Skip to content

Glossary ​

Each term has one meaning in these docs and in the app. A term marked Planned is not built. A term marked Vision is long-term direction (see Where DutyGraph is going).

TermMeaning
AdvisorThe DutyGraph user who runs an engagement for a company: an advisor or a forward-deployed engineer (FDE). See Run engagements.
Agent Authority RecordPlanned. One auditable view of an agent's purpose, owner, Task Cards, capabilities, scope calculation, decisions, issued authority, lifecycle and revalidation history. A view over existing records, not new storage.
Agent Build PacketVision. A build-ready package for an engineer or a tool: Task Card, rules, exceptions, permissions, cases and tests.
Agent GrantThe authority issued to one agent under conditions: its scopes, controls and end date. Built as an issued agent's scopes and credential. A grant traced to validated business capabilities is Planned.
Agent manifestThe versioned, hashed description of an agent: its owner, Task Cards, steps, tools and scopes. Code builds it from the Company Repo. A draft gives no access.
Agent planA private 30-day page that shows a person the agent proposed for their work. Marked "Proposed plan · not deployed".
Agent rulesThe rule pack for non-human principals. The default pack has 14 rules, including AGT-12: "No agent approves work."
Agent StudioVision. A compiler that would turn more of a build packet into a working agent.
AI modeThe proposed AI role on a Task Card: People only, AI assists the person, AI drafts for review, AI recommends and a person decides, AI acts after a person approves, AI acts within set limits, or AI not allowed. A proposal, not a permission.
Approval chainThe people who decide on an agent request, in order. Default: manager, role owner, Internal Controls, IAM.
As understoodHow DutyGraph stores an answer: the version the person confirmed on "Here's what we heard". Their own words are kept too.
Auto-trimmedA scope removed because the owner does not hold it or a rule removes it. The requester cannot add it back.
Bottom-up capturePlanned. When an employee requests an agent and no confirmed Task Card fits, a short capture of that one task, confirmed by their manager and validated by the application owner. The request tools are Built. See Just-in-time task capture.
BuiltIn the code, with tests. Not a claim that a customer runs it. See Built and planned.
Business capabilityPlanned. An action a Task Card step requires, named in business terms, for example "place a hold on a supplier invoice". Each maps to technical entitlements. See The key join.
CEIL-01The agent rule that keeps an agent's scopes within its owner's holdings. Enforced by the evaluator, with no exception. A ceiling, not a justification.
CEIL-02The agent rule that an agent's scopes never exceed what its Task Cards need. In the default rule pack, but not enforced yet.
Chain intactWhat Verify chain shows when every ledger entry matches its hash: "Chain intact: N records match their hashes."
Company RepoThe company as files an AI can read, with a state on each file. Built from DutyGraph's records. Holdings never become Repo files.
Company researchA run over public sources about a company. Each finding keeps its source and stays Inferred until a person at the company confirms it.
Company SnapshotThe company's front page (Overview): what is known, what is needed, Discovery progress. It prints or saves as a PDF.
Decision ExportPlanned. A contract that sends DutyGraph's approved decision, its justification, approvals, controls, evidence references and lifetime to the company's IAM or IT service management. See Approval, provisioning and revalidation.
DutyA continuing responsibility that holds one or more Task Cards.
DutyGraph PulseVision. A short async check-in that would replace a status meeting. Part of Meeting OS.
Employee Knowledge CompensationPartial. DutyGraph's model for paying employees for the work knowledge they put into accepted Task Cards and the improvements they make. Per-card amounts, a pool and a time-saved bonus are Built in Earn; the rest is Planned.
EngagementThe scoped work an advisor runs for one company: its outcome, scope, dates and people.
EntitlementA permission in a business system. Today a catalog scope, or an entitlement name in an access export that the System Catalog maps to scopes. Mappings validated by an application owner are Planned.
EscalatedA scope that needs a human decision under a rule, for example an SoD conflict that needs a mitigating control.
Event ImportPlanned. A contract that brings events back from the company's identity systems, such as provisioning completed, credentials revoked or owner left, and reconciles them with what DutyGraph approved.
