Entities (Workbench Meta)

Namespace: https://ecosystem.ai/ontology/entities#

These classes map to Mongo collections (and embedded deployment config) managed in Workbench2. The Project is the hub; other entities attach via asset arrays and foreign keys.

Project hub

Class	Pref label	Definition
`Project`	Project	Hub that groups predictors, frames, dynamic engagements, offer matrices, campaigns, agents, and deployment steps

Key properties: projectId, projectDescription, purpose, projectType (a label, not the classifier), projectOwner, engagementKind (dynamic | static | hybrid)

engagementKind is derived from hasDynamicEngagement, hasPredictor, hasFeatureStoreFrame, and each deployment step’s scoringMode. A stored projectType such as recommender or algorithm does not decide the kind. dynamic_recommender is the review canonical label for a dynamic project. It is still not the classifier.

Dynamic, static, and hybrid

Example project	Stored `projectType`	`engagementKind`	Why
`outreach_message_recommender`	`dynamic`	dynamic	Dynamic engagement links, no predictors, no feature-store frames. The label agrees.
`dynamic_recommender_convergence`	`recommender`	dynamic	Six dynamic engagements and no static models. The label does not agree.
`sample_credit_hybrid`	`solution_template`	hybrid	Static steps and predictors together with dynamic engagement steps.

Spend Personality and Money Personality are lookup-only. They are not Dynamic Engagement approaches and they are not learningAlgorithm values. PostScoreSpendingPersonality (including generated PostScoreSpendingPersonality_* copies) and PostScoreMoneyPersonality copy precomputed scores from whatever database and collection the deployment lookup points at. Any store is valid when the document has the fields that post-score reads.

On spend_personality, the spend_personality and money_personality deployments are lookup-only, while dynamic_recommender_behavior is dynamic engagement. options.is_multi_armed_bandit does not turn a personality plugin into a dynamic step. The collection names on one installation are examples, not required names.

Spend lookup fields: personality, personality_score, and when present trait and trait_score. Optional dimensions include Experiential, Extrovert, Enthusiastic, Industrious, Intentional, and Introvert.

Money lookup fields: money_personality or personality, money_trait or trait, a matching score (money_personality_score or personality_score), and dimension fields such as conservative, carefree, adventurous, borrower, cautious, indulger, prudent, and adventurer.

Static models

Class	Collection / role	Definition
`Frame`	`frames`	Training or reference H2O / data frame
`Predictor`	`predictions`	Prediction case (AutoML / model parms); uses a Frame

Key properties: frameId; predictId, modelId; link usesFrame

Dynamic engagement

Class	Definition
`DynamicEngagement`	Online learning / bandit config (`dynamic_engagement`)
`FeatureStoreRef`	DB + collection for customer/context lookup
`OptionsStoreRef`	Options / arms store (alpha, beta, rewards)

Key properties: engagementName, learningAlgorithm, behavioralType

learningAlgorithm is randomisation.approach and must be one of the Dynamic Engagement dropdown values: epsilonGreedy, binaryThompson, naiveBayes, QLearning, behaviorAlgos, Network.

behavioralType is randomisation.sub_approach. Set it only when learningAlgorithm is behaviorAlgos: lossAversion, riskAversion, prospectTheory, sentimentalEquilibrium, coverageAwareThompson, longTailBoostMF, generative. generative is a behavioral type, not a top-level approach.

Links: hasFeatureStore, hasOptionsStore

Offer catalog

Class	Definition
`OfferMatrix`	First-class offer catalog (`offer_matrix`) with spec, version, status
`OfferMatrixEntry`	Single offer row — fixed six fields for runtime scoring

See Offer Matrix for the canonical row JSON.

Campaigns & contacts

Class	Definition
`Campaign`	Binds patron sources, offer matrices, recommender deployments
`ContactCatalog`	Catalog of contact sources (`contacts_catalog`)
`ContactRecord`	Individual contact / patron document
`ChannelProfile`	Delivery channel for ecogentic / engagement
`AgentConfig`	Ecogentic agent (`agent_configs`)

Supporting assets

Class	Typical collection
`DataConnection`	`data_connections`
`Prompt`	`prompt_library`
`FactInjectionConfig`	`chat_config`
`Calendar`	`calendar`
`Simulation`	`simulation`
`ResponderConfig`	`responder_config`
`Dashboard`	analytics dashboards
`EnrichmentPipeline`	enrichment pipelines
`DeploymentStep`	Embedded on Project — pushed to runtime. `scoringMode` is `dynamic`, `static`, or `lookup`. `stepType` is the stored label and does not override `scoringMode`.
`SpendPersonalityLookup`	Lookup document copied by `PostScoreSpendingPersonality`. Collection is per installation.
`MoneyPersonalityLookup`	Lookup document copied by `PostScoreMoneyPersonality`. Collection is per installation.

Project asset links (object properties)

Property	From → To	Workbench field
`hasPredictor`	Project → Predictor	`project_predictors`
`hasFrame`	Project → Frame	`project_models`
`hasFeatureStoreFrame`	Project → Frame	`feature_stores` (static H2O frames; leave empty on a dynamic project)
`hasDynamicEngagement`	Project → DynamicEngagement	`project_dynamic_interactions`
`hasOfferMatrix`	Project → OfferMatrix	`project_offer_matrices`
`hasCampaign`	Project → Campaign	`project_campaigns`
`hasAgent`	Project → AgentConfig	`project_agents`
`hasDeploymentStep`	Project → DeploymentStep	`deployment_step`
`hasPrompt`	Project → Prompt	`project_prompts`
`hasCalendar`	Project → Calendar	`project_calendars`
`hasSimulation`	Project → Simulation	`project_simulations`
`hasResponderConfig`	Project → ResponderConfig	`project_api_configs`
`hasFactInjectionConfig`	Project → FactInjectionConfig	`project_fact_injection_configs`
`hasDataConnection`	Project → DataConnection	`presto_connections`
`linkedToProject`	ContactCatalog → Project	`project_ids`
`usesOfferMatrix`	Campaign → OfferMatrix	campaign actions
`belongsToCatalog`	ContactRecord → ContactCatalog	—

Full tables (every class and property): Entities Catalog · Concept Index