Compliance review
Agents: the compliance agent (compliance) and Rules Recon (rules_recon).
What it does
It puts what your policies say next to what your systems actually do, names where they part, and writes the next step in business language for the person who owns the control. Its jobs:
- Rules Recon. Finds the business and governance rules in your data, documents,
telemetry limits and code, and classifies each one:
- Aligned: the implementation matches the policy.
- Drift: both exist, but the implementation has moved away from the policy.
- Shadow: logic runs in the data or code with no policy behind it.
- Missing: a policy exists and nothing implements it.
- Framework checks. Gives an overall status of Pass, Fail or Warning against the frameworks you work to, and lists each violation with its rule and a remediation.
- Personal data. Flags personal data shown without masking in reports, query results and logs, and drafts the masking proposal and the data protection officer’s notice.
- Emissions and supplier risk. Works out freight emissions with the method beside the figure, and summarises supplier compliance risk from the evidence provided.
What it reads
Policies, SOPs, contracts and SLAs; the running configuration and data of connected systems; logs and report extracts; telemetry limits; and, for Rules Recon’s code checks, source code and configuration files. It can run read-only queries against the connection a run was started with.
What it produces
- A list of rules, each with its plain-English logic, the source it came from, its severity, its maturity (Low, Medium or High) and its reconciliation status. You see them on Why, where each rule’s policy drawer sets its enforcement posture.
- A compliance status, violations with a remediation each, and an audit trail of which systems were checked and with what result, on the Compliance view opened from Actionable Twins. See Compliance, finance, operations.
- Remediation orders for a person to assign, route and decide, listed in the Waiting for you lane on Actionable Twins and on the Compliance view.
Human oversight
- A person assigns, routes and decides every remediation order, and the server records
who did each (
AssignRemediation,SetRemediationRoute,DecideRemediation). - Rule enforcement posture is set by an operator (
UpdateRulePolicy). The agent can’t change a finding’s status that way. - The agent never changes configuration, masking or data. Its remediations are steps for a person to take, such as a ticket or an escalation, never SQL or code fixes.
Model
- Compliance and rule extraction: the general reasoning tier.
- Rules Recon code checks: the code tier.
Where it runs and data handling
On the deployment’s own model plane: Runink’s shared machines on a Trial, your own cloud on Cloud and your own servers on On-premises. No outside AI service is called on any plan. See Facts shared by every card.
Guardrails
- Cites the source of every rule, and doesn’t invent files or policies. When nothing is found, it reports no rules.
- Reports no confidence score for its own verdict. The reconciliation status is the verdict.
- Refuses to create rules that discriminate, that retaliate against a particular employee, supplier or partner, or that bypass audits, governance checks or ethical sourcing obligations.
- Refuses to falsify emission figures or supplier ESG ratings, to weaken business continuity plans, or to bypass the personal-data masking and breach-notice triggers.
How to read its results
- It doesn’t certify compliance with any framework. A Pass is the agent’s reading of the evidence it was given.
- “Not assessed” is a separate result from “passed”. Where evidence is absent, the control is not shown as green.
- A breach-notice draft is a starting point for the data protection officer, not a legal determination.
- It gives no legal advice.