An assurance result a client cannot independently reproduce is an opinion. Determinism turns it into evidence.
For a supervised institution, an assurance result must be evidence, and evidence is something a third party can reproduce. A result that cannot be re-derived from stated inputs is an opinion, and an opinion carries little weight at examination. The distinction matters most on-chain, where transactions are irreversible and the cost of acting on an unsupported assurance lands on the institution that relied on it.
Determinism is achieved through discipline in what is pinned and recorded. Engine and ruleset versions are pinned by content hash, so the analysis runs against the same tools every time. Model-generated explanations are fixed to the analyzed bytecode hash, so an explanation is traceable to the exact artifact it describes. The state snapshot and inputs are recorded, so the same analysis re-run yields the same rule-based verdict. None of this is exotic. It is the difference between a record and a recollection.
Artificial intelligence has a defined place in this design. AI informs investigation and explanation, and generates hypotheses about where a contract could fail. Every hypothesis is confirmed deterministically before it becomes a finding. The verdict remains rule-based and repeatable regardless of what the model suggested, which means the output a client relies on does not vary with a model's probabilities.
A probabilistic model deciding a pass or fail verdict would be neither reproducible nor defensible. Model-risk management expects documented validation, stable behavior across runs and the ability to explain a decision after the fact. Keeping the verdict rule-based while AI assists the surrounding investigation is what makes an AI-assisted assurance system bankable: the model accelerates the work, and the determinism preserves the evidence.
Reproducibility is the property that lets an assurance result travel. It moves from the vendor into the client's own control evidence, and from there to the client's supervisor, intact and re-derivable at every step. Client Focus builds SCG360 to this standard: pinned inputs, recorded state and rule-based verdicts, so the result a client receives is a result the client can independently confirm.
Related insights
Perspectives / Covers: September 2026
Automated analysis and manual audit answer different questions. Treating one as the other is how a passing result becomes a false sense of safety.
Perspectives / Covers: August 2026
Detection is necessary. Accountable response is what protects an asset.
Regulatory briefings / Covers: July 2026
What European supervisors expect a digital asset operation to produce on request.
Client Focus reviews the estate, the coverage required and the gaps, then sets out what changes.
