Getting Started

How Rosetta Works

From application to audit receipt — understanding the policy evaluation, risk scoring, and evidence generation pipeline.

The Rosetta Flow

Application
A financial workflow triggers an AI-assisted decision — a loan application, fraud alert, or underwriting review.
Rosetta — Policy Evaluation
The interaction is evaluated against defined policies. Risk rules, compliance requirements, and business constraints are checked.
AI Model
The AI model processes the request and produces a decision or recommendation within policy boundaries.
Evidence Record
Rosetta generates a tamper-evident audit receipt capturing the full decision context, policy evaluation, and risk assessment.
Human Review (Optional)
Depending on risk level, the decision may be flagged for human review before finalization.

Policy Evaluation

When a request reaches Rosetta, the first step is policy evaluation. Policies define the rules and constraints that govern AI-assisted decisions.

Policies can cover regulatory requirements, internal compliance rules, risk tolerance thresholds, and business-specific constraints. Each policy evaluation produces a result that feeds into the overall decision documentation.

Regulatory Policies

Fair lending, anti-discrimination, consumer protection rules

Risk Policies

Risk tolerance thresholds, exposure limits, signal-based rules

Business Policies

Product eligibility, pricing rules, operational constraints

Risk Scoring

After policy evaluation, Rosetta produces a risk score for the interaction. This score reflects the overall risk profile based on policy results, model outputs, and contextual signals.

The risk score determines whether the decision requires human review, additional verification, or can proceed automatically. All risk metadata is captured in the evidence record.

Risk Score Factors
Policy ComplianceDid the decision comply with all applicable policies?
Model ConfidenceWhat was the model's confidence in its output?
Signal DetectionWere any risk signals or anomalies detected?
Transaction ValueWhat is the value or impact of the decision?
Entity RiskWhat is the risk profile of the involved entities?

Audit Receipts

The final output of every Rosetta evaluation is a tamper-evident audit receipt. This receipt captures the complete decision context, including policy evaluation results, risk scores, model outputs, and review status.

Each receipt is designed to serve as authoritative evidence for compliance, audit, and regulatory review. Receipts are immutable records that provide a complete chain of custody for every AI-assisted decision.

Review Workflows

Not all AI decisions should be automated. Rosetta supports review workflows that route decisions to human reviewers based on risk level, policy requirements, or business rules.

Review workflows ensure that high-risk decisions receive appropriate human oversight while low-risk decisions can proceed efficiently. Every review action is also captured in the evidence record, creating a complete chain of accountability.

Auto-Approved

Low-risk decisions that pass all policy checks proceed automatically.

Flagged for Review

Medium-risk decisions are routed to a human reviewer for verification.

Requires Escalation

High-risk decisions require senior reviewer approval.