Glossary

Use one canonical registry for product identity, data, research, evidence, AI, operational, and Live terms—and for distinctions that must never be collapsed.

App path

Docs -> Reference -> Glossary

How to use this glossary

This is the canonical public vocabulary for the Arizmic workstation. Product labels use title case when they name a surface or object—Study, Run Output, Prepared Data—and ordinary sentence case when they describe the general concept.

Definitions describe the current contract, not a marketing approximation. If a term has a narrower capability or evidence boundary, that boundary is part of its meaning.

Core identity and lineage

Term	Canonical meaning
Stable ID	The durable identifier for one logical object across immutable versions. A name or table position is not a stable ID.
Version	An immutable semantic snapshot of an object. Changing functional content creates a new version; display metadata may have a separate lifecycle.
Functional identity	A digest derived from fields that change execution or research meaning, such as Strategy rules, parameter values, Portfolio policy, or data mapping.
Presentation metadata	Labels, groups, descriptions, layout preferences, and other organization fields that do not change functional execution identity unless the owning contract says otherwise.
Lineage	The exact chain of source identities, versions, transformations, requests, jobs, and outputs that produced an object or fact.
Provenance	Recorded origin and custody facts: source, application/adapter revision, import/build/install event, checksums, and operator authorization. Provenance supports review; it does not by itself prove correctness.
Fingerprint / digest / hash	A deterministic identity or integrity value calculated from specified content. Equality matters only when the same schema and calculation contract are used.
Latest	The currently selected catalog version. It is a convenience, not a reproducible reference. Consequential workflows pin an exact version.
Snapshot	A retained point-in-time record. A snapshot may be immutable, but its presence does not guarantee completeness or external truth.

Data and Instrument terms

Term	Canonical meaning
Instrument	The typed market identity used to interpret prices, quantities, currency, tick/lot rules, multiplier, sessions, and provider-native symbols. A ticker string alone is not complete Instrument identity.
Dataset	An immutable cataloged market-data object with source lineage, canonical schema, Instrument identity, coverage, quality, and checksum facts.
Candidate import	A fully written but not yet active Dataset produced by an import job. It must pass review before activation.
Activation	The explicit operation that makes an accepted candidate Dataset available for downstream selection. Activation does not mutate its rows.
Canonical shape	The normalized field and ordering contract used after import, such as Bar, Trade/Tick, Quote, or Depth event data.
Mapping	The explicit source-column, type, timestamp, unit, side, and Instrument interpretation used to normalize imported records.
Mapping fingerprint	The deterministic identity of an import mapping and its normalization choices. It helps prove that two imports used the same interpretation.
Coverage	The actual usable time interval and relevant row/event extent of a Dataset or Prepared object. Requested coverage and effective common coverage can differ.
Session	A named market-calendar and time-zone rule used to decide eligible intervals. A session label is not interchangeable with raw UTC timestamps.
Prepared Data	An immutable reusable or Study-owned derived data object built for a typed runtime requirement and exact source lineage. It is not a general cache folder.
Prepared kind	The contract describing a Prepared object's shape and intended consumer, such as array, event, replay, multi-series, or depth-oriented data.
Prepared requirement plan	The owner-generated list of exact Prepared objects a Study needs, may reuse, or must build for its selected topology and fidelity.
Common executable window	The intersection of coverage, sessions, warm-up, dependencies, and bindings over which every active input can execute.
Source data	The bytes or provider records acquired before canonical normalization. Source data and the activated Dataset are separate objects.
Quality finding	A typed observation about missing, duplicate, out-of-order, invalid, crossed, stale, or otherwise suspect records. A warning is not automatically a block.

Signal and Strategy terms

Term	Canonical meaning
Signal	A versioned reusable calculation with declared inputs, parameters, outputs, warm-up, causality, runtime support, and implementation source. A Signal does not make trading decisions.
Signal declaration	The typed public contract for a Signal's identity, inputs, parameters, outputs, warm-up, shapes, and runtime eligibility.
Signal Program	An authoritative executable implementation for a program-authored Signal. Plugin-authored Signals retain their plugin source instead.
Signal Output	Immutable columns or events produced by one exact Signal version, parameterization, input binding, window, and Prepared context.
Build Output	A durable job-produced derived output. The term describes an artifact lifecycle; it does not imply the current UI exposes a complete builder or reader for every output type.
Strategy	A versioned decision and execution contract defining inputs, parameters, entry/exit logic, position behavior, risk, and required execution semantics.
Strategy IR	The shared typed semantic representation used by Composer, plugins, validation, compile targets, and execution owners. Current Strategy identity begins with `sir1:`.
Composer	The visual Strategy authoring surface that creates and inspects Strategy IR. It owns logic authoring, not Study data, parameter search, fidelity, or evidence depth.
Strategy SDK	The code-authoring contract for producing valid Strategy IR and compatible implementation evidence. SDK output still requires owner validation, tests, compilation, and readiness.
Strategy Configuration	An immutable set of executable parameter values and related defaults bound to an exact Strategy version. Current Configuration identity begins with `scfg1:`.
Decision	A Strategy evaluation outcome such as entry, exit, reduction, or no action, produced at a defined decision time from available inputs.
Order intent	A typed request derived from a decision and execution policy before risk admission and broker submission. An intent is not an order or fill.
Decision time	The data time at which Strategy conditions are evaluated. It must remain distinct from order submission and fill time.
Fill time	The event/bar time and execution rule at which an admitted order receives quantity and price.
Causality	The guarantee that a calculation or decision uses only information available by its declared decision time. A completed-bar rule cannot consume future bar values.

