graflo.architecture.evolution

Manifest evolution: apply high-level schema + ingestion transforms to :class:~graflo.architecture.contract.manifest.GraphManifest.

Use :func:~graflo.migrate.io.manifest_hash to compare contract identity before and after.

Modules:

Name	Description
alignment	Derived identity: lower a merged class's identity branches to fundamental ops.
apply	Apply manifest evolution operations to a copy of a :class:~graflo.architecture.contract.manifest.GraphManifest.
autogenerate	Derive a contract change set from two manifests.
canonical	Canonical vocabulary maps, and the shared pieces of merge resolution.
canonicalize	Canonical form: the payload a content hash is taken over.
codec	(De)serialization for contract operations.
commit
db_profile	Update :class:~graflo.architecture.schema.database_features.DatabaseProfile after vertex changes.
equivalence	The resolved shape of a merge's groups.
hashing	Content hashes over manifest blocks.
history	The commit DAG, and commits on disk.
identity	Identity-plane evolution ops.
ingestion	Ingestion-side evolution ops: mutations whose primary effect is the pipeline.
inverse	Inverses for the subset of contract operations that have one.
inverse_edges	Realize declared inverses as explicit logical edges (add_inverse_edges).
inverse_plan	Planning what to do about declared inverses: realize, repair, switch, withdraw.
merge	Binary merge of two :class:~graflo.architecture.contract.manifest.GraphManifests.
merge3	Three-way merge over manifest change sets, and tracked re-merges.
merge_commit	Recording a merge as a commit.
merge_core	Pure merge helpers for logical vertices and edges.
merge_types	Merged property types: declared on a merge, refused when they disagree undeclared.
naming_graph	One naming graph over the original names of both sides of a merge.
ops	Typed manifest evolution operations.
physical	Typing and physical-profile evolution ops.
preview	What a merge would do, and every way it could refuse — without refusing.
project	Pure planning for :class:~graflo.architecture.evolution.ops.ProjectManifestOp.
rewrite	Structured rewrite of vertex names in pipeline dicts and related resource fields.
sanitize	Physical naming for a target database flavor.
semantics	Applying grounding changes to a manifest.
state_core	state-core: lifting a manifest into one that tracks state and measurements.
structure	Structure-plane evolution ops: introducing and retargeting graph entities.
version	Semantic version bump helpers for manifest evolution.

Attributes

INGESTION_REWRITING_OPS = frozenset({'add_inverse_edges', 'add_resource_transforms', 'add_resources', 'canonicalize', 'remove_resources', 'replace_resources', 'ensure_extracted_fields', 'merge_edges', 'merge_vertices', 'project_manifest', 'remove_edge_properties', 'remove_edges', 'remove_vertex_properties', 'remove_vertices', 'rename_edge_properties', 'rename_relations', 'rename_resources', 'rename_vertex_properties', 'rename_vertices', 'replace_identity', 'retarget_edges', 'sanitize', 'set_inverse_emission'}) module-attribute

IdentityBranchDecl = RawBranch | DerivedBranch | LocalKeyBranch module-attribute

IdentityTarget = Annotated[NaturalIdentityTarget | HashIdentityTarget | FunnelIdentityTarget | AssignedIdentityTarget | BlankIdentityTarget, PydanticField(discriminator='mode')] module-attribute

ManifestOp = Annotated[RemoveVerticesOp | AddResourceTransformsOp | EnsureExtractedFieldsOp | AddResourcesOp | RemoveResourcesOp | ReplaceResourcesOp | AddVerticesOp | AddEdgesOp | RetargetEdgesOp | AddSecondaryIdentitiesOp | RemoveSecondaryIdentitiesOp | ReplaceEdgeIdentitiesOp | ChangeFieldTypesOp | AddVertexIndexesOp | RemoveVertexIndexesOp | AddEdgeIndexesOp | RemoveEdgeIndexesOp | SetEdgeDirectedOp | SetBindingsOp | SetDbProfileOp | SetVertexSemanticsOp | SetVertexDescriptionsOp | SetEdgeSemanticsOp | SetFieldSemanticsOp | MergeVerticesOp | CanonicalizeOp | RenameVertexPropertiesOp | RemoveVertexPropertiesOp | AddVertexPropertiesOp | RenameVerticesOp | RenameRelationsOp | RenameResourcesOp | RemoveEdgesOp | MergeEdgesOp | RenameEdgePropertiesOp | RemoveEdgePropertiesOp | AddEdgePropertiesOp | DeclareEdgeInversesOp | RetractEdgeInversesOp | AddInverseEdgesOp | SetNativeInversesOp | SetInverseEmissionOp | ProjectManifestOp | ReplaceIdentityOp | SanitizeOp | MergeManifestsOp, PydanticField(discriminator='op')] module-attribute

