Skip to main content

Glossary

Terms used throughout this site, defined as they are used here rather than as they appear in product documentation.

ACP — App Customization Policy​

A Microsoft Teams tenant policy governing whether app metadata (name, icon, description) may be changed. Raybot's Teams distribution is blocked because ACP consent was never granted in the tenant; the edit-details request carries isConsentProvidedToChangeACPToAny=false. Requires a tenant admin to resolve.

Agent ID​

The platform-minted GUID identifying an agent. Created on the first successful save, not when the create dialog is submitted. Raybot's is 4de4ada9-ad61-4c22-a0d0-a0bd1c189f4e.

Agentic retrieval​

Retrieval that plans queries, executes searches, and synthesises an answer before returning, rather than returning raw passages. Microsoft IQ performs agentic retrieval; a plain public-website knowledge source does not.

Confidence level​

The audit's evidence grade. One of:

  • CONFIRMED — observed live in a running system.
  • SOURCE-INSPECTED — read from source code or a published artifact, not observed executing.
  • UNCONFIRMED — inferred or asserted without direct evidence.

This audit is CONFIRMED throughout.

Coral​

The internal codename for Copilot Studio's next-generation runtime and maker surface. Distinguished from the classic experience by the unified Build designer, the right-hand Agent configuration rail, inline publishing with no confirmation dialog, and the coral.* test-id convention.

data-testid​

The DOM attribute the audit keys on. 983 of the 1,627 catalogued elements carry one, resolving to 177 unique values, of which 76 appear on more than one surface. Test ids are the most stable automation handle in Coral, but they are not unique and not universal.

Derived label​

A heuristic accessible-name approximation computed by the audit harness. The heuristic that produced each value is recorded in labelSource. It is not a strict accessible-name computation, so the 228 elements with no derivable label are a lower bound on accessibility gaps, not a finding.

Foundry IQ​

The Azure AI Search feature backing Microsoft IQ. Exposes /knowledgeSources and /knowledgeBases on the 2025-11-01-preview REST surface. Requires Basic tier or higher for knowledge bases to appear in the Copilot Studio picker.

Interactive​

A computed property in the element audit, derived from native control type, presence of href, ARIA role, tabindex, or contenteditable. Computed independently of data-testid. 1,115 of 1,627 elements are interactive.

Island Gateway​

The host serving channel management operations, at powervamg.{region}.gateway.prod.island.powerapps.com. Separate from Dataverse. The split explains how a channel can report enabled in a Dataverse-backed UI while gateway-side registration has failed.

Knowledge base​

In Foundry IQ, the queryable aggregate built over one or more knowledge sources. Raybot's is raybot-kb, built over the ray-docs-web source with outputMode: answerSynthesis.

Knowledge source​

In Foundry IQ, a single origin of content. Raybot's is ray-docs-web, of kind: web, which requires no additional parameters.

MCP — Model Context Protocol​

The protocol used to drive the browser during the build. playwright-browser_* tools performed every UI action; filesystem tools handled local files. MCP also appears as a tool catalog tab inside Copilot Studio itself.

Microsoft IQ​

The Copilot Studio right-rail section that binds an agent to a Foundry IQ knowledge base. The add dialog offers exactly one option, Foundry IQ (Preview). It binds; it does not create.

Publish v2​

The asynchronous publish mechanism. A POST to a publish-operations endpoint returns an operation identifier which is then long-polled. Coral has no publish confirmation dialog, so the command-bar button text is the only progress signal.

PvaProvision​

The Dataverse create model observed on POST /api/data/v9.2/bots when an agent is created.

Ray​

The distributed computing framework Raybot is an expert on. Its documentation at https://docs.ray.io is the agent's sole knowledge source. Relevant subsystems: Ray Core, Data, Train, Tune, Serve, RLlib. Relevant cluster topics: head and worker nodes, autoscaling, KubeRay, the Dashboard, and placement groups.

Schema name​

The generated Dataverse logical name for an agent, carrying a random suffix. Read-only from the moment it is assigned. Raybot's is Default_Raybot_vyN1Rz.

Semantic ID​

A stable identifier assigned by the audit. Two layers exist:

  • UI — coral.<area>.<surface>.<component>[.<state>]
  • API — {namespace}:{surface}:{group}:{operation}

See Naming schema.

Skill​

A named, instruction-driven capability authored in the skill editor. Created either blank or from a generated draft. Raybot's is ray-cluster-troubleshooter, authored blank.

Surface​

One audited screen or dialog. Sixteen were catalogued. A surface is the unit the element audit and the naming schema both organise around.

Test set​

A CSV-imported collection of question-and-expected-response pairs used by Evaluate. Conversation-type sets accept at most six pairs; the Reference field caps at 1000 characters. Raybot's is "Ray Knowledge Eval - 6Q".