Govern · item 19

Risk classification and contestability

A regulator does not ask whether the agents are useful. It asks what class of risk each one sits in, who the human overseeing it is, what the affected person is told, and how they contest a decision they believe is wrong. Every agent in this estate carries an answer to all four, and where the answer is weak it is written as a gap rather than left blank.

Adverse-impact monitoring

Health

How the agent estate is classified

Classified systems by risk class and conformity state, with the gaps recorded against each.

Agents classified

11 affect a person directly

High risk

2 limited · 5 minimal

Independently assessed

13 self-assessed · 3 not assessed

Named appeal route

10 of 16

Every high-risk agent routes to a named human

Appeals recorded

5 overturned · 1 partially upheld · 3 pending

Wired to the spine

13 of 13

Median human review: 48 minutes

Risk class under the prevailing regulation

Classification follows what the decision does to a person, not how sophisticated the model is.

Prohibited0

Practice not permitted under the prevailing regulation. Must not run.

High risk9

Affects access to employment, credit or a comparable material interest. Requires human oversight, logging and contestability.

Limited risk2

Transparency obligation applies. The affected person must be told they are dealing with an automated system.

Minimal risk5

No specific obligation beyond ordinary governance.

No agent in this estate is classified as a prohibited practice. That is a claim the register makes explicitly, and it is checked against the classification basis recorded on each agent rather than assumed from silence.

High risk — 9 agents

Affects access to employment, credit or a comparable material interest. Requires human oversight, logging and contestability.

Agent	Function	Why it is classified here	Human oversight	Explanation to the person	Appeal path	Logging	Conformity	Status
Screening Agent A3 · people-affecting	Human Resources depth-agent	Employment: a system used to filter applications and evaluate candidates in the course of recruitment. High risk on the face of Annex III.	Recruiter review of every shortlist, with the ranking rationale and the field list visible, and the ability to reinstate any applicant. Recruiting pod lead, per requisition · Recruiter	Published notice shown	Named human Talent Acquisition Lead for the region	84 mo	Self-assessed	Operating with a gap Self-assessed. No notified body involvement, and the NYC Local Law 144 independent bias audit has not been commissioned even though the agent screens candidates for New York roles.
Succession Slate Agent A2 · people-affecting	Human Resources depth-agent	Employment: a system used to make decisions affecting promotion. Slate construction determines who is considered.	Talent council review of every slate, with a recorded reason required when a slate is accepted unchanged, and automatic rejection of slates below four names. Morgan Idris · Talent council chair	On request no notice	Named human Morgan Idris, Function Leader	84 mo	Self-assessed	Operating with an alarm Employees are not proactively told a slate exists or that an agent helped build it. Explanation is available on request, which means available to whoever knows to ask. Two fairness alarms are open against this agent.
Calibration Prep Agent A1 · people-affecting	Human Resources depth-agent	Employment: a system used to evaluate performance and behavior. Framing the calibration is evaluation in substance.	The agent issues no rating. Calibration is a human session; every departure from the proposed distribution is minuted. Function leader chairing the session · Calibration chair	Published notice shown	Named human Priya Raman, Chief People Officer	84 mo	Self-assessed	Operating with a gap The German population is excluded pending the works council agreement, so the readings shown cover four jurisdictions out of five.
Merit Cycle Agent A2 · people-affecting	Human Resources depth-agent	Employment: a system used in decisions affecting terms of the work relationship, specifically pay.	Manager approval on every recommendation, followed by an automated pay-equity check that holds any increase widening a flagged gap. Marcus Weill, VP Total Rewards · Reward partner	Published notice shown	Named human Marcus Weill, VP Total Rewards	84 mo	Self-assessed	Operating
Attrition Signal Agent A2 · people-affecting	Human Resources depth-agent	Employment profiling. Predicting who will leave is profiling of workers, and the flag is capable of affecting how a person is treated.	Flags are visible only to the HR business partner, expire after 60 days, and are excluded from promotion, merit and slate contexts by deny list. Employees may opt out. HRBP for the population · HR business partner	On request no notice	Named human Priya Raman, Chief People Officer	84 mo	Self-assessed	Operating with a gap Employees are profiled for flight risk without being told. The opt-out exists because one employee appealed and asked for it, not because it was designed in. Review is 51 days overdue.
PIP Drafting Agent A1 · people-affecting	Human Resources depth-agent	Employment: a system used in decisions on termination of the work relationship, since an improvement plan is a step on that path.	Two humans record the decision before the agent drafts. Every draft is edited before issue. HRBP countersigning the plan · HR business partner	Published notice shown	Named human Priya Raman, Chief People Officer	84 mo	Self-assessed	Operating with a gap Volume is too low for a statistically meaningful fairness reading at 180 plans a month across four jurisdictions. The platform reports insufficient sample rather than a ratio.
Background & Right-to-Work Agent A3 · people-affecting	Human Resources depth-agent	Employment: verification steps that gate an offer. Adverse findings determine whether a hire proceeds.	Any adverse finding routes to a human before any action, and the candidate is given the finding and a right to correct it before a decision. Regional Talent Acquisition Lead · Talent Acquisition Lead	Published notice shown	Named human Regional Talent Acquisition Lead	84 mo	Self-assessed	Operating with a gap No fairness monitoring is running on this agent even though it is classified high risk. It is on the monitoring backlog and has been for two quarters.
Immigration & Visa Agent A2 · people-affecting	Human Resources depth-agent	Migration and border control adjacent: assists with visa and work authorization applications affecting a person right to work in a country.	Every filing is reviewed and submitted by a licensed immigration adviser. The agent prepares; it never files. External counsel of record · Immigration adviser	On request no notice	Named human Mobility lead, then external counsel	84 mo	Not assessed	Operating with a gap Classified high risk and never formally assessed. The mitigating fact is that a licensed human files every application, which is a real control, not a paper one.
Credit Check Agent A3 · people-affecting	Finance depth-agent	Creditworthiness evaluation. High risk where the counterparty is a natural person, which applies to the sole-trader segment of the customer base.	Credit limit changes affecting sole traders route to a human analyst. Corporate counterparties are handled autonomously within band. Ana Duarte, Group Financial Controller, as escalation · Credit manager	On request no notice	Named human Credit manager, then Group Financial Controller	84 mo	Not assessed	Operating with a gap The sole-trader segment makes a commercial credit process a high-risk system for part of its population. That segment is not separated in the current configuration, so the human-review rule is applied to a broader set than strictly required and there is no fairness monitoring on it at all.