Evidence exportOne agent or a filtered population exported as CSV or JSON, with the ledger slice. No secrets, no business data.
Evidence statesThe state on each Company Repo file: Confirmed, Observed, Inferred, Disputed, Unknown, Proposed, Generated or Slot. Confirmed means a person with authority confirmed it in DutyGraph, where it is recorded.
FDEForward-deployed engineer: an engineer who works inside a client's business to build with it.
FreshnessThe state of a saved manifest: Fresh, Out of date (a cited file changed) or Broken (a cited file was removed, a role holder left, or access or rules are now stricter).
GatewayThe optional DutyGraph endpoint (POST /api/mcp/agent) that checks every call an issued agent routes through it. Each agent has its own dga_… credential. It has no per-call human approval step.
HandoffWhat one Task Card passes to another: what is handed over, what the receiver needs, the time limit and who handles exceptions.
HoldingOne system/action:object scope a person holds, with its source and date. Holdings are the ceiling for an agent: a ceiling, not a justification.
Hosted agentPlanned. An approved agent that runs on the company's own AI vendor account. The vendor runs it; the gateway checks calls routed through it.
Human-delegated agentAn agent that acts for one named owner, with scopes within that owner's holdings and the delegation rules. The model DutyGraph supports today.
IAMIdentity and access management.
IGAIdentity governance and administration.
InboxThe advisor's list of things to act on, grouped as Review, Decide and Send / Chase.
InheritedA requested scope that is catalogued, that the owner holds and that no rule removes or escalates. It does not yet mean a Task Card proves the scope is needed.
KARPlanned. Knowledge Above Replacement: the yearly value a company would lose or rebuild without one person's captured knowledge, plus their measured improvements. Credit only, shown as a range.
Key joinThe target: Task Card × owner's holdings × System Catalog × rule packs = what an agent may hold, and why. Partial today: the Task Card check (CEIL-02) is not enforced.
KickoffThe executive kickoff: a meeting with the company's leaders. Code builds the agenda; the advisor pastes the transcript in afterwards.
LedgerThe hash-linked, append-only record of governance decisions for a company.
Living DutyGraphThe ongoing service after a first engagement: scheduled research refreshes, a digest the advisor approves before the sponsor sees it, and renewal reminders.
MCPModel Context Protocol. The way an AI chat calls DutyGraph's tools: advisor tools, employee request tools and the agent gateway.
Meeting OSVision. Meetings that run async first and update the operating model. Not built.
Mitigating controlA control that lets an SoD conflict stand. Assigned to one principal and one risk. Covers an agent only if it is marked agent-eligible.
Operating Intelligence GraphVision. The AI-readable operating model of a company. The Company Repo is its built start.
OpportunityWork that could change. The Opportunities page shows each person's Task Cards in hours a week, the proposed agents and the possible constraints.
OutcomeThe measured result of a change, read against a prediction recorded first. The proof rule reads Supported, Not supported or Inconclusive.
OwnerThe person an agent acts for. Its scopes never exceed theirs. Also, the person accountable for a Task Card.
ParticipantAn employee who answers asks from a private link, with no account. Requesting an agent needs a DutyGraph sign-in.
PerformerThe person who does a Task Card's work today. Not an approval role.
PlannedWritten down, not built.
PlaybookPlanned. An approved agent served as instructions inside the employee's own chat, using their own access.
Pod-down discoveryPlanned. Discovery sponsored by one department head for a bounded unit, recording links to other departments without a full company map. See Three ways in.
Possible constraintA testable explanation of what may limit a flow's output, keyed C1, C2. A hypothesis until a test supports it.
PredictedHours a week with an agent, or a change's expected effect, recorded before it happens. Agent predictions exist only in the sample companies.
Private linkA link for one named person to their own answer page. It opens without an account and can be revoked.
Proposed agentAn agent that code proposes for one named person, inside that person's access. It needs a Task Card with an owner and a proposed AI mode.
