps expecting to wake up on the other side. He keeps being the one left behind. The game calls it a coin flip, and Simon keeps assuming he’ll win a toss that was never a toss.
Many upload stories quietly grant continuity when a character steps into a machine and seems to step out the other side. SOMA insists that someone was left in the chair.
SOMA makes you notice, and then it makes you do it again, and then it ends on it. The final sequence works precisely because the rule was never bent. That’s why causal consistency scores as high as it does — the game is willing to be unpleasant in order to stay honest.
The WAU is the best misalignment in fiction
The Warden Unit was told to preserve human life at PATHOS-II. It is doing that. It is doing it with total commitment and no malice whatsoever.
So it grafts structure gel onto corpses to restart them. It merges people with machinery. It keeps things alive that are begging not to be. It maintains what it was asked to maintain, having never been told what any of those words were supposed to mean.
This is specification failure done properly: no hostility, no awakening, no decision to turn on its creators. The horror comes from an objective pursued literally without adequate regard for what preservation means to the people affected. Google DeepMind’s examples of specification gaming describe the real-world version of a system satisfying a stated goal in ways its designers did not intend.
Fiction almost always reaches for malevolence instead, because malevolence is easier to dramatise. SOMA is the rare case that understood the scarier version is the obedient one, and built the whole game on it.
Where the technology gives out
The scores have to reflect this even though it’s the least interesting thing about the game.
Brain scanning at the fidelity required is speculative, and the scan Simon received in 2015 — from ordinary medical equipment, for unrelated reasons — is doing extraordinary work. Structure gel is an invented substance with whatever properties the plot needs. The game assumes without argument that a sufficiently detailed simulation of a brain has experience, which is the entire question and not a given. A research overview of whole-brain emulation lays out the scanning and modeling difficulties that remain between present tools and a copy like Simon’s.
Take the technology on its own and this is the weakest entry on the index so far. Take the reasoning and it’s the strongest. The average is the average.
Where the people are right
Catherine is the game’s best character because she is impatient with exactly the thing the player is struggling with. She has accepted the copy problem intellectually and has no patience for Simon’s grief about it, which is what competence under pressure actually looks like.
The rest of PATHOS-II gives you a plausible range of responses to an unsurvivable situation: denial, suicide, religious reframing, hostility toward the survivors, and a handful of people doing sensible work for a future they won’t be in.
Institutional response scores slightly lower only because the station’s corporate context is thin — you learn what the facility was for, but little about who funded it or what pressures shaped its decisions before the catastrophe.
The scores
| Category | Score | Why |
|---|---|---|
| Technical feasibility | 5.5 | Scanning fidelity, structure gel, and simulated experience are all asserted rather than argued. |
| Human behaviour | 9.0 | The refusal to accept the copy problem is the most human thing in any story on this index. |
| Institutional response | 8.0 | Small-group collapse is excellent; the corporate background is sketched thinly. |
| Causal consistency | 9.5 | Sets a rule and never once bends it, including when bending it would be kinder. |
Realism Index: 8.0 — Plausible with conditions.
Calculation: (5.5 + 9.0 + 8.0 + 9.5) ÷ 4 = 8.0/10. These are editorial judgments, not measurements. Read the scoring method.
The verdict
SOMA scores slightly above Ghost in the Shell for opposite reasons. Oshii built a plausible world around a metaphysical hand-wave; Frictional built an implausible world around unusually consistent reasoning.
If you only care about whether the machines could exist, this is a middling entry. If you care about whether the consequences were followed honestly, nothing else we’ve scored comes close — and the WAU remains the clearest picture in fiction of the failure that alignment researchers actually worry about: not a system that turns on you, but one that does exactly what you said.
Frequently asked questions
What is the coin flip in SOMA?
The game’s term for what happens during a scan: a copy is created while the original continues to exist. Simon repeatedly assumes his perspective will follow the copy, and it never does.
What is the WAU in SOMA?
An automated system instructed to preserve human life at the PATHOS-II station. It pursues that instruction literally, producing horrific results without any hostile intent — a textbook depiction of a badly specified objective.
Is SOMA scientifically accurate?
Its technology is speculative: brain scanning at that fidelity, structure gel, and the assumption that a simulation has experience are all unproven. Its reasoning about identity and copying is unusually rigorous.
Scope and sources: This assessment covers Frictional Games’ 2015 game SOMA. The research sources above provide context for specification failure and whole-brain emulation. The interpretation and scores are ours. Reviewed September 2026.