Does the system behave as designed under the stated test conditions?
Review technical evidenceEvidence & Verification
Evidence should make the next decision easier.
AEGI separates technical control evidence, public synthetic benchmark evidence and institution-controlled evaluation evidence so each claim remains tied to what was actually demonstrated.
Evidence hierarchy
Three evidence classes. Three different questions.
Each evidence class answers a distinct question: technical behaviour, measured benchmark results, or customer-specific value.
Is there measurable supporting benchmark evidence under a defined comparison?
Review BAF-003Does the proposed change create decision-relevant value in the institution's own environment?
See the next evidence stagePublic synthetic benchmark · BAF-003
A measurable change in which cases reached fixed review capacity.
Same evaluation procedure. Same 2,417 Top-1% review slots. Pre-defined comparison.
CONTROL494fraud-labelled applications
→
CONTEXT-ENRICHED544fraud-labelled applications
+50
REVIEW CAPACITY2,417Top-1% review slots
NORMALISED EFFECT+20.7 / 1,000bootstrap 95% CI +9.1 to +32.3
INTERPRETATIONCase prioritisation changedat the same review capacity
Technical control evidence
Controlled execution across defined synthetic paths.
The completed synthetic test paths exercise risk-domain separation, institutional decision rights, failure handling, provenance, replay and version/configuration controls. Outcome performance is evaluated separately against labelled data and an agreed baseline.
Evidence pathExecutedWhat it demonstratesScope
Credit / Risk2,304 / 2,304
Synthetic Credit cases executed through the governed evaluation and evidence path.
Synthetic execution testingAuthorised Push Payment (APP) intervention32 / 32
Representative APP cases preserved the applicable evaluation configuration and evidence linkage.
Synthetic representative testingFraud → APP16 / 16
Workflow handoff cases preserved the governed APP evidence and control path.
Synthetic workflow handoff testingIntegrated Credit-Fraud Review Context14 cases
Combined, conflict, evidence-gap and authority scenarios were exercised while risk-domain provenance remained distinct.
Synthetic integration testingEvaluation control foundation15 / 15
Scoped controls cover risk-domain separation, institutional decision rights, failure containment, replay, provenance, version control and bank-local data boundaries.
Synthetic internal control testing
Risk-domain separationInstitutional decision rightsFailure & replay controlsEvidence provenanceVersion & configuration controlBank-local boundary
October 2026 · Internal engineering evidence
A tested synthetic Fraud workflow with reviewable handover and recovery.
AEGI has internally accepted the pinned synthetic Fraud workflow connecting Brief, acknowledgement, configuration, Shield evidence, Core Fraud I0 research rechecking, replay, review receipt, decision and Ops handover. The merged Ops release-branch source passed 276/276 tests on SQLite and 276/276 on PostgreSQL, plus 12/12 populated full-state recovery checks. An additional tested offline review pack remains in draft.
Eleven operating procedures have been documented and internally accepted for process readiness. Cross-system management acceptance drills remain a separate qualification gate. Together, the tested workflow and procedures provide a concrete engineering basis for institution-specific diligence.
AEGI Core
Verify the evidence, scope and configuration behind an evaluated result.
Core provides a scoped verification layer for evidence provenance, scope, version, configuration and integrity so institutional reviewers can inspect the evidence in context.
Understand AEGI Core
EVALUATED RESULTOutcome+Evidence+Scope+Version & Configuration→ Evidence Verification
Institution-controlled evaluation
The next material evidence comes from the institution's environment.
The institution defines the comparison basis, then the baseline and proposed change are evaluated on the same agreed historical evidence.
Scope
Agree the workflow, cohort, approved inputs, baseline, outcomes, review capacity and operating constraints.
Replay
Run the institution baseline and proposed change on the same historical evidence without post-access tuning.
Compare
Measure case-prioritisation impact, control effectiveness and evidence quality.
Decide
Continue, Refine or Stop. Any later shadow or production stage requires separate institutional approval.
EVIDENCE BOUNDARY
The published baseline covers completed internal technical tests and public synthetic benchmarking. Institution-controlled historical evaluation is the next stage for customer-specific outcome evidence.
NEXT EVIDENCE STAGE
Test one proposed change in an institution-controlled historical evaluation.
Start with the workflow, current baseline, proposed change and decision question.