Designing a dispute-resolution state machine for adversarial user outcomes

I’m exploring a backend architecture problem that comes up in a lot of adversarial systems:

Two parties agree to a set of rules, an outcome occurs outside the system, and the platform must deterministically decide “what happened” in a way that’s defensible, auditable, and repeatable.

The scope I’m looking at is intentionally narrow and infrastructure-only:

• no matchmaking
• no pricing or odds
• no payment custody
• no control over the underlying activity

The system only:

• locks terms once both parties accept
• accepts outcome submissions
• supports dual confirmation or evidence-based disputes
• resolves outcomes via a deterministic state machine
• emits an auditable decision trail explaining why a resolution was reached

What’s interesting is how often platforms that already handle users, identity, and money explicitly avoid owning this layer.

For folks who’ve worked on marketplaces, trust & safety, moderation tooling, or dispute systems:

where do designs like this usually fail in practice? Is it operational cost, abuse patterns, regulatory ambiguity, internal policy drift, or something else?

I’m looking for failure modes and reasons not to build this, rather than validation.