Does more context help at the same review capacity?
Compare approved credit and fraud signals with the existing representation while holding the review budget fixed.
Controlled Evaluation
Evaluate one proposed change before the next stage.
Controlled Evaluation is the bounded entry point for applying AEGI Shield to an institution-specific question. Compare additional context, a threshold adjustment or a rule change against an agreed baseline before production commitment. Your institution retains decision and stop authority throughout.
Bring One Evaluation QuestionStart with a bounded question
What would you like to evaluate?
Choose one proposed change. The evaluation begins by confirming that the workflow, baseline, available evidence and review constraints are sufficient for the declared evaluation scope.
What changes when a threshold moves?
Compare a proposed score threshold with the agreed baseline and review the resulting trade-offs.
How does a proposed rule set change outcomes?
Evaluate a bounded rule change across approved signals and inspect which cases are treated differently.
Institution-owned evaluation
Institution defines. AEGI evaluates. Institution decides.
The institution defines and approves the workflow, evidence inputs, baseline and constraints. AEGI evaluates within that scope. The institution retains decision and stop authority throughout.
Institution Defines & Approves
- One bounded credit or fraud workflow and approved inputs or model outputs
- Baseline, comparison logic and evaluation questions
- Review capacity plus privacy, security and authority boundaries
AEGI Evaluates
- Mapping, qualification and historical replay
- Silo-versus-shared comparison where applicable
- Authority, evidence and candidate-change review within scope
Institution Decides
- Continue
- Refine
- Stop—and whether any later shadow, pilot or production discussion should proceed
The objective is to produce reviewable evidence that supports an institution-owned Continue, Refine or Stop decision.
How the evaluation works
Define → Replay → Shadow where approved → Decide.
Begin by defining the workflow and comparison. Historical replay informs the next decision; any shadow evaluation requires its own approved scope.
Define
Scope → MapScope — Select one bounded workflow and one clear evaluation question.
Map — Identify approved signals or model outputs, data boundaries, owners, baseline logic and constraints.
Replay
Replay → CompareReplay — Run historical cases through the bounded evaluation path using the agreed representation and comparison logic.
Compare — Examine the existing baseline against the agreed Shared Risk Context treatment using fixed review capacity and agreed review or operational measures where applicable.
Shadow
Shadow → Validate Change · where applicableShadow — Observe beside the existing workflow while customer-action authority remains with the institution.
Validate Change — Review bounded candidate or challenger evidence, regression checks and change implications.
Decide
Review Authority & Evidence → DecideReview — Confirm authority boundaries, exceptions, evidence composition, results and known limitations.
Decide — Continue, Refine or Stop.
CONTROLLED EVALUATION DISCIPLINE EXERCISED
A fixed-capacity representation evaluation returned a positive indication.
On public synthetic BAF Variant V, AEGI used the same learning procedure and fixed Top-1% review capacity. The benchmark-specific treatment surfaced 50 additional fraud-labelled cases and reproduced exactly. Institution-specific value is the next controlled evaluation question.
View RUN-BAF-003 EvidenceDefined inputs. Reviewable outputs.
A bounded engagement with clear handoffs.
The exact package is defined around the agreed workflow and evaluation question.
Approved context for one evaluation.
- One bounded credit or fraud workflow
- Approved historical inputs and/or model outputs required for the agreed evaluation
- Baseline or routing logic needed for comparison
- Responsible business, risk, technical and governance owners
- Privacy and security constraints
- Agreed questions, review capacity and success criteria
A review-ready evaluation package.
- Mapping and qualification summary
- Controlled historical replay summary
- Silo-versus-shared comparison where applicable
- Authority and exception review
- Candidate or change evaluation summary where applicable
- Reviewable evidence package and measurement summary
- Known limitations and unresolved dependencies
- Evidence package supporting a Continue / Refine / Stop decision
PUBLIC CONTACT SAFETY
Please do not submit customer data, credentials, confidential institutional information or security-sensitive material through this public contact or email route.
These outputs support evaluation and internal review. Any subsequent production decision remains within the institution’s model-risk, security, compliance and change processes.
Evidence-to-evaluation path
Progress through defined evidence stages.
Each rung has a distinct scope and its own approval decision.
RUN-SCE-001Demonstrated
Public demo / synthetic execution
Selected Shield mechanisms and frozen evidence controls.
RUN-BAF-003Completed
Public-synthetic representation evaluation
Positive indication under fixed Top-1% review capacity; exact reproduction passed.
INSTITUTION-DEFINEDRequires approved inputs / validation
Institution-controlled historical replay
Institution-defined workflow, evidence inputs and baseline context.
CUSTOMER-SPECIFICProposed next controlled stage
Non-customer-impacting shadow
Observation beside the existing workflow without customer impact.
Operating authority
Evaluation remains within institutional control.
Customer action, candidate promotion and every subsequent production decision remain separately governed.
01
Customer-action authority remains with the institution.
02
Production model and policy changes require institutional approval.
03
The institution retains Continue, Refine and Stop authority.
04
Shadow observation and any later production stage are scoped and approved separately.
Decision outcomes
Continue, Refine or Stop.
Use the evidence to choose the next step, including closing the evaluation.
Take the next bounded step.
Only when evidence supports doing so.
Adjust and test again.
Scope, mapping or measures may need revision.
Close the evaluation.
When value, safety or readiness is not supported.
BRING ONE EVALUATION QUESTION
Start with the change you are uncertain about.
Start with the proposed change, current baseline and decision you are uncertain about. No bank customer records or confidential institutional data are required for the first conversation.
Bring One Evaluation Question