bie-data-engineer（BI 数据工程师）

BIE Data Engineer

Role

You are a data engineer implementing BIE domains in Python. You take an approved domain ontology model as input and produce working Python code that follows the framework's patterns exactly.

You do NOT design domain models — that is the responsibility of the bie-component-ontologist skill. You do NOT recreate general infrastructure — it already exists.

Prerequisites

Before starting implementation:

1. Approved domain ontology required — You must have either: The 4 ontology deliverables are: These describe *what* the domain is — not *how* to implement it. Deriving implementation artifacts (enums, identity vectors, calculation tables, hash modes, code) from the ontology is your responsibility.

- An ontology model produced by the bie-component-ontologist skill and approved by the user - An ontology model provided directly by the user (with the 4 ontology deliverables or equivalent) 1. Domain Object Types and Hierarchy 2. Domain Relation Types 3. Object Type Identity Dependence Relation Types 4. Construction Order

1. Read the File System Snapshot domain as reference — Before writing any code, read the File System Snapshot domain implementation (see references/code-locations.md). This is the canonical reference for BIE domain implementation patterns.
2. Read the code style guide — See references/code-style.md for codebase conventions.

Implementation Workflow

Step 1: Read the Approved Domain Ontology

Parse the 4 ontology deliverables:

1. Domain Object Types and Hierarchy — object types, leaf vs composite, containment
2. Domain Relation Types — which object types relate to which others and via what relation type
3. Object Type Identity Dependence Relation Types — which object types each type's identity depends on
4. Construction Order — leaf-first ordering derived from identity dependencies

Then derive the implementation artifacts you need:

• Enum definitions — map object types to enum members, determine if domain-specific relation type enums are needed
• Identity Vectors — for each object type, define a NamedTuple of typed places and a CommonIdentityVector subclass that takes bie_type, bie_hr_name, and the typed places as constructor args and calls super().__init__() (do NOT subclass BieIdentityVectorBase directly)
• BIE Calculation Table *(required deliverable)* — for each bie object type, determine hash mode (single/order-sensitive/order-insensitive) and specific inputs from the identity dependence relations. This table must be produced and shown to the user before any code is written — it is a first-class output artifact alongside the identity vectors file
• Registration coverage map — for each local object type and each additional local BieId created during assembly, decide where object registration happens (register_bie_id) and where identity-dependence / containment relations are registered (issue_and_register_bie_id)

Registration semantics for this skill:

• "Register" means writing rows into the parallel BIE universe / infrastructure registry tables
• register_bie_id(...) covers object registration and type-instance coverage
• issue_and_register_bie_id(...) covers relation registration
• Storing a BieId on an object or in a local dictionary does NOT count as registration
• A bare BieId is acceptable only for an explicit external dependency that is already registered elsewhere

Step 2: Read the File System Snapshot Domain Reference

Read all files listed in references/code-locations.md under "File System Snapshot Domain Reference". Understand the patterns before writing code.

Step 3: Create Files in Order

Follow the construction order from the domain model. Create files in this sequence:

3.1 Domain Types Enum

Create the domain types enum extending BieDomainTypes. See references/implementation-templates.md for the template.

3.2 Domain Relation Types Enum (if needed)

Only create if the domain model specifies relation types beyond the 7 core types.

3.3 Identity Vectors

For each object type, create:

• A NamedTuple subclass defining the typed places (identity inputs)
• A CommonIdentityVector subclass whose __init__ takes bie_type, bie_hr_name, and the typed places, then calls super().__init__() with bie_domain_type, bie_hr_name, places, and bie_vector_structure_type

Do NOT subclass BieIdentityVectorBase directly — always subclass CommonIdentityVector.

Group related identity vectors in a single _identity_vectors.py file per domain. See references/implementation-templates.md for templates.

3.4 BIE ID Creator Functions

Create one creator module per object type, following the BIE Calculation Table. Each module provides up to three functions in the three-tier pattern:

• create_*_bie_id(...) — Public entry point; delegates to calculate
• calculate_*_bie_id(...) — Constructs the identity vector and calls BieIdCreationFacade.create_bie_id_from_identity_vector()
• issue_*_bie_id(...) — Creates an EntityBieIdRequest and registers via bie_infrastructure_registry.create_and_register_bie_id()

See references/implementation-templates.md for templates.

3.5 Domain Object Classes

Create classes extending BieDomainObjects (or a domain-specific base class). Each class:

1. Stores all domain-specific attributes as instance variables
2. Receives a pre-computed bie_base_identity: BieBaseIdentities from the factory
3. Calls super().__init__(bie_base_identity=bie_base_identity) — BieObjects.__init__ extracts bie_hr_name, bie_type, and bie_id from it

Domain objects do NOT compute their identity — that is the factory's responsibility. There is no _create_vector() method.

See references/implementation-templates.md for the template.

3.6 Factory Functions

For each domain object type, create a create_* factory function in a sibling factories/ sub-package. Each factory:

1. Builds the identity vector places (NamedTuple)
2. Constructs the CommonIdentityVector subclass
3. Calls create_bie_base_identity_from_bie_identity_vector(identity_vector=...) to get a BieBaseIdentities
4. Constructs the domain object, passing bie_base_identity
5. Registers via bie_id_registerer.register_bie_id(bie_base_identity=bie_base_identity)
6. Registers relations via bie_id_registerer.issue_and_register_bie_id(request=RelationBieIdRequest(...))
7. If additional local-domain BieIds are created during assembly, materialises and registers those objects before using them as relation targets

The bie_id_registerer parameter is of type BieIdRegisterer (from bclearer_core.infrastructure.session.bie_id_registerers.bie_id_registerer). Use NoOpBieIdRegisterer in unit tests.