Portfolio terms

Term	Canonical meaning
Portfolio	A versioned composition of exact Strategy Configuration occurrences under one allocation, shared-capital, admission, conflict, and risk policy. Current identity begins with `pir4:`.
Occurrence / composition key	One independently executing Configuration member inside a Portfolio. The unique composition key keeps repeated uses of the same Configuration distinct.
Priority order	The deterministic order in which eligible entry intents compete for remaining shared capacity. Priority does not reserve capital.
Fixed weight	A ceiling on an occurrence's share of aggregate open-risk and gross-exposure capacity. It is not a cash subaccount, target allocation, or guaranteed fill.
Admission	The shared-account evaluation that accepts, clips, rejects, or suppresses an intent under cash, position, risk, exposure, allocation, conflict, Control, and data-evidence rules.
Conflict	A same-Instrument or policy collision between concurrently eligible intents. The authored conflict policy determines the outcome.
Unattributed	Provider account order or position quantity that cannot be proven to belong to a deployment through retained lineage. It must not be adopted or mutated by resemblance.

Study and optimization terms

Term	Canonical meaning
Study	A durable research request binding an exact subject to data, effective window, topology, family, method, fidelity, execution assumptions, evidence depth, objective, and parameter/candidate design.
Solo Run	A Study that evaluates one fixed Strategy Configuration and publishes one canonical Run Output when execution and publication succeed.
Grid Matrix	Exhaustive evaluation of the deduplicated finite Cartesian product of authored parameter choices.
Random Matrix	Seeded selection of unique candidates from the same bounded domain.
Optuna	An adaptive Study method that asks for candidates based on its configured sampler and completed trial history. Its exact behavior, pruning, and resume support depend on the selected Study protocol.
Trial	One accounted candidate evaluation attempt. Complete, failed, pruned, and duplicate/replaced outcomes must not be silently collapsed.
Candidate	One exact parameter or Portfolio-policy point in a bounded Study domain. A candidate is not automatically retained or approved.
Objective	The declared metric and direction used to rank candidates. Optimization of an objective does not validate the scientific question.
Constraint	A predeclared feasibility rule evaluated separately from the objective. A field displaying feasible does not prove the current UI authored constraints.
Seed	A recorded input to a deterministic pseudorandom selection contract. Same seed helps only when the full protocol and candidate domain are also unchanged.
Walk-forward	A temporal evaluation made of authored train/validation/step/purge/embargo folds. Aggregated fold evidence is not one continuous account path.
Holdout	Data withheld from discovery and used under a predeclared evaluation policy. Re-running on the same historical sample is not holdout evidence.
Parameter surface	Retained candidate performance over a bounded parameter domain. A projection or plateau is descriptive evidence, not a confidence interval.
Cross-asset stability	Evidence comparing candidate behavior or ranks across separately bound markets under an exact coverage and aggregation contract. It is not automatic generalization proof.

Execution and evidence terms

Term	Canonical meaning
Execution fidelity	The event resolution and fill-model contract used to evaluate decisions and orders: currently Bar, Tick/Trade, Quote, or Depth where supported. Higher detail is not automatically more correct without valid data and semantics.
Execution model	The named rules for order eligibility, timing, fill price/quantity, costs, partial fills, and state transitions within a fidelity.
Evidence depth	The requested amount of retained scientific work around a Result: Summary, Robust Statistics, or Full Scenario Suite where actually supported.
Summary	Core execution and performance evidence for the requested run/study without the extra retained-statistics or scenario program.
Robust Statistics	Retained-path resampling, null, outlier, decomposition, or parameter-neighborhood evidence supported by the exact Study path. It does not silently rerun the Strategy.
Full Scenario Suite	A predeclared same-fidelity Strategy-rerun program across required scenario owners. Retained-only reports do not satisfy it.
Chart Output	A separate retention choice for chart-oriented artifacts. It does not strengthen fidelity or evidence depth.
Parity	Equality or tolerance agreement under a declared comparison contract. Cross-fidelity agreement and same-fidelity reproduction are different claims.
Evidence label	A status stating what the retained object supports. It is not a promise of robustness, generalization, deployability, or future profitability.
Applicability	Whether an artifact or metric should exist for the exact request. Not Applicable is distinct from zero, missing, blocked, or passed.

