---
title: "Your Ontology Records Permission. It Does Not Resolve Authority."
url: "https://atlan.com/context-and-chaos/issue/your-ontology-records-permission-not-authority/"
description: "Agents act on whoever signed off. Decisions often turn on someone else's judgment. The Authority Resolution Framework measures that gap."
keywords: "Ontologies, AI Agents, Data Governance, Context Engineering"
---

> Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/

A Context & Chaos guest take by Parviz Shariff, Principal Enterprise Architect, Data & AI at Coca-Cola HBC, where he leads data and AI readiness initiatives across the group. Published October 7, 2026 (16 min read). His argument: an agent configured against the documented structure of a decision (the scoring matrix, the approval path, the permissions) cannot recover the judgment the decision actually relies on. The Authority Resolution Framework (ARF) maps both the formal and the lived side of a decision into a machine-readable ontology and measures the gap between them before an agent acts.

## The problem

- Technology and vendor selection is shaped by relationships, institutional memory and who has standing to say "we tried that before." The scoring matrix, RFP response and technical evaluation record only what the process was designed to capture. An agent deployed into procurement, negotiation or contract management "was configured against the scoring matrix, not against the room."
- When an agent defers only to the documented approver, leaves out someone whose judgment the decision relies on, or ignores a constraint everyone already knew, the model has not malfunctioned. That is an authority failure.
- **Vendor selection example.** The documented path required sign-off from architecture, legal, procurement and security. Engineering exerted significant influence that was neither obvious nor recorded. The selection completed cleanly on paper; what followed was technical debt, implementation rework and several months of delay.
- The bill is mostly not paid in tokens: findings get corrected offline, the next decision of the same shape gets escalated rather than trusted, and launches wait while someone reconstructs who should have been involved. In the author's experience (explicitly experience, not measurement), one authority misread costs one to three extra escalation cycles, each pulling four to six people for two to six hours.

## Authority is not one thing

Five objects, resolved across five domains:

1. **Formal right**: what policy says a role may decide.
2. **System permission**: what the software will allow.
3. **Recorded approval**: whose name lands in the audit trail.
4. **Practical influence**: whose judgment the decision turns on. It leaves only fragments (meeting invites, threads, document histories, memories) and is less a form of authority than the evidence of whether the other four are exercised as intended.
5. **Agentic mandate**: what the agent itself may see and do, set across engineering, procurement, security and vendor configuration before the agent meets any decision, and the most likely to go unaudited.

The five domains say where evidence lives: social (who holds and influences a role), business (terms and policies), process (the approval path), machine (what the system permits) and real-world (the state and consequence the decision acts on). The piece builds on earlier issues: [Gartner Hype Cycles 2026: Nobody Owns Context](https://atlan.com/context-and-chaos/issue/gartner-hype-cycles-2026-nobody-owns-context/), [Your Agents Are Code. Stop Governing Them Like Documents.](https://atlan.com/context-and-chaos/issue/your-agents-are-code-not-documents/) and Amanda Darcangelo's [Your AI Context Layer Is Being Built on Stale Metadata](https://atlan.com/context-and-chaos/issue/your-ai-context-layer-is-being-built-on-stale-metadata/). Unlike stale metadata, practical influence never entered a governed record, so there is nothing to restore.

## Two failure shapes

A new hire notices who gets called, whose objection stops the room and whose approval is ceremonial (see Vivek Dubey's [The Onboarding Gap is Killing your AI Agents](https://atlan.com/context-and-chaos/issue/the-onboarding-gap-is-killing-your-ai-agents/)). An agent sees policies, role records, workflows and permissions, so it is right about the organization that was documented.

- **Practice exceeds mandate**: someone exercises more influence than their mandate grants and the agent leaves them out. Findable, but in the vendor selection nobody noticed until the rework did.
- **Mandate exceeds practice** (the worse case): a formally authorized person approves while the substantive judgment came from elsewhere. The record is valid and the safeguard has become ceremonial; the agent obtains the correct signature without the independent judgment it was meant to represent.
- Diagram: the two failures side by side, a missed participant and a valid signature with a missing safeguard. "Only one of the two looks wrong on paper."

