The bounded world model and command surface behind every KUON EVM analyst. Not a diagram of the domain — a typed, replayable, reversible runtime that turns intent into evidence-led action against Empower, the source of record. The EVM Ontology Viewer is what the principle looks like when an analyst can drive it.
The EVM Ontology Viewer is the visual surface of the Agentic Ontology. Same typed commands, same evidence chain, same rollback verbs — rendered as a navigable world model the analyst can drive. Below: the opening session, with every region named in the analyst's vocabulary.
A · T · C · R · S — All / Technical / Cost / Risk / Schedule. Each glyph is color-tracked to the branch family it connects to. The rail is a typed query: it shows only the named paths directly reachable from the currently selected node, not the whole ontology as a checklist.ROOT · EVM Analyst Reasoning Ontology · Type Root · Degree 126 · 16 key relationship chips. Every chip on every tab is a typed, actionable control. Click one and the viewer builds the exact Empower CLI command and posts it to the local bridge — Empower is the source of record, and the ontology never invents the number.The Agentic Ontology is not documentation. It is not a graph database behind a query language. It is an operational runtime — bounded, typed, auditable, reversible — and for EVM analysts using Empower, that means every variance claim, every evidence citation, and every narrative draft traces back to a named command against Empower. Not an interpretation. Not an inference. A typed action whose trail any reviewer can walk.
A declared, typed set of objects — branches, evidence classes, reasoning patterns, artifacts. Nothing implicit. Nothing hidden. If the ontology cannot name it, the analyst cannot claim it, and the reviewer will not be asked to accept it.
Every relationship has a type. Every command has a schema. The ontology refuses to skip a step — and that refusal is the point. It is what makes the question "where did this number come from?" answerable in one chip click.
Every selection, every chip click, every CLI call is logged with its actor, intent, command, and result. The CAM and the reviewer walk the same path. No silent moves. No invisible state. The audit is not a separate document — it is the world model in motion.
Every state change carries a one-verb rollback. Nothing the analyst does in the viewer is irreversible. The ontology stays honest because undo is always one command away, and the audit log shows when undo was needed and what it restored.
The principle is small; the daily-life consequences for the CAM, the analyst, and the reviewer are large. Four workflows collapse into one continuous, evidence-led surface — and the world model never drifts from Empower.
Every claim in the CAM's draft is a typed relationship to a named Empower artifact. Reviewers stop asking "where did this number come from?" — the chip shows them, the trail logs when, and the variance explanation owns its evidence from the first sentence to the last.
Each branch, pattern, and conclusion carries its evidence class forward. When the analyst clicks a chip on the right panel, the ontology walks the chain — Empower export → artifact → narrative claim. The chain is the audit, and the audit is always one click away from the analyst, the reviewer, and the contract lead.
"What are the cost drivers off-plan?" is a single rail selection. "What evidence backs this narrative?" is one chip click. "What pattern applies to this corrective action?" is one tab. The hardest review questions become a path through the world model, not a search through exports.
When a chip posts an Empower CLI command, the response is logged with the source ontology node. The CAM owns the explanation, and the explanation owns its trail. The narrative draft and the audit are the same document — not two documents that have to be reconciled under review pressure.
Empower is the source of record. The Agentic Ontology never invents a number. When the analyst clicks a chip in the right panel, the viewer builds the exact Empower CLI command the analyst would type — empower artifact open --artifact-id "BCWR by CAM" — then routes it through the local Empower CLI bridge running in the analyst's environment. If the bridge isn't reachable, the viewer says so and copies the command to the clipboard. Never a fake success. Never a phantom result. The audit trail is honest because the failure mode is honest.
empower artifact open --artifact-id "BCWR by CAM" POST {empower-cli-bridge}/empower-cli // local bridge in the analyst's environment Content-Type: application/json { "command": "empower artifact open --artifact-id \"BCWR by CAM\"", "action": "open-artifact", "artifact": { "id": "bcwr-by-cam", "label": "BCWR by CAM", "nativeId": "empower:...", "alias": "bcwr-by-cam", "family": "chart", "branches": [], "patterns": [], "evidenceClasses": [] }, "context": { "id": "selected-node-id", "label": "selected-node-label" } }
The ontology is live. The principle is committed. Talk to KUON about how the Agentic Ontology can land inside your EVM review cycle — for the CAM, the analyst, and the reviewer, with Empower as the source of record.