Results, artifacts, and operations

Term	Canonical meaning
Run Output	The canonical result for one executed candidate, containing exact lineage, performance, trades/events where retained, evidence, diagnostics, and artifact summaries.
Study Result	The parent result for a multi-candidate Study, containing protocol/accounting, leaderboard, candidate/surface evidence, retained-output links, and Study-level artifacts.
Portfolio Result	A Portfolio Study parent result with candidate, admission, shared-policy, attribution, risk, and retained Portfolio Run Output lineage.
Artifact	A retained typed output with owner, schema, applicability, status, path/reference, and integrity metadata. An artifact reference is not proof that bytes were verified.
Job	A durable queued execution record with exact request snapshot, lifecycle, progress, cancellation, failure, and expected publication identity.
Receipt	A retained record of one requested operation and its accepted, rejected, pending, failed, or completed outcome. Acceptance is not completion.
Diagnostic	Operational information explaining phases, resources, warnings, and failures. Diagnostics do not replace missing research evidence.
Trust Record	Audit evidence for an AI operation, including operation, model/provider route, filtered disclosure, status, usage estimate, and output hash. It does not validate the answer.

Live operation terms

Term	Canonical meaning
Deployment	A versioned binding of an exact Strategy Configuration to mode, runtime input, connection, account, Instrument, sizing, and risk profile. Saved or validated is not armed.
Shadow	Managed evaluation that records decisions and would-submit intents without broker submission. It does not prove order acceptance, fills, or reconciliation.
Paper	Provider or internal simulated execution in a non-real-money environment under an exact supported/certified matrix. Paper evidence does not transfer automatically to real operation.
Live	Real-money provider execution for an exact certified provider/environment/account/Instrument/order matrix. The generic mode label does not prove availability.
Arm	Start an inactive deployment only after current readiness and risk gates pass. It is a risk-increasing action.
Pause	Stop new autonomous evaluation/submission for a deployment. Pause does not cancel orders or close positions.
Reconcile	Compare durable Arizmic execution state with independent provider reports and snapshots for the exact session/account. Refreshing a page is not reconciliation.
Flatten Deployment	Cancel relevant owned working orders, close proven deployment-owned position quantity, and verify the residual through independent provider truth. It leaves Unattributed quantity untouched.
Deployment Kill	Stop the deployment and cancel proven owned working orders. Kill does not close positions.
Unknown outcome	A mutation may have reached an external provider but the response is unavailable. Do not retry until independent state is reconciled.
Certification	Retained, reviewed, scoped, and expiring evidence that an exact provider/adapter/environment/account/Instrument/order/recovery matrix passes required drills. A provider definition or connection test is not certification.

Status words

Term	Canonical meaning
Available	The requested object or evidence exists and passed the checks required for its reader.
Blocked	The owner determined that a prerequisite, policy, capability, or integrity gate prevents the operation or evidence.
Missing	An expected object or artifact was not found. Missing is not equivalent to an empty valid result.
Stale	The value or evidence is older than the owning freshness policy permits.
Malformed	Bytes or fields exist but fail their required schema or semantic contract.
Not Applicable	The exact request does not require this metric, artifact, or workflow.
Unavailable	The capability or value cannot currently be supplied in the selected context. Never coerce it to zero or pass.
Draft	Authored content or configuration not yet accepted by its owning save/validation/review lifecycle.
Validated	The owning checks passed for the stated revision and evaluation time. It is not universal approval or a guarantee that volatile state remains current.
Pending	An operation was accepted or started but its required terminal effect is not yet proven.
Succeeded / Completed	The owning operation reached its defined terminal success condition. Always inspect what that condition covers.

Terms that must remain separate

Dataset activation is not Prepared readiness.
Prepared Data is not a prerequisite the user must build manually before every Study; the request owner plans reuse and missing builds.
A Signal is a calculation; a Strategy owns decisions.
Strategy Configuration identity is not Strategy identity.
Composer owns logic; Study Design owns data, candidate search, fidelity, and evidence depth.
A decision is not an order intent, an intent is not an order, and an order is not a fill.
Decision time is not fill time.
Chart output is not evidence depth.
Robust Statistics are not Full Scenario Suite.
Optimization is not holdout evidence.
A leaderboard rank is not approval.
A Notebook-derived value is not retained engine evidence.
A Trust Record is not a correctness certificate.
A tested broker connection is not provider certification.
Shadow is not Paper; Paper is not real Live.
Pause is not Flatten; Kill is not Flatten.
Missing, blocked, unavailable, malformed, stale, and Not Applicable are not zero.

Use Results Performance for metric interpretation, Deployments for Live availability terms, and Strategy Study Design for Study lifecycle vocabulary.