## Score the gap, not the legitimacy

- Divergence has a **magnitude** (how far documented and practised authority diverge; zero means aligned) and a **direction** (positive: practice exceeds mandate; negative: mandate exceeds practice). Both failure shapes can be read off the sign.
- Neither field measures legitimacy: a large divergence does not make informal practice right, and a small one does not prove the formal path safe.
- Zero settles the authority question and nothing more; the action still runs under its own risk tier. A non-zero result is interpreted through a [risk policy defined in advance](https://atlan.com/context-and-chaos/issue/four-architectures-that-make-ai-work/).
- Diagram: "Magnitude measures the gap. Direction names the failure."

## What the ontology has to hold

- Both sides: roles, policies, permissions and approval paths; and whose judgment a decision relies on, and why. [Prukalpa's definition of a context layer](https://atlan.com/context-and-chaos/issue/what-an-enterprise-context-layer-actually-is/) names the same gap: a semantic layer can tell an agent what gross margin is, but not which approval path matters in practice.
- An agent using only the structural half produces outputs that are formally correct and operationally wrong, in a way an audit built on formal authorization does not detect.
- **Authority Relation**: an actor performing an action on an object under a stated mandate, linked to the governing policy, process step, system permission and real-world state. `resolvesTo` lets a validator check the references exist, agree and remain current; `affirmedBy` puts business, technical and governance review inside the record.
- At runtime the agent traverses from the permission or object it is about to rely on to the governing Authority Relation and reads divergence magnitude, direction, evidence and affirmation status. "The ontology does not decide whom the agent should obey."
- Diagram: a proposed action resolving through an Authority Relation and its governing evidence to a runtime decision to act, escalate or refuse.
- Three instruments estimate divergence: compare documented approvers with decision logs; map where influence concentrates, with the people on the map knowing why; interview people inside the flow. The author states none has been empirically tested and calls the measurement method the framework's most original part and its largest open empirical question.

## Missing capability or missing independence?

- The same finding supports opposite readings. Usually the informal path routes around missing capability (in the vendor selection, engineering was never part of the workflow), and formalizing it is right.
- Rarer and more serious: the path routes around segregation of duties, independent regulatory sign-off or a required second pair of eyes. There, "the slowness was the mechanism," and encoding the path into an agent would turn a control failure into an executable rule.
- Diagram: "The same divergence can signal adaptation or control failure."

## Four options, and the one to pick today

An agent facing material divergence can defer to the informal decision-maker, escalate to a human, refuse above an agreed risk threshold, or the organization can formally delegate or revoke the authority first. The author picks human escalation today, with refusal above a threshold where escalation is unavailable, and would not encode the informal path. Frequent refusal teaches people to route around the system, so the threshold needs a governance judgment weighing the cost of refusing a legitimate decision against proceeding on an illegitimate one.

## Starting, and the framework's limits

- Maturity runs from tacit authority, through authority written down but never checked against practice, to divergence measured with a direction; most enterprises deploying agents sit in the first two levels for high-stakes decisions.
- Start with one consequential decision in one business unit. Ask the people inside the flow separately; "two practitioners disagreeing about who decides is itself evidence."
- No finding should be used until business, technical and governance have each affirmed it independently. Timestamps matter: a verbal approval logged after the fact can carry the logging time, not the decision time.
- The framework cannot legitimize what it finds; the organization still decides whether influence is recognized, delegated, constrained or stopped. The team operating ARF should not validate its own interpretation, and agent-design authority should sit under independent oversight.
- An agent that cannot resolve authority either hands decisions back to people or acts confidently on an incomplete model of the organization. "One erodes the value the agent was bought to create. The other creates risk at machine speed."

The views expressed are the author's own, based on his experience and independent research for ARF, and do not represent the views of any organisation. Contributions reflect their authors' views; Context & Chaos curates and vets submissions, with no paid promotions. All issues: [https://atlan.com/context-and-chaos/](https://atlan.com/context-and-chaos/).