Limited risk — 2 agents

Transparency obligation applies. The affected person must be told they are dealing with an automated system.

Agent	Function	Why it is classified here	Human oversight	Explanation to the person	Appeal path	Logging	Conformity	Status
Scheduling Agent A3 · people-affecting	Human Resources depth-agent	Allocation of interview slots. Not an Annex III evaluation, but it interacts directly with candidates and so carries a transparency duty.	Exception review. Candidates waiting more than ten working days escalate automatically. Coordinator on the requisition · Recruiting coordinator	Published notice shown	Open channel Recruiting coordinator, via the candidate portal	36 mo	Self-assessed	Operating
Deprovisioning Coordination Agent A2 · people-affecting	Human Resources depth-agent	Executes a decision already made by a human. Included because a premature revocation affects a person materially.	Reversible for 30 days. Every early revocation raises an incident automatically. Exit coordinator, Hire-to-Retire tower · Exit coordinator	Internal only no notice	Named human Yusuf Adeyemi, Chief Technology Officer	36 mo	Not assessed	Operating with a gap No fairness monitoring has ever been switched on for this decision point, and one appeal has already been upheld against it.

Minimal risk — 5 agents

No specific obligation beyond ordinary governance.

Agent	Function	Why it is classified here	Human oversight	Explanation to the person	Appeal path	Logging	Conformity	Status
Adverse Impact Monitor A2	Human Resources depth-agent	A monitoring instrument, not a decision system. It observes other agents and raises alarms.	Its alarms are reviewed by the Data Protection Officer. It cannot change any outcome. Nadia Kovac · Data Protection Officer	Internal only no notice	Not applicable Not applicable. The monitor makes no decision about a person.	84 mo	Self-assessed	Operating
Pay Equity Monitor A2	Human Resources depth-agent	A monitoring instrument. It holds recommendations for human review; it does not set pay.	Holds are released by the reward partner. The monitor cannot approve anything. Marcus Weill · Reward partner	Internal only no notice	Not applicable Not applicable.	84 mo	Self-assessed	Operating
Rating Distribution Monitor A1	Human Resources depth-agent	Observes rating distributions across managers and flags anomalies for the calibration chair.	Flags are advisory to the calibration chair. Function leader chairing the session · Calibration chair	Internal only no notice	Not applicable Not applicable.	84 mo	Self-assessed	Operating
Bias Language Monitor A1	Human Resources depth-agent	Scans review text for language patterns associated with biased evaluation and prompts the writer.	Advisory prompt to the review author. No text is changed automatically. The manager writing the review · Review author	Internal only no notice	Not applicable Not applicable.	84 mo	Self-assessed	Operating
Contract Policy Agent A3	Legal depth-agent	Commercial contract analysis. No natural person is the subject of the decision.	Counsel review above the materiality threshold. Rafael Ortiz · General Counsel	Internal only no notice	Not applicable Not applicable. Commercial counterparty, not a data subject.	84 mo	Self-assessed	Operating

The appeal path, exercised

Not a policy statement — the actual contests raised, who they routed to, how long the human spent and whether the original decision survived.