__all__ = ['CANON_VERSION', 'COMMIT_KINDS', 'INGESTION_REWRITING_OPS', 'IRREVERSIBLE', 'LIST_ORDER', 'AddEdgeIndexesOp', 'AddEdgePropertiesOp', 'AddEdgesOp', 'AddInverseEdgesOp', 'AddResourceTransformsOp', 'AddResourcesOp', 'AddSecondaryIdentitiesOp', 'AddVertexIndexesOp', 'AddVertexPropertiesOp', 'AddVerticesOp', 'AlignmentConflictError', 'AssignedIdentityTarget', 'BlankIdentityTarget', 'CanonicalMap', 'CanonicalizeOp', 'ChangeFieldTypesOp', 'Cluster', 'ClusterIndex', 'Commit', 'CommitError', 'Completion', 'ConflictResolution', 'DanglingEntry', 'DeclareEdgeInversesOp', 'DeclaredMaps', 'DerivationSpec', 'DerivedBranch', 'EdgeFieldSemanticsTarget', 'EdgeIdentitiesEntry', 'EdgeIndexEntry', 'EdgeRetargetEntry', 'EdgeSelector', 'EnsureExtractedFields', 'EnsureExtractedFieldsOp', 'FieldSemanticsTarget', 'FieldTypeSpec', 'FileCommitStore', 'FunnelIdentityTarget', 'HashIdentityTarget', 'History', 'IdentityBranchDecl', 'IdentityPlan', 'IdentityReplacement', 'IdentityTarget', 'InversePlan', 'ListOrder', 'LocalKeyBranch', 'LocalKeySource', 'ManifestOp', 'MergeCanonicalConflictError', 'MergeConflict', 'MergeEdgesOp', 'MergeError', 'MergeFieldTypes', 'MergeIdentityError', 'MergeIncompleteError', 'MergeManifestsOp', 'MergeNameConflictError', 'MergeNamingError', 'MergeNamingIncompleteError', 'MergeRecipe', 'MergeRecipeRef', 'MergeRenames', 'MergeResult', 'MergeVerticesOp', 'NamingFinding', 'NamingGraph', 'NamingResult', 'NaturalIdentityTarget', 'ProjectManifestOp', 'PropertyEquivalence', 'RelationCluster', 'RelationEquivalence', 'RemoveEdgeIndexesOp', 'RemoveEdgePropertiesOp', 'RemoveEdgesOp', 'RemoveResourcesOp', 'RemoveSecondaryIdentitiesOp', 'RemoveVertexIndexesOp', 'RemoveVertexPropertiesOp', 'RemoveVerticesOp', 'RenameEdgePropertiesOp', 'RenameHints', 'RenameRelationsOp', 'RenameResourcesOp', 'RenameVertexPropertiesOp', 'RenameVerticesOp', 'ReplaceEdgeIdentitiesOp', 'ReplaceIdentityOp', 'ReplaceResourcesOp', 'RetargetEdgesOp', 'RetractEdgeInversesOp', 'RevisionOp', 'SanitizeOp', 'SetEdgeDirectedOp', 'SetEdgeSemanticsOp', 'SetFieldSemanticsOp', 'SetInverseEmissionOp', 'SetNativeInversesOp', 'SetVertexDescriptionsOp', 'SetVertexSemanticsOp', 'SideMaps', 'SideRenames', 'UnclassifiedListField', 'VertexEquivalence', 'apply_add_edge_indexes', 'apply_add_edge_properties', 'apply_add_edges', 'apply_add_inverse_edges', 'apply_add_resource_transforms', 'apply_add_resources', 'apply_add_secondary_identities', 'apply_add_vertex_indexes', 'apply_add_vertex_properties', 'apply_add_vertices', 'apply_canonicalize', 'apply_change_field_types', 'apply_declare_edge_inverses', 'apply_evolution', 'apply_merge_edges', 'apply_merge_vertices', 'apply_project_manifest', 'apply_remove_edge_ids', 'apply_remove_edge_indexes', 'apply_remove_edge_properties', 'apply_remove_edges', 'apply_remove_resources', 'apply_remove_secondary_identities', 'apply_remove_vertex_indexes', 'apply_remove_vertex_properties', 'apply_remove_vertices', 'apply_rename_edge_properties', 'apply_rename_relations', 'apply_rename_resources', 'apply_rename_vertex_properties', 'apply_rename_vertices', 'apply_replace_edge_identities', 'apply_replace_identity', 'apply_replace_resources', 'apply_retarget_edges', 'apply_retract_edge_inverses', 'apply_sanitize', 'apply_set_edge_directed', 'apply_set_edge_semantics', 'apply_set_field_semantics', 'apply_set_inverse_emission', 'apply_set_native_inverses', 'apply_set_vertex_descriptions', 'apply_set_vertex_semantics', 'build_commit', 'build_merge_commit', 'build_merge_recipe', 'build_multi_parent_commit', 'build_naming', 'build_recipe', 'build_revert_commit', 'build_root_commit', 'canonical_map_to_ops', 'canonical_near_collisions', 'canonical_payload', 'canonicalize_ops', 'checkout', 'checkout_parent', 'compose_canonical_maps', 'compute_commit_id', 'compute_root_commit_id', 'dangling_entries', 'describe_slot', 'diff_manifests', 'diff_manifests_verified', 'find_commit_by_tree', 'find_merge_base', 'fold_declared_maps', 'full_hash', 'graph_hash', 'identity_to_ops', 'ingestion_hash', 'invert_op', 'invert_ops', 'irreversible_reason', 'is_reversible', 'manifest_hash', 'merge_manifests', 'merge_three_way', 'naming_table', 'op_from_dict', 'op_slots', 'op_to_dict', 'ops_from_dicts', 'ops_from_yaml', 'ops_reaching_ingestion', 'ops_to_dicts', 'ops_to_yaml_str', 'plan_declare_symmetric', 'plan_realize_inverses', 'plan_repair_inverses', 'plan_switch_realization', 'plan_withdraw_realization', 're_merge', 'rehash_trees', 'schema_hash', 'stable_hash', 'subject', 'suggest_merge_op', 'take_left', 'take_right', 'trim_canonical_map', 'validate_and_complete_canonical_map', 'validate_identity', 'verify_history'] module-attribute

Classes

AddEdgeIndexesOp

Bases: ConfigBaseModel

Author secondary indexes on edge physical specs.

Source code in graflo/architecture/evolution/ops.py

class AddEdgeIndexesOp(ConfigBaseModel):
    """Author secondary indexes on edge physical specs."""

    op: Literal["add_edge_indexes"] = "add_edge_indexes"
    edges: list[EdgeIndexEntry] = PydanticField(
        ...,
        description="Per-spec indexes to add (``indexes`` on each entry).",
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_entries(self) -> AddEdgeIndexesOp:
        _validate_edge_index_entries(
            self.edges, kind="add_edge_indexes", carries="indexes"
        )
        return self

Attributes

edges = PydanticField(..., description='Per-spec indexes to add (``indexes`` on each entry).', min_length=1) class-attribute instance-attribute

op = 'add_edge_indexes' class-attribute instance-attribute

AddEdgePropertiesOp

Bases: ConfigBaseModel

Add edge properties for each relation in schema/profile defaults.

