VERSION & EVIDENCE
See the work. Inspect the evidence.
Trustera demonstrates how an oversight team could connect a model or policy problem to affected decisions, investigate it, support human review and check the response.
Start with a short walkthrough
- Product overview explains the oversight workflow and its components.
- Banking example follows a decline-reason mismatch through the current workflow, Trustera’s checks and the proposed impact. Insurance example follows a change in a cyber-insurance vendor’s data definitions.
- Decision support shows why an explanation can match its trace while the supplied policy basis conflicts. Inspect the historical rule, exception authority, missing evidence and review/retest sequence.
- Model explanation and candidate comparison demonstrates a disclosed teaching model, feature contributions and a separate later-cohort evaluation.
What runs in this demonstration
The workflow examples calculate with constructed data in your browser. The Decision support section displays and replays actual exported Python assessments: 155 supplied synthetic records, with separately identified challenge cases. It does not call a live Python service.
All 120 original lending declines remain unchanged. Their explanation-matching example moves from 96 to 114 matching traces, leaving six unknown. The separate policy/evidence assessment identifies insufficient substantive evidence for those original records.
Policy limits and authority records are fictional demonstration specifications. Review actions and customer effects are simulated. Staff-hour inputs and projected benefits are unmeasured assumptions; actual customer remedies and institutional adoption are not established. Independent practitioner validation remains pending.
The workflow check page runs 31 authored checks across 19 synthetic scenarios. It covers the existing six-case JavaScript engine. It is separate from the newer Python decision-support implementation and is not an independent or production validation.
Release details
Hashes identify file bytes. They do not authenticate external facts or establish the date or independence of a review. The same Python evaluator is used for lending and insurance decision support; the older JavaScript workflow engine contains case-specific branches.