Configure Accountable Oversight
Your people provide oversight. RateLoop supports the configured workflow and records its evidence.
Start with who has authority
Customer-invited reviewers can be people your organization designates and authorizes for its oversight workflow. RateLoop records the scope you assign to them, but your organization remains responsible for their competence, training, authority, and support.
Controls the workflow exposes
Capability 01
See operation and exceptions
Your designated people monitor operation from the oversight dashboard: sampling coverage, response latency, disagreement, and blocked outputs, per scope. In-app, email, and browser alerts flag disagreement spikes, coverage-floor hits, blocked outputs, failed or expired reviews, and workspace stops, and event webhooks feed your own monitoring. Per-agent evidence summaries show the declared provider and model alongside observed workflows and risk tiers — declared metadata labelled host-reported, not independently verified.
You remain responsible for watching those surfaces, understanding the agent's capacities and limitations, and acting on what they show for your use case.
Relevant where a provider addresses Article 14(4)(a) monitoring.
Capability 02
Collect blinded judgments
Blinded reviewers record separate judgments before your decision: separate submissions keep early judgments private, so reviewers cannot anchor on each other. The decision prompt has no preselected choice, and available disagreement and calibration signals appear above the decision buttons. The dashboard's override-rate metric is a workspace-wide 30-day aggregate, not the deciding person's own trend.
You remain responsible for staying aware of the pull to over-rely on the system and keeping each decision a considered one.
Relevant where a provider addresses Article 14(4)(b) automation bias.
Capability 03
Put the output in context
The owner case view shows the oversight person the actual output, its source context, reviewer rationales, and surfaced disagreement before their decision. Access to case material remains limited to authorized workspace members.
You remain responsible for correctly interpreting the output within your domain, workflow, and context.
Relevant where a provider addresses Article 14(4)(c) interpretation.
Capability 04
Record the human decision
Every go, revise, and stop decision is recorded against the case. Per-output override records carry a required reasons field and join the workspace audit chain, and the override rate is a first-class metric on the dashboard and in coverage exports.
You remain responsible for deciding when to disregard, override, or reverse an output.
Relevant where a provider addresses Article 14(4)(d) disregard, override, or reversal.
Capability 05
Control intervention and stop
Codex, plugin, and MCP integrations are advisory: they report the review lifecycle but do not verify interception or withheld delivery. RateLoop's workspace stop blocks new review-triggered release authorizations. Your workflow must honor that state when delivering outputs. Releasing the stop restores no agent grant automatically.
You remain responsible for choosing which outputs are gated, when to intervene, and when to halt.
Relevant where a provider addresses Article 14(4)(e) intervention or stop controls.
The deployer's people and process
For Annex III high-risk systems, Articles 14 and 26 apply from 2 December 2027 under Regulation (EU) 2026/1744. Article 26(2) will require deployers to assign oversight to people with the necessary competence, training, authority, and support.
RateLoop records oversight designations with competence basis, training completed, authority scope, and expiry. It records the reviews, decisions, overrides, and stops those people make. These records can support your evidence of how oversight was organized; they do not decide whether the Act applies or establish that the people or process meet Article 26(2).
Article 26(3) preserves the deployer's freedom to organize its own resources and activities. That is consistent with using an outside service, but it does not turn an outside reviewer into your authorized oversight person; your organization must still make and support the designation required by Article 26(2).
If you also provide the AI system
From the applicable date, Article 14 will bind the provider of a high-risk AI system and require the system to be designed and developed so that natural persons can effectively oversee it. RateLoop operates around the customer's system and does not by itself satisfy that provider design duty. These capabilities may support an oversight process for a system that was designed to expose the necessary controls and information.
A customer that is also the system provider should address Article 25(4) only where RateLoop supplies services used or integrated in that high-risk system. In that case the provider and RateLoop must document the required information, capabilities, technical access, assistance, and relevant expertise in a written agreement. The supplier schedule in the service terms records that boundary; it is not an attestation or a claim that RateLoop is integrated into every customer system.
Designation, competence, and literacy
RateLoop records oversight designations with attestation records — competence basis, training completed, and authority granted — exportable as an assignment record, and emits audit events on every role assignment and change. Reviewer and oversight-person training and calibration records can support measures under Article 4 to develop AI literacy. Article 4, as amended by Regulation (EU) 2026/1744, requires measures that support that development and does not require guaranteeing any specific level for any individual. Choosing those people, supporting their literacy, and ensuring their competence, training, and authority remains yours.
Audit and evidence exports map to the Commission's draft Article 73 serious-incident reporting template — labelled draft-aligned until the template is final — and the workspace's oversight configuration exports as a factual description of the implemented oversight measures, usable as input for an Article 27 fundamental-rights impact assessment.
The shared-responsibility matrix and the exportable evidence behind each capability live on Evidence reference.