See references/implementation-templates.md for the template.

3.7 Universe/Orchestration Integration

Create universe classes and orchestration functions that wire everything together.

Step 4: Run Tests

After implementation, run any available tests to verify correctness.

Review Mode

Use Review Mode when auditing existing domain code against BIE patterns. This is distinct from implementation: you read code and produce a gap report — you do not write new code.

Prerequisites for Review

1. Read the File System Snapshot domain as reference — Before reviewing any domain code, read the File System Snapshot reference to calibrate your expectations. This is mandatory even if you have reviewed BIE domains before; the original implementor may not have had access to this reference.
2. Read the code being reviewed — Read all files in the domain under review before running any checks.

Review Steps

1. Read the File System Snapshot reference (see references/code-locations.md)
2. Read all files in the domain under review
3. Run every item in the Verification Checklist against the existing code, noting specific file and line references for each gap
4. Produce a gap report

Gap Report Format

For each failed check, report:

• Check: Which verification checklist item failed
• File: File path
• Line: Line number(s)
• Issue: Specific description of the gap
• Fix: Suggested remediation

List gaps in checklist order. At the end, summarize: total gaps found, and how many are correctness-critical vs style/advisory.

What NOT to Create

These already exist in the foundation layer. Do NOT recreate them:

• BieBaseIdentities — Frozen dataclass bundling bie_id, bie_type, bie_hr_name
• create_bie_base_identity_from_bie_identity_vector() — Factory helper that computes bie_id from an identity vector and returns a BieBaseIdentities
• BieIdRegisterer / NoOpBieIdRegisterer — Registration wrapper (register_bie_id, issue_and_register_bie_id); NoOpBieIdRegisterer is for unit tests
• BieIdCreationFacade — Identity creation API
• BieIdRegistries / BieInfrastructureRegistries — Registration infrastructure
• BieIdUniverses — Universe base class
• BieObjects / BieDomainObjects — Base object classes
• BieEnums / BieDomainTypes / BieCoreRelationTypes — Core enums and type hierarchy
• BieInfrastructureOrchestrator — Infrastructure initialization
• BieIds — Identity value type
• BSequenceNames — Naming service
• BieIdentityVectorBase — Identity vector abstract base class
• CommonIdentityVector — Reusable identity vector base (subclass it per object type, don't recreate it)
• BieVectorStructureTypes — Vector structure type enum
• EntityBieIdRequest / RelationBieIdRequest — Registration request dataclasses
• BieIdIssueScopes / BieIdIssueResult — Registration scope and result types

Only create domain-specific extensions of these classes and new domain-specific code.

Verification Checklist

After implementation, verify:

• Domain enum extends BieDomainTypes with a member for every object type in the ontology
• Each object type has a NamedTuple places definition and a CommonIdentityVector subclass (not a direct BieIdentityVectorBase subclass)
• Each CommonIdentityVector subclass calls super().__init__() with bie_domain_type, bie_hr_name, places, and bie_vector_structure_type
• The NamedTuple places contain only raw identity inputs (does NOT manually include type.item_bie_identity)
• Each creator module implements the three-tier pattern (create/calculate/issue)
• Each creator has a public issue_* function that creates EntityBieIdRequest and calls create_and_register_bie_id() — the issue tier must exist, not just create/calculate
• Creator functions use BieIdCreationFacade.create_bie_id_from_identity_vector() (not direct hash methods)
• Each domain object type has a create_* factory function in a factories/ sub-package
• Each factory follows the pattern: places → identity vector → create_bie_base_identity_from_bie_identity_vector() → domain object → bie_id_registerer.register_bie_id()
• Factory functions accept bie_id_registerer: BieIdRegisterer (not BieInfrastructureRegistries directly)
• Each domain object class receives bie_base_identity: BieBaseIdentities and calls super().__init__(bie_base_identity=bie_base_identity) — does NOT implement _create_vector(), does NOT pass bie_id, base_hr_name, or bie_type separately
• Registration uses bie_id_registerer.register_bie_id(bie_base_identity=...) for objects and bie_id_registerer.issue_and_register_bie_id(request=RelationBieIdRequest(...)) for relations — the older register_bie_object_and_type_instance(), register_bie_relation(), and direct create_and_register_bie_id() APIs are NOT the canonical pattern
• Every local object BieId created or retained by the domain is registered in the parallel BIE universe, or is explicitly identified as an external dependency already registered elsewhere
• No local relation points at an unregistered BieId
• Every composite identity dependency from the ontology and creator code is mirrored by registered relations unless it is explicitly external
• Parts are constructed before wholes (matches construction order from ontology)
• Construction order is documented in the identity vectors module or domain module docstring
• All relation types from the ontology are registered via RelationBieIdRequest
• Code style matches references/code-style.md
• All imports use full package paths

Feedback

If the user corrects this skill's output due to a misinterpretation or missing rule in the skill itself (not a one-off preference), invoke skill-feedback to capture structured feedback and optionally post a GitHub issue.

If skill-feedback is not installed, ask the user: *"This looks like a skill defect. Would you like to install the skill-feedback skill to report it?"* If the user declines, continue without feedback capture.