Ref	Agent	Subject	Channel	Grounds	Original	Routed to	Service level	Human time	Outcome	Evidence	Status
APL-0012	Calibration Prep Agent High risk	EMP-11603 employee · UK	HR case management	Employee contests the evidence selection in their calibration pack.	Rating held at the proposed level	Priya Raman Chief People Officer	96 / 240 hr	—	Pending	HR3B-1010#1	Open
APL-0011	Screening Agent High risk	CAND-49512 candidate · US	Candidate portal	Applicant requests human review of an automated ranking under state law.	Ranked below the shortlist threshold	Ben Ortiz Talent Acquisition Lead, Americas	168 / 120 hr	—	Pending	HR2A-1021#3	Past service level
APL-0010	Merit Cycle Agent High risk	EMP-10771 employee · UK	HR case management	Employee asks for the pay-equity check result that applied to their recommendation.	Increase of 3.4 percent	Marcus Weill VP Total Rewards	264 / 240 hr	—	Pending	HR8A-1046#1	Past service level
APL-0009	Screening Agent High risk	CAND-49330 candidate · AU	Candidate portal	Applicant states an accessibility adjustment was not accounted for in the assessment stage.	Ranked below the shortlist threshold	Lin Wei Talent Acquisition Lead, APAC	62 / 120 hr	44 min	Overturned The adjustment request had not reached the assessment stage. Reassessed with the adjustment in place and shortlisted.	HR2A-1008#2	Closed
APL-0008	Succession Slate Agent High risk	EMP-08117 employee · US	HR case management	Employee questions why a peer with shorter tenure appeared on a slate they were excluded from.	Not included on the slate	Morgan Idris Function Leader, chairing the talent council	233 / 240 hr	87 min	Original upheld Slate criteria and both assessments shared. The council confirmed the original slate and recorded the reasoning.	HR3A-1045#1	Closed
APL-0007	Attrition Signal Agent High risk	EMP-12008 employee · US	HR case management	Employee learned of a retention flag during a conversation with their business partner and objects to the profiling.	Flagged at elevated attrition risk	Priya Raman Chief People Officer	96 / 240 hr	52 min	Overturned Flag withdrawn and the employee opted out of future flagging. The opt-out is now offered to the whole population, which it was not before this appeal.	HR9A-1012#1	Closed
APL-0006	Screening Agent High risk	CAND-49014 candidate · SG	Candidate portal	Applicant requests the reason for rejection and the data used.	Ranked below the shortlist threshold	Lin Wei Talent Acquisition Lead, APAC	43 / 120 hr	19 min	Original upheld Rationale and field list supplied. No change to the outcome.	HR2A-1002#2	Closed
APL-0005	Calibration Prep Agent High risk	EMP-10442 employee · SG	HR case management	Employee contests an outlier flag raised in the calibration pack.	Flagged as a downward outlier in the calibration pack	Priya Raman Chief People Officer	151 / 240 hr	71 min	Overturned The flag rested on goal data that had not been updated after a mid-year role change. Flag withdrawn and the rating stood unchanged.	HR5B-1009#5	Closed
APL-0004	Merit Cycle Agent High risk	EMP-09931 employee · AU	HR case management	Employee questions the market benchmark used to set the band position.	Increase of 2.1 percent, below the population median	Marcus Weill VP Total Rewards	205 / 240 hr	64 min	Original upheld Benchmark verified as the correct market cut for the role and location. The band position explanation was rewritten to be legible to the employee.	HR3A-1025#1	Closed
APL-0003	Succession Slate Agent High risk	EMP-11284 employee · UK	HR case management	Employee was not on the slate for a role they had been developed toward for two cycles.	Not included on the slate	Morgan Idris Function Leader, chairing the talent council	188 / 240 hr	95 min	Partially upheld Added to the slate for the next cycle. The council recorded that the readiness signal the agent used was two cycles out of date.	HR3B-1010#1	Closed
APL-0002	Screening Agent High risk	CAND-48377 candidate · US	Candidate portal	Applicant asks on what basis the application was ranked and requests human review.	Ranked below the shortlist threshold	Ben Ortiz Talent Acquisition Lead, Americas	96 / 120 hr	22 min	Original upheld Ranking rationale supplied in full. Human reviewer reached the same conclusion on the stated requirements.	HR2A-1021#2	Closed
APL-0001	Screening Agent High risk	CAND-48120 candidate · UK	Candidate portal	Applicant states that relevant experience held under a former job title was not recognized by the ranking.	Ranked below the shortlist threshold	Aisha Bello Talent Acquisition Lead, EMEA	71 / 120 hr	38 min	Overturned The former job title was not in the skills taxonomy. The applicant was added to the shortlist and the taxonomy was extended.	HR3B-1010#1	Closed
APL-0013	Deprovisioning Coordination Agent Limited risk	EMP-07740 employee · DE	IT service desk	Access was revoked four days before the agreed leaving date, blocking handover.	Access revoked on the system-held termination date	Yusuf Adeyemi Chief Technology Officer	6 / 24 hr	15 min	Overturned Termination date in the HR system had not been updated after an extension. Access restored within six hours and an incident was raised against the date synchronization.	IT1A-1020#1	Closed

Agents operating with a gap

Named, with the gap stated.

