Concepts

Glossary

The main technical terms, and the role names used to explain them, one line each. The specification holds the precise definitions; where the two differ, the specification is right.

TermMeaning
AuthorThe person or agent who writes a Program.
CapabilityWhat a Connector can offer on behalf of its system: the reads and writes it supports, and what the system can compute itself. The basis for pushdown. Not a permission.
CarrierA kind of typed collection that operations work on: Relation, Graph, Vectors or Events. See Beyond tables.
CheckpointA durable stop where protected work waits for a decision, with what is being decided and how the Run continues both fixed. Can also occur during preparation, before a Run starts.
CIDAn identifier computed from normalised content. Distinguishes a Program's meaning, a Plan, an observation and a Preview from one another.
ConnectorThe component that connects ActQL to one external system and declares its sources, capabilities and Effects.
CoreThe part of ActQL that owns meaning: it checks, plans and computes, and runs the Run state machine.
DeciderWhoever answers a delegated policy decision: a person, an automatic rule, an approval application, or a combination.
EffectA declared operation that changes the state of an external system — update, insert, delete, a typed command, a store. Applying it requires the Host's permission.
Effect outcomeWhat became of one operation sent for an Effect: applied, accepted, failed or unknown. Items never sent are unattempted; an item undone after a failure is rolled_back. An Effect with several items records one outcome per item and can end partly applied.
GuardA limit the author adds to a Program on top of those set by the systems and the Host.
HostThe part that runs ActQL in your environment and holds policy, credentials and the Store. Embedded in an application, behind the CLI, on a server; ActQL Cloud is a managed Host.
INFERThe stage that gives each row's model-task inputs to a model and returns a typed answer per row.
InvocationThe semantic input to one unit of computation: its bound values, the definitions it relies on and the observations it has captured. A Program with several stages has one Invocation per stage.
LimitA ceiling on a quantity — rows, bytes, calls, Effects — fixed before the work it applies to starts. Exceeding it fails the work with an error.
Model taskA versioned definition of what to ask a model: the instructions, the input types and the output types. A Program refers to it by name in INFER.
OperatorWhoever runs a Host and sets up its connections, credentials and policy.
PlanThe execution contract the Core builds from a checked Program: operations, types, the boundaries of external work and the limits. Where and how each part physically runs is chosen within that meaning.
PolicyHost configuration that answers, per user, connection, resource and operation, whether a read, model call or Effect is allowed, denied or delegated.
PreviewThe targets and changes an Effect would apply, fixed from recorded results before anything is written. Whether the current values are shown too depends on the Connector.
ProgramOne piece of work written in ActQL: reads, computation, model calls, Results and Effects.
PushdownHanding part of the work to the external system that holds the data, when results, failure behaviour, permissions and the scope of observed data all stay the same.
ReconciliationResolving an unknown outcome — of a Run or of an Effect — from the evidence of the original operation. May remain unresolved; never resends without evidence.
ResultA named output a Program publishes. It is kept and can be read by the caller, or by later work, with the required permission.
RunOne execution of a Program and its durable record: what was read, which answers were received, what was sent and what was decided.
SourceWhere a Program reads from: a Connector's SourceView, or typed data handed to the Program locally.
SourceViewA Connector's versioned contract for one resource: its types, how it is read, and its constraints.
StageA finite unit of work within a Program. A later stage can use the Results and receipts of earlier ones.
StoreWhere the Host keeps what execution and recovery need: Run state, observations, Results, Checkpoints and the Effect ledger.

Source: actql/actql/docs/concepts/glossary.md at d904416674e4 · Edit on GitHub