An entry may be a bare name (an untyped property) or a full :class:~graflo.architecture.schema.vertex.Field, as on :class:AddVertexPropertiesOp, so a typed or grounded edge property is one replayable step rather than an add followed by a type change.

Source code in graflo/architecture/evolution/ops.py

class AddEdgePropertiesOp(ConfigBaseModel):
    """Add edge properties for each relation in schema/profile defaults.

    An entry may be a bare name (an untyped property) or a full
    :class:`~graflo.architecture.schema.vertex.Field`, as on
    :class:`AddVertexPropertiesOp`, so a typed or grounded edge property is
    one replayable step rather than an add followed by a type change.
    """

    op: Literal["add_edge_properties"] = "add_edge_properties"
    additions: dict[str, list[str | Field]] = PydanticField(
        ...,
        description=(
            "Per-relation edge property additions: "
            "``{relation_name: [field_name | Field, ...]}``."
        ),
        min_length=1,
    )

    def field_names(self, relation: str) -> list[str]:
        """The names added to *relation*, whichever shape they were written in."""
        return [
            entry if isinstance(entry, str) else entry.name
            for entry in self.additions.get(relation, [])
        ]

Attributes

additions = PydanticField(..., description='Per-relation edge property additions: ``{relation_name: [field_name | Field, ...]}``.', min_length=1) class-attribute instance-attribute

op = 'add_edge_properties' class-attribute instance-attribute

Methods:

field_names(relation)

The names added to relation, whichever shape they were written in.

Source code in graflo/architecture/evolution/ops.py

def field_names(self, relation: str) -> list[str]:
    """The names added to *relation*, whichever shape they were written in."""
    return [
        entry if isinstance(entry, str) else entry.name
        for entry in self.additions.get(relation, [])
    ]

AddEdgesOp

Bases: ConfigBaseModel

Introduce new logical edge relations between existing vertex types.

Source code in graflo/architecture/evolution/ops.py

class AddEdgesOp(ConfigBaseModel):
    """Introduce new logical edge relations between existing vertex types."""

    op: Literal["add_edges"] = "add_edges"
    edges: list[Edge] = PydanticField(
        ...,
        description="Full edge definitions, in the shape the schema block accepts.",
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_unique_ids(self) -> AddEdgesOp:
        edge_ids = [(edge.source, edge.target, edge.relation) for edge in self.edges]
        if len(edge_ids) != len(set(edge_ids)):
            raise ValueError(
                "add_edges entries must be unique by (source, target, relation)"
            )
        return self

Attributes

edges = PydanticField(..., description='Full edge definitions, in the shape the schema block accepts.', min_length=1) class-attribute instance-attribute

op = 'add_edges' class-attribute instance-attribute

AddInverseEdgesOp

Bases: ConfigBaseModel

Realize declared inverse pairs as explicit logical edges (portable to every backend).

Edges are derived from the relation map: for each directed edge (S, T, r) whose relation has a declared pair inv, adds the logical edge (T, S, inv) unless it exists and its physical spec, and sets emit_inverse on the edge steps that write r, so the same rows write both. No step is generated: the inverse is mirrored at assembly, after the relation is resolved, which covers every way a step can name its relation. A resource that already writes the inverse with a step of its own is left alone. The pair must be declared first (:class:DeclareEdgeInversesOp); this op never declares. A symmetric relation has no inverse edge to add -- its edges are undirected. Refused for a relation whose inverse is native (:class:SetNativeInversesOp), since both would store the same fact.

Withdraw with :class:RemoveEdgesOp on the inverse edges: removing an edge clears the emit_inverse flags that fed it.

Source code in graflo/architecture/evolution/ops.py

class AddInverseEdgesOp(ConfigBaseModel):
    """Realize declared inverse pairs as explicit logical edges (portable to every backend).

    Edges are derived from the relation map: for each directed edge ``(S, T, r)``
    whose relation has a declared pair ``inv``, adds the logical edge
    ``(T, S, inv)`` unless it exists and its physical spec, and sets
    ``emit_inverse`` on the edge steps that write ``r``, so the same rows write
    both. No step is generated: the inverse is mirrored at assembly, after the
    relation is resolved, which covers every way a step can name its relation.
    A resource that already writes the inverse with a step of its own is left
    alone. The pair must be declared first (:class:`DeclareEdgeInversesOp`);
    this op never declares. A symmetric relation has no inverse edge to add --
    its edges are undirected. Refused for a relation whose inverse is native
    (:class:`SetNativeInversesOp`), since both would store the same fact.

    Withdraw with :class:`RemoveEdgesOp` on the inverse edges: removing an edge
    clears the ``emit_inverse`` flags that fed it.
    """

    op: Literal["add_inverse_edges"] = "add_inverse_edges"
    relations: list[str] | None = PydanticField(
        default=None,
        description=(
            "Paired relations whose edges get their inverse edge. Omitted: every "
            "relation in ``edge_config.inverses``, both sides."
        ),
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_unique(self) -> AddInverseEdgesOp:
        if self.relations is not None and len(set(self.relations)) != len(
            self.relations
        ):
            raise ValueError("add_inverse_edges: relations must be unique")
        return self

Attributes

op = 'add_inverse_edges' class-attribute instance-attribute

relations = PydanticField(default=None, description='Paired relations whose edges get their inverse edge. Omitted: every relation in ``edge_config.inverses``, both sides.', min_length=1) class-attribute instance-attribute

AddResourceTransformsOp

Bases: ConfigBaseModel

Append transform steps to named resources' pipelines.

The first op whose primary effect is ingestion: graph_schema is untouched. Steps land at the root level of each pipeline unless at names a deeper one; at whichever level they land, actor type-priority sorting (transform runs before vertex extraction at the same level) makes the position safe.