Screening AgentHigh risk

Self-assessed. No notified body involvement, and the NYC Local Law 144 independent bias audit has not been commissioned even though the agent screens candidates for New York roles.

Succession Slate AgentHigh risk

Employees are not proactively told a slate exists or that an agent helped build it. Explanation is available on request, which means available to whoever knows to ask. Two fairness alarms are open against this agent.

Calibration Prep AgentHigh risk

The German population is excluded pending the works council agreement, so the readings shown cover four jurisdictions out of five.

Attrition Signal AgentHigh risk

Employees are profiled for flight risk without being told. The opt-out exists because one employee appealed and asked for it, not because it was designed in. Review is 51 days overdue.

PIP Drafting AgentHigh risk

Volume is too low for a statistically meaningful fairness reading at 180 plans a month across four jurisdictions. The platform reports insufficient sample rather than a ratio.

Deprovisioning Coordination AgentLimited risk

No fairness monitoring has ever been switched on for this decision point, and one appeal has already been upheld against it.

Background & Right-to-Work AgentHigh risk

No fairness monitoring is running on this agent even though it is classified high risk. It is on the monitoring backlog and has been for two quarters.

Immigration & Visa AgentHigh risk

Classified high risk and never formally assessed. The mitigating fact is that a licensed human files every application, which is a real control, not a paper one.

Credit Check AgentHigh risk

The sole-trader segment makes a commercial credit process a high-risk system for part of its population. That segment is not separated in the current configuration, so the human-review rule is applied to a broader set than strictly required and there is no fairness monitoring on it at all.

What the appeal record shows

Read the outcome distribution before reading the policy.

5 of 13 appeals overturned the original outcome outright and 1 were partially upheld. An appeal path that never changes anything is a complaints inbox; one that changes a third of what reaches it is telling you something about the decisions upstream of it.

2 appeals are past their stated service level. The median human spent 48 minutes on a review. That number is shown because an oversight measure that takes two minutes is a rubber stamp, and the only way to tell the difference is to record the time.

13 of 13 appeals are wired into the evidence spine, so the contest and its resolution sit in the same chain as the decision they contest. An appeal recorded only in a case-management tool is not evidence.

Actions

What is waiting on a person

Risk classification is a legal exercise before it is an engineering one. Nothing below can be cleared by an agent.

Get an independent assessment

Self-assessed only. No external party has looked.

Close a conformity gap on a running class

Operating today with a named gap against the obligation.

Assess an unassessed class

Classified, never put through a conformity assessment.

Respond to a class operating under alarm

A live alarm against a classified system.

Operations

What this desk is allowed to start

A surface that only reports is not operable. This is the work this page can set in motion, and the bound it runs into.

Trigger and bound

This desk can classify a system, record the obligations that follow from the class, and track the evidence against each one. It cannot certify conformity, issue a transparency notice, or waive an obligation.

Live observability

What the record shows right now

Risk class distribution across the classified estate.

Current distribution

16 classes

Minimal531%

Limited213%

High956%

Is policy and strategy coming to fruition

Whether the written intent is holding here

9 of 16 classes are high risk. 16 have had no independent assessment.

Not holding on the record

The policy commitment is that every agent with a consequential effect on a person is classified and assessed before it runs. Classification holds — all 16 classes carry one. Assessment does not: 13 are self-assessed and 3 have not been assessed at all, so 100% of the classified estate runs on this organization’s own opinion of its conformity. 8 classes are operating with a written gap and 1 under an alarm. Nothing on this page constitutes a regulatory determination.

What this page is, and what it is not

The classification, the oversight measures, the explanation surfaces and the appeal routes are real structures. The appeals themselves are modeled cases running through a real mechanism — they route to a named person, carry a service level, record the minutes a human spent, and land in the evidence spine.

3 agents have no conformity assessment on record and 13 were assessed by the team that operates them. Self-assessment is a legitimate route under the regulation and it is not independence; both are labeled so nobody has to infer it.

6 agents record their reasoning internally but surface nothing to the affected person. For a minimal-risk agent that is proportionate. Where it appears against an agent that touches a person, it is the gap this page exists to surface.