Open research question
Where does determinism help—and where does it create false confidence?
What can deterministic machinery genuinely establish, and which uncertainties remain outside its tests, specifications, and repeatable outputs?
- Status
- open
- Connected essays
- 3
- Projects
- 3
- Last activity
- July 21, 2026
Working hypothesis
Determinism is most valuable at boundaries: schemas, contracts, invariants, evidence capture, and repeatable checks. It becomes misleading when repeatability is treated as proof that the system represents the right world, measures the right outcome, or deserves authority in a particular domain.
What remains unresolved
- Which boundaries around an exploratory agent should be deterministic?
- When does a larger test suite increase assurance, and when does it only preserve a mistaken model?
- How should systems communicate uncertainty that their deterministic checks cannot observe?
Why it matters
AI systems make an old engineering problem visible: software can behave consistently while remaining wrong about the domain it acts within. Assurance therefore has to connect implementation evidence to a defensible decision about use.