Model governance | Praman Labs

Score and band
Decision state
The response records the personhood score, policy band and recommended action that were active when the application was evaluated.
Ranked reason codes
Readable explanation
The strongest supporting and contradicting signals are returned in readable language with their contribution to the result.
Model and policy version
Reproduction key
The inference version, threshold set and institution policy reference are attached so the decision can be reproduced later.
Evidence and overrides
Audit context
The record includes available raw signal values, any hard override and the ring identifier where network evidence affected the result.

Thin-file status
Results are compared for borrowers with and without established bureau history so an absent file is not silently treated as proof of a false identity.
Geographic performance
Evaluation is segmented by relevant geography and onboarding channel where the institution has enough labelled outcomes to make the comparison meaningful.
Outcome and review threshold
Performance is reported against a defined fraud or loss label and at the share of applications the institution can review, not as an isolated accuracy figure.
Change comparison
A proposed model is compared with the active version on a fixed evaluation set before thresholds or authority change.

Human review path
Thin and inconsistent cases can be routed to step-up verification or an analyst queue. The institution defines when a human decision is required.
No unexplained decline
A production policy must not issue a decline without reason codes and a record of the threshold or override that caused it.
Member challenge
An institution can challenge a score with the application identifier, decision record and corrected outcome. The response becomes part of the governance log.
Authority can be withdrawn
If drift, data loss or segment harm breaches an agreed limit, the institution can return the model to shadow mode while the issue is investigated.

Release record
Each operational release states its training window, participating signal groups, evaluation cohorts, known limitations and approved policy thresholds.
Approval route
A member’s approval route determines whether and when a new version can affect live action.
Historical integrity
Outcome feedback can improve later models, but a corrected label does not rewrite the historical decision. The original record remains available with the later challenge and resolution attached.

Validation is local
Before granting decision authority, an institution validates the model on a defined cohort, outcome label and review threshold. This site does not present one customer’s result as a universal benchmark.
Accountability is named
Operational use requires the governance record to identify the responsible Praman Labs model owner, the institution’s policy owner and the approval route for the use case.
Thresholds fit the use case
Drift, performance and disparity limits are agreed against the institution’s data, review capacity and action policy rather than copied from a public default.
Challenge terms are explicit
The institution’s governance pack records the review route, responsible contacts, required evidence and escalation expectations. Any response-time commitment belongs in the applicable agreement.

Governance contact

Use this public route for general governance questions, not individual application reviews.

A score must leave an inspectable record.