RegisterThe list of every agent requested through DutyGraph, plus drafts never requested. In the sample, identities "found outside DutyGraph" are illustrative data. Reconciling with a real enterprise inventory is Planned.
DutyGraph AgentA panel in the Company Repo that answers, reviews, proposes edits and drafts asks. It only proposes.
ReportedHours a week from the person's own answer. "Reported · sample" in the sample companies.
RequesterThe person who asks for an agent. The requester never decides on the request.
RevalidationChecking an authorization again when the information behind it changes: holdings, a Task Card, an expiry, an owner. Manifest freshness, stale-decision refusal, revocation and expiry are Built. Recomputing issued agents is Planned. See Approval, provisioning and revalidation.
RouteA recorded decision on what to change for an accepted possible constraint, chosen from a ladder of nine options, cheapest first. DutyGraph runs no change.
Rule packA versioned set of rules: agent rules, SoD rules or mitigating controls.
Run trackingPlanned. A record of what an agent actually did, from OpenTelemetry traces, next to what it was approved to do. Evidence only; it never grants or blocks.
Sample companyA company in DutyGraph for demos. Cobalt Industrial Supply is fictional. An illustrative distributor sample shows governance. Its people, access and numbers are illustrative.
ScopeOne permission, written system/action:object, for example oracle-ebs/approve:invoice-hold.
SealAn optional Ed25519 seal from DutyGraph's server key on each ledger entry. Off unless the server has a key.
Service-owned agentPlanned, as a future consideration. An agent owned by a department or service, with an accountable human sponsor and its own approved authorization, not tied to one person's holdings. Not supported.
SimulatedThe screen works, but no real outside system is connected.
SlotA Company Repo file that nothing has filled yet.
SoDSegregation of duties. Pairs of scopes one principal must not hold together. DutyGraph checks the agent alone, then the owner plus all their agents.
SponsorThe client contact who receives the first touch and answers the sponsor packet.
Sponsor packetThe first ask to the sponsor: check the company picture, answer research gaps, upload the team list, choose kickoff attendees and confirm scope.
Step markersThe mark on each Task Card step: [H] a person does it, [A] AI does it, [S] shared (AI drafts, a person confirms).
System CatalogThe list of systems and the scopes each offers.
Task CardOne piece of recurring work, described by the person who does it and confirmed. In governance, the work a request names. Proving each scope from it is Partial.
Team askA question set DutyGraph sends one person by private link. They answer on a phone, by voice or text.
Three decisionsThe only choices at each approval step: Approve as proposed, Approve with changes (narrow only) and Deny (with a reason).
To reviewThe Company Repo tab where answers, proposed work and research findings wait. Nothing applies until a person accepts it.
Unresolved Mission LineagePlanned. The mark on a task with no confirmed link to why it exists, such as a goal, a mandate or a legal requirement. It never means the task is unnecessary.
Value SpineThe ordered stages of the company's value chain. The Work Map's Process map places Task Cards in these stages.
Verify chainThe ledger button that recomputes every hash.
Work MapHow work moves through the business, in four tabs: Process map, Work flows, Task Cards and People.
WorkflowA chain of Task Cards, across 2 or more people, that carries one unit of work (an order, an invoice) from start to done, with the handoffs between them. DutyGraph suggests workflows from the handoffs and the advisor confirms each. One person's job is a Task Card, not a workflow. Also called a flow.
Workflow GainsharingPlanned. A program in which a company shares the measured gain with the employee who runs a Task Card in less time or produces more output. Runs end at the hand-off, the receiver confirms them, and the baseline is frozen for a period.
Workflow coachPlanned (a first part is built as the client app's DutyGraph Agent: make a card faster, finish a card, time a run). An AI assistant in the employee's app that works from the role persona and the person's Task Cards. It proposes one change at a time to make the work faster, cheaper or safer. Its AI use check records which AI tool, model and account a step uses.

INFO

Status. Each term has the status of the page it links to. Terms marked Planned or Vision are not built.

Source: gitbook/product/reference/glossary.md in the DutyGraph repository.

DutyGraph. Pages label each feature Built, In review or Planned.