The level is load-bearing rather than cosmetic. An actor reads its transform buffer at its own LocationIndex with no ancestor fallback, and a descend subtree runs before its own level's transforms, so a step appended at the root is invisible to a vertex produced under a descend — and a transform whose declared inputs are missing skips silently by default. Target the level that produces the vertex.

Steps may reference a registry transform via call.use (resolved against the manifest's existing ingestion_model.transforms union the op's own transforms) or carry a fully inline call (module + foo + params), which cannot collide by name.

Source code in graflo/architecture/evolution/ops.py

class AddResourceTransformsOp(ConfigBaseModel):
    """Append transform steps to named resources' pipelines.

    The first op whose primary effect is ingestion: ``graph_schema`` is
    untouched. Steps land at the root level of each pipeline unless ``at``
    names a deeper one; at whichever level they land, actor type-priority
    sorting (transform runs before vertex extraction at the same level) makes
    the position safe.

    The level is load-bearing rather than cosmetic. An actor reads its
    transform buffer at its own ``LocationIndex`` with no ancestor fallback,
    and a ``descend`` subtree runs *before* its own level's transforms, so a
    step appended at the root is invisible to a vertex produced under a
    ``descend`` — and a transform whose declared inputs are missing skips
    silently by default. Target the level that produces the vertex.

    Steps may reference a registry transform via ``call.use`` (resolved
    against the manifest's existing ``ingestion_model.transforms`` union the
    op's own ``transforms``) or carry a fully inline ``call``
    (``module`` + ``foo`` + ``params``), which cannot collide by name.
    """

    op: Literal["add_resource_transforms"] = "add_resource_transforms"
    additions: dict[str, list[dict[str, Any]]] = PydanticField(
        ...,
        description=(
            "Per-resource transform steps to append: "
            "``{resource_name: [step_dict, ...]}``."
        ),
        min_length=1,
    )
    at: dict[str, list[int]] = PydanticField(
        default_factory=dict,
        description=(
            "Per-resource pipeline level to append into: "
            "``{resource_name: [step_index, ...]}``. Each index must address a "
            "``descend`` step, descending one level per element; an omitted or "
            "empty path means the root level."
        ),
    )
    transforms: list[ProtoTransform] = PydanticField(
        default_factory=list,
        description=(
            "Named transforms to register in ``ingestion_model.transforms`` "
            "for steps that reference them via ``call.use``. A name already "
            "registered with a different body is an error at apply time."
        ),
    )

    @model_validator(mode="after")
    def _validate_steps(self) -> AddResourceTransformsOp:
        from graflo.architecture.contract.ingestion.steps.models import (
            TransformActorConfig,
        )
        from graflo.architecture.contract.ingestion.steps.normalize import (
            normalize_actor_step,
        )

        for resource_name, steps in self.additions.items():
            if not steps:
                raise ValueError(
                    f"add_resource_transforms: empty step list for resource "
                    f"{resource_name!r}"
                )
            for step in steps:
                normalized = normalize_actor_step(step)
                if (
                    not isinstance(normalized, dict)
                    or normalized.get("type") != "transform"
                ):
                    raise ValueError(
                        f"add_resource_transforms: step for resource "
                        f"{resource_name!r} is not a transform step: {step!r}"
                    )
                TransformActorConfig.model_validate(normalized)
        stray = sorted(set(self.at) - set(self.additions))
        if stray:
            raise ValueError(
                f"add_resource_transforms: `at` names resources {stray} with no "
                "steps to append"
            )
        for resource_name, path in self.at.items():
            if any(index < 0 for index in path):
                raise ValueError(
                    f"add_resource_transforms: `at` path for resource "
                    f"{resource_name!r} has a negative index: {path}"
                )
        for proto in self.transforms:
            if not proto.name:
                raise ValueError(
                    "add_resource_transforms: registry transforms must define "
                    "a non-empty name"
                )
        return self

Attributes

additions = PydanticField(..., description='Per-resource transform steps to append: ``{resource_name: [step_dict, ...]}``.', min_length=1) class-attribute instance-attribute

at = PydanticField(default_factory=dict, description='Per-resource pipeline level to append into: ``{resource_name: [step_index, ...]}``. Each index must address a ``descend`` step, descending one level per element; an omitted or empty path means the root level.') class-attribute instance-attribute

op = 'add_resource_transforms' class-attribute instance-attribute

transforms = PydanticField(default_factory=list, description='Named transforms to register in ``ingestion_model.transforms`` for steps that reference them via ``call.use``. A name already registered with a different body is an error at apply time.') class-attribute instance-attribute

AddResourcesOp

Bases: ConfigBaseModel

Introduce ingestion resources, in the shape the ingestion block accepts.

The unary way to grow ingestion_model: without it a change set could only ever rename or narrow the resources it started with, and the differ had to report an added resource as inexpressible.

Source code in graflo/architecture/evolution/ops.py

class AddResourcesOp(ConfigBaseModel):
    """Introduce ingestion resources, in the shape the ingestion block accepts.

    The unary way to grow ``ingestion_model``: without it a change set could
    only ever rename or narrow the resources it started with, and the differ
    had to report an added resource as inexpressible.
    """

    op: Literal["add_resources"] = "add_resources"
    resources: list[ResourceConfig] = PydanticField(
        ...,
        description="Full resource definitions.",
        min_length=1,
    )
    transforms: list[ProtoTransform] = PydanticField(
        default_factory=list,
        description=(
            "Named transforms to register in ``ingestion_model.transforms`` for "
            "steps of the new resources that reference them via ``call.use``. "
            "Unioned by name exactly as ``add_resource_transforms`` does: an "
            "identical body already registered dedupes, a different one is an "
            "error at apply time."
        ),
    )

    @model_validator(mode="after")
    def _validate_unique_names(self) -> AddResourcesOp:
        names = [resource.name for resource in self.resources]
        if len(names) != len(set(names)):
            raise ValueError("add_resources entries must be unique by name")
        return self

Attributes

op = 'add_resources' class-attribute instance-attribute

resources = PydanticField(..., description='Full resource definitions.', min_length=1) class-attribute instance-attribute

transforms = PydanticField(default_factory=list, description='Named transforms to register in ``ingestion_model.transforms`` for steps of the new resources that reference them via ``call.use``. Unioned by name exactly as ``add_resource_transforms`` does: an identical body already registered dedupes, a different one is an error at apply time.') class-attribute instance-attribute

AddSecondaryIdentitiesOp

Bases: ConfigBaseModel

Declare alternate lookup keys on existing vertices.

Secondary identities are lookup-only: upserts keep using the primary identity. Each declared field-set automatically gains a non-unique index at :meth:Schema.finish_init, so this op is how an edge-only source gains a way to reference endpoints by a business key without touching the primary identity.

Source code in graflo/architecture/evolution/ops.py

class AddSecondaryIdentitiesOp(ConfigBaseModel):
    """Declare alternate lookup keys on existing vertices.

    Secondary identities are lookup-only: upserts keep using the primary identity.
    Each declared field-set automatically gains a non-unique index at
    :meth:`Schema.finish_init`, so this op is how an edge-only source gains a way to
    reference endpoints by a business key without touching the primary identity.
    """

    op: Literal["add_secondary_identities"] = "add_secondary_identities"
    additions: dict[str, list[SecondaryIdentity]] = PydanticField(
        ...,
        description=(
            "Per-vertex secondary identities to declare: "
            "``{vertex_name: [{name, fields}, ...]}``. A bare field list is accepted "
            "for each entry and auto-named."
        ),
        min_length=1,
    )

Attributes

additions = PydanticField(..., description='Per-vertex secondary identities to declare: ``{vertex_name: [{name, fields}, ...]}``. A bare field list is accepted for each entry and auto-named.', min_length=1) class-attribute instance-attribute

op = 'add_secondary_identities' class-attribute instance-attribute

AddVertexIndexesOp

Bases: ConfigBaseModel

Author secondary indexes on vertices in the database profile.

Source code in graflo/architecture/evolution/ops.py

class AddVertexIndexesOp(ConfigBaseModel):
    """Author secondary indexes on vertices in the database profile."""

    op: Literal["add_vertex_indexes"] = "add_vertex_indexes"
    indexes: dict[str, Annotated[list[Index], PydanticField(min_length=1)]] = (
        PydanticField(
            ...,
            description="``{vertex_name: [Index, ...]}``.",
            min_length=1,
        )
    )

Attributes

indexes = PydanticField(..., description='``{vertex_name: [Index, ...]}``.', min_length=1) class-attribute instance-attribute

op = 'add_vertex_indexes' class-attribute instance-attribute

AddVertexPropertiesOp

Bases: ConfigBaseModel

Add vertex properties to existing logical vertex types.

An entry may be a bare name or a full :class:~graflo.architecture.schema.vertex.Field. Bare names were the original shape and still mean what they meant -- an untyped property -- but they cannot express a property that arrives with a type and a grounding, which is what any op stream that adds a measured or temporal property has to say in one replayable step.

Source code in graflo/architecture/evolution/ops.py

class AddVertexPropertiesOp(ConfigBaseModel):
    """Add vertex properties to existing logical vertex types.

    An entry may be a bare name or a full :class:`~graflo.architecture.schema.vertex.Field`.
    Bare names were the original shape and still mean what they meant -- an
    untyped property -- but they cannot express a property that arrives with a
    type and a grounding, which is what any op stream that adds a *measured* or
    *temporal* property has to say in one replayable step.
    """

    op: Literal["add_vertex_properties"] = "add_vertex_properties"
    additions: dict[str, list[str | Field]] = PydanticField(
        ...,
        description=(
            "Per-vertex property additions: ``{vertex_name: [field_name | Field, ...]}``."
        ),
        min_length=1,
    )

    def field_names(self, vertex: str) -> list[str]:
        """The names added to *vertex*, whichever shape they were written in."""
        return [
            entry if isinstance(entry, str) else entry.name
            for entry in self.additions.get(vertex, [])
        ]

Attributes

additions = PydanticField(..., description='Per-vertex property additions: ``{vertex_name: [field_name | Field, ...]}``.', min_length=1) class-attribute instance-attribute

op = 'add_vertex_properties' class-attribute instance-attribute

Methods:

field_names(vertex)

The names added to vertex, whichever shape they were written in.

Source code in graflo/architecture/evolution/ops.py

def field_names(self, vertex: str) -> list[str]:
    """The names added to *vertex*, whichever shape they were written in."""
    return [
        entry if isinstance(entry, str) else entry.name
        for entry in self.additions.get(vertex, [])
    ]

AddVerticesOp

Bases: ConfigBaseModel

Introduce new logical vertex types.

The unary counterpart to what :class:MergeManifestsOp can only do binarily. A replayable change set that cannot introduce a type could only ever describe a shrinking graph, which is why this exists alongside remove_vertices.

Source code in graflo/architecture/evolution/ops.py

class AddVerticesOp(ConfigBaseModel):
    """Introduce new logical vertex types.

    The unary counterpart to what :class:`MergeManifestsOp` can only do binarily.
    A replayable change set that cannot introduce a type could only ever describe a
    shrinking graph, which is why this exists alongside ``remove_vertices``.
    """

    op: Literal["add_vertices"] = "add_vertices"
    vertices: list[Vertex] = PydanticField(
        ...,
        description="Full vertex definitions, in the shape the schema block accepts.",
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_unique_names(self) -> AddVerticesOp:
        names = [vertex.name for vertex in self.vertices]
        if len(names) != len(set(names)):
            raise ValueError("add_vertices entries must be unique by name")
        return self

Attributes

op = 'add_vertices' class-attribute instance-attribute

vertices = PydanticField(..., description='Full vertex definitions, in the shape the schema block accepts.', min_length=1) class-attribute instance-attribute

AssignedIdentityTarget

Bases: ConfigBaseModel

Target an assigned identity: an intentional UUID primary key.

Source code in graflo/architecture/evolution/ops.py

class AssignedIdentityTarget(ConfigBaseModel):
    """Target an assigned identity: an intentional UUID primary key."""

    mode: Literal["assigned"] = "assigned"

Attributes

mode = 'assigned' class-attribute instance-attribute

BlankIdentityTarget

Bases: ConfigBaseModel

Target a blank identity: an auto-generated placeholder ID.

Source code in graflo/architecture/evolution/ops.py

class BlankIdentityTarget(ConfigBaseModel):
    """Target a blank identity: an auto-generated placeholder ID."""

    mode: Literal["blank"] = "blank"

Attributes

mode = 'blank' class-attribute instance-attribute

CanonicalizeOp

Bases: ConfigBaseModel

Relabel classes, attributes and relations by one vocabulary map, in one step.

The map is a partial function on names — identity where unmapped — applied simultaneously over the original schema, so a chain ({X: Z, Z: Q}) and a swap resolve without an intermediate state, and the fibers of the map are exactly the groups that merge. A target that already exists and does not move must be declared a member of its own group with a self entry (Company: Company); otherwise the op refuses rather than merging into it silently. properties is keyed by the source class name and is applied before the class relabel.

This is the single lowering of a :class:~graflo.architecture.evolution.canonical.CanonicalMap, and the per-side step of :func:~graflo.architecture.evolution.merge.merge_manifests.

Source code in graflo/architecture/evolution/ops.py

class CanonicalizeOp(ConfigBaseModel):
    """Relabel classes, attributes and relations by one vocabulary map, in one step.

    The map is a partial function on names — identity where unmapped — applied
    simultaneously over the original schema, so a chain (``{X: Z, Z: Q}``) and
    a swap resolve without an intermediate state, and the fibers of the map
    are exactly the groups that merge. A target that already exists and does
    not move must be declared a member of its own group with a self entry
    (``Company: Company``); otherwise the op refuses rather than merging into
    it silently. ``properties`` is keyed by the *source* class name and is
    applied before the class relabel.

    This is the single lowering of a
    :class:`~graflo.architecture.evolution.canonical.CanonicalMap`, and the
    per-side step of
    :func:`~graflo.architecture.evolution.merge.merge_manifests`.
    """

    op: Literal["canonicalize"] = "canonicalize"
    vertices: dict[str, str] = PydanticField(
        default_factory=dict,
        description="Class map: ``{source_class: canonical_class}``.",
    )
    properties: dict[str, dict[str, str]] = PydanticField(
        default_factory=dict,
        description=(
            "Per-source-class attribute map: "
            "``{source_class: {source_attr: canonical_attr}}``."
        ),
    )
    relations: dict[str, str] = PydanticField(
        default_factory=dict,
        description="Relation map: ``{source_relation: canonical_relation}``.",
    )
    allow_merges: bool = PydanticField(
        default=False,
        description=(
            "Accept a group of more than one class or relation collapsing onto "
            "one target. A merge fuses entities and can create self-relations, "
            "so it is acknowledged here rather than inferred from the map."
        ),
    )
    allow_self_relations: bool = PydanticField(
        default=False,
        description=(
            "Accept a merge whose sources are connected by an edge that becomes "
            "a self-relation once both endpoints land on the same class."
        ),
    )
    allow_observation_fusion: bool = PydanticField(
        default=False,
        description=(
            "Accept a merge whose sources are produced in one accumulator slot "
            "(the same pipeline level and the same ``role``, or both bare), "
            "fusing those observations into one node."
        ),
    )

    @model_validator(mode="after")
    def _validate_map(self) -> CanonicalizeOp:
        validate_vocabulary_map(
            self.vertices,
            self.relations,
            self.properties,
            allow_merges=self.allow_merges,
            kind="canonicalize",
            merge_hint="set allow_merges=true",
        )
        return self

    @property
    def vertex_groups(self) -> dict[str, list[str]]:
        """``{target: [members]}`` over ``vertices``, self entries included."""
        return vocabulary_groups(self.vertices)

    @property
    def relation_groups(self) -> dict[str, list[str]]:
        """``{target: [members]}`` over ``relations``, self entries included."""
        return vocabulary_groups(self.relations)

    @property
    def merges(self) -> bool:
        """Whether any class or relation group has more than one member."""
        return any(
            len(members) > 1
            for groups in (self.vertex_groups, self.relation_groups)
            for members in groups.values()
        )

Attributes

allow_merges = PydanticField(default=False, description='Accept a group of more than one class or relation collapsing onto one target. A merge fuses entities and can create self-relations, so it is acknowledged here rather than inferred from the map.') class-attribute instance-attribute

allow_observation_fusion = PydanticField(default=False, description='Accept a merge whose sources are produced in one accumulator slot (the same pipeline level and the same ``role``, or both bare), fusing those observations into one node.') class-attribute instance-attribute

allow_self_relations = PydanticField(default=False, description='Accept a merge whose sources are connected by an edge that becomes a self-relation once both endpoints land on the same class.') class-attribute instance-attribute

merges property

Whether any class or relation group has more than one member.

op = 'canonicalize' class-attribute instance-attribute

properties = PydanticField(default_factory=dict, description='Per-source-class attribute map: ``{source_class: {source_attr: canonical_attr}}``.') class-attribute instance-attribute

relation_groups property

{target: [members]} over relations, self entries included.

relations = PydanticField(default_factory=dict, description='Relation map: ``{source_relation: canonical_relation}``.') class-attribute instance-attribute

vertex_groups property

{target: [members]} over vertices, self entries included.

vertices = PydanticField(default_factory=dict, description='Class map: ``{source_class: canonical_class}``.') class-attribute instance-attribute

ChangeFieldTypesOp

Bases: ConfigBaseModel

Set the logical type of existing vertex or edge properties.

Makes the differ's CHANGE_VERTEX_FIELD_TYPE / CHANGE_EDGE_FIELD_TYPE authorable. Targets are validated against the profile's db_flavor so an unsupported type fails here rather than at define time.

Source code in graflo/architecture/evolution/ops.py

class ChangeFieldTypesOp(ConfigBaseModel):
    """Set the logical type of existing vertex or edge properties.

    Makes the differ's ``CHANGE_VERTEX_FIELD_TYPE`` / ``CHANGE_EDGE_FIELD_TYPE``
    authorable. Targets are validated against the profile's ``db_flavor`` so an
    unsupported type fails here rather than at define time.
    """

    op: Literal["change_field_types"] = "change_field_types"
    vertices: dict[str, dict[str, FieldTypeSpec]] = PydanticField(
        default_factory=dict,
        description="``{vertex_name: {field_name: {type, item_type}}}``.",
    )
    edges: dict[str, dict[str, FieldTypeSpec]] = PydanticField(
        default_factory=dict,
        description="``{relation_name: {field_name: {type, item_type}}}``.",
    )

    @model_validator(mode="after")
    def _require_a_target(self) -> ChangeFieldTypesOp:
        if not self.vertices and not self.edges:
            raise ValueError(
                "change_field_types requires at least one of vertices or edges"
            )
        return self

Attributes

edges = PydanticField(default_factory=dict, description='``{relation_name: {field_name: {type, item_type}}}``.') class-attribute instance-attribute

op = 'change_field_types' class-attribute instance-attribute

vertices = PydanticField(default_factory=dict, description='``{vertex_name: {field_name: {type, item_type}}}``.') class-attribute instance-attribute

DeclareEdgeInversesOp

Bases: ConfigBaseModel

Declare inverse pairs and symmetric relations in edge_config.

Purely logical: it records how relation names read one fact from its two endpoints and creates nothing. inverses pairs two distinct names; a pair is unordered, so {a: b}, {b: a} and {a: b, b: a} declare the same thing. symmetric names relations that are their own inverse. Together they must give every relation at most one inverse (no a: b with b: c), in the op and against what is already declared.

Realize a pair with :class:AddInverseEdgesOp (explicit, portable inverse edges) or :class:SetNativeInversesOp (TigerGraph maintains the pair) -- never both for one relation. A symmetric relation is realized by its edges being undirected (:class:SetEdgeDirectedOp), which the schema requires.

Source code in graflo/architecture/evolution/ops.py

class DeclareEdgeInversesOp(ConfigBaseModel):
    """Declare inverse pairs and symmetric relations in ``edge_config``.

    Purely logical: it records how relation names read one fact from its two
    endpoints and creates nothing. ``inverses`` pairs two distinct names; a pair
    is unordered, so ``{a: b}``, ``{b: a}`` and ``{a: b, b: a}`` declare the same
    thing. ``symmetric`` names relations that are their own inverse. Together
    they must give every relation at most one inverse (no ``a: b`` with
    ``b: c``), in the op and against what is already declared.

    Realize a pair with :class:`AddInverseEdgesOp` (explicit, portable inverse
    edges) or :class:`SetNativeInversesOp` (TigerGraph maintains the pair) --
    never both for one relation. A symmetric relation is realized by its edges
    being undirected (:class:`SetEdgeDirectedOp`), which the schema requires.
    """

    op: Literal["declare_edge_inverses"] = "declare_edge_inverses"
    inverses: dict[str, str] = PydanticField(
        default_factory=dict,
        description="Pairs to declare: ``{relation: inverse_relation}``, either order.",
    )
    symmetric: list[str] = PydanticField(
        default_factory=list,
        description="Relations to declare as their own inverse.",
    )

    @model_validator(mode="after")
    def _validate_table(self) -> DeclareEdgeInversesOp:
        if not self.inverses and not self.symmetric:
            raise ValueError(
                "declare_edge_inverses: nothing to declare; give inverses or symmetric"
            )
        normalize_inverse_table(
            self.inverses.items(), self.symmetric, kind="declare_edge_inverses"
        )
        return self

Attributes

inverses = PydanticField(default_factory=dict, description='Pairs to declare: ``{relation: inverse_relation}``, either order.') class-attribute instance-attribute

op = 'declare_edge_inverses' class-attribute instance-attribute

symmetric = PydanticField(default_factory=list, description='Relations to declare as their own inverse.') class-attribute instance-attribute

DerivationSpec

Bases: ConfigBaseModel

How one resource derives a canonical attribute from its raw doc fields.

Source code in graflo/architecture/evolution/ops.py

class DerivationSpec(ConfigBaseModel):
    """How one resource derives a canonical attribute from its raw doc fields."""

    input: list[str] = PydanticField(
        ...,
        min_length=1,
        description=(
            "RAW source-doc field names fed to the function, in order. "
            "Documents keep their original keys after property renames, so "
            "canonical property names are usually wrong here."
        ),
    )
    module: str = PydanticField(
        default="graflo.util.transform",
        description="Module holding the derivation function.",
    )
    foo: str = PydanticField(
        default="normalized_key",
        description=(
            "Function name; called as ``foo(*values, **params)``. The default "
            "trims and casefolds one value."
        ),
    )
    params: dict[str, Any] = PydanticField(
        default_factory=dict,
        description="Keyword parameters for the function.",
    )
    when: TransformGuardConfig | None = PydanticField(
        default=None,
        description=(
            "Run the step only for documents whose RAW field holds one of the "
            "listed values (``{field, in}``). A merge derives this guard from "
            "how the resource produces the class; one set here replaces it. "
            "Not allowed on a spec keyed by member: the member decides."
        ),
    )

Attributes

foo = PydanticField(default='normalized_key', description='Function name; called as ``foo(*values, **params)``. The default trims and casefolds one value.') class-attribute instance-attribute

input = PydanticField(..., min_length=1, description='RAW source-doc field names fed to the function, in order. Documents keep their original keys after property renames, so canonical property names are usually wrong here.') class-attribute instance-attribute

module = PydanticField(default='graflo.util.transform', description='Module holding the derivation function.') class-attribute instance-attribute

params = PydanticField(default_factory=dict, description='Keyword parameters for the function.') class-attribute instance-attribute

when = PydanticField(default=None, description='Run the step only for documents whose RAW field holds one of the listed values (``{field, in}``). A merge derives this guard from how the resource produces the class; one set here replaces it. Not allowed on a spec keyed by member: the member decides.') class-attribute instance-attribute

DerivedBranch

Bases: ConfigBaseModel

A funnel branch over an attribute each source derives from its own columns.

name is the canonical attribute the branch keys on: the digest's input, not where the digest is stored (that is :attr:VertexEquivalence.digest_field). sources is keyed by resource, because derivation inputs are that resource's raw column names. An entry is either

• one :class:DerivationSpec — the resource derives the attribute the same way for every record of the class it produces; or
• a dict keyed by member class — the resource produces several members and which one a record is decides the derivation. A member is keyed by its own name on its side or by its canonical one.

The merge reads how the resource produces the class on its side and guards each step with when: a member-keyed spec on the discriminator values that route onto its member, a single spec behind a vertex_router on the values that route onto the class, and nothing for a plain vertex step. An explicit :attr:DerivationSpec.when replaces the derived guard.

Source code in graflo/architecture/evolution/ops.py

class DerivedBranch(ConfigBaseModel):
    """A funnel branch over an attribute each source derives from its own columns.

    ``name`` is the canonical attribute the branch keys on: the digest's
    input, not where the digest is stored (that is
    :attr:`VertexEquivalence.digest_field`). ``sources`` is
    keyed by resource, because derivation inputs are that resource's raw
    column names. An entry is either

    * one :class:`DerivationSpec` — the resource derives the attribute the
      same way for every record of the class it produces; or
    * a dict keyed by member class — the resource produces several members
      and which one a record *is* decides the derivation. A member is keyed by
      its own name on its side or by its canonical one.

    The merge reads how the resource produces the class on its side and guards
    each step with ``when``: a member-keyed spec on the discriminator values
    that route onto its member, a single spec behind a ``vertex_router`` on
    the values that route onto the class, and nothing for a plain ``vertex``
    step. An explicit :attr:`DerivationSpec.when` replaces the derived guard.
    """

    name: str = PydanticField(
        ...,
        description=(
            "Derived attribute this branch digests; its funnel branch id. Not "
            "where the key is stored: see `VertexEquivalence.digest_field`."
        ),
    )
    sources: dict[str, DerivationSpec | dict[str, DerivationSpec]] = PydanticField(
        ...,
        min_length=1,
        description=(
            "Per-resource derivation: ``{resource: spec}``, or ``{resource: "
            "{member_class: spec}}`` when the member a document is must decide "
            "the derivation."
        ),
    )

    @model_validator(mode="after")
    def _validate_sources(self) -> DerivedBranch:
        _refuse_member_keyed_guards("derived branch", self.sources, self.name)
        return self

    def specs_for(self, resource: str) -> list[DerivationSpec]:
        """Derivations *resource* contributes to this branch, in order."""
        spec = self.sources.get(resource)
        if spec is None:
            return []
        return [spec] if isinstance(spec, DerivationSpec) else list(spec.values())

    def members_for(self, resource: str) -> list[str] | None:
        """Member classes keying *resource*'s specs, or ``None`` if unkeyed."""
        spec = self.sources.get(resource)
        return list(spec) if isinstance(spec, dict) else None

Attributes

name = PydanticField(..., description='Derived attribute this branch digests; its funnel branch id. Not where the key is stored: see `VertexEquivalence.digest_field`.') class-attribute instance-attribute

sources = PydanticField(..., min_length=1, description='Per-resource derivation: ``{resource: spec}``, or ``{resource: {member_class: spec}}`` when the member a document is must decide the derivation.') class-attribute instance-attribute

Methods:

members_for(resource)

Member classes keying resource's specs, or None if unkeyed.

Source code in graflo/architecture/evolution/ops.py

def members_for(self, resource: str) -> list[str] | None:
    """Member classes keying *resource*'s specs, or ``None`` if unkeyed."""
    spec = self.sources.get(resource)
    return list(spec) if isinstance(spec, dict) else None

specs_for(resource)

Derivations resource contributes to this branch, in order.

Source code in graflo/architecture/evolution/ops.py

def specs_for(self, resource: str) -> list[DerivationSpec]:
    """Derivations *resource* contributes to this branch, in order."""
    spec = self.sources.get(resource)
    if spec is None:
        return []
    return [spec] if isinstance(spec, DerivationSpec) else list(spec.values())

EdgeFieldSemanticsTarget

Bases: ConfigBaseModel

One property of one edge triple, and the grounding to put on it.

Edge properties carry the same FieldSemantics as vertex properties -- a since on an edge has a unit as much as a temperature on a vertex does -- but until this target existed no op could reach them.

Source code in graflo/architecture/evolution/ops.py

class EdgeFieldSemanticsTarget(ConfigBaseModel):
    """One property of one edge triple, and the grounding to put on it.

    Edge properties carry the same ``FieldSemantics`` as vertex properties --
    a ``since`` on an edge has a unit as much as a ``temperature`` on a vertex
    does -- but until this target existed no op could reach them.
    """

    source: str = PydanticField(..., description="Source vertex type name.")
    target: str = PydanticField(..., description="Target vertex type name.")
    relation: str | None = PydanticField(
        default=None,
        description="Relation name; ``None`` matches the edge with no relation set.",
    )
    field: str = PydanticField(..., description="Property name on that edge.")
    semantics: FieldSemantics | None = PydanticField(
        default=None,
        description="Grounding for the property; ``None`` clears it.",
    )

    def edge_id(self) -> tuple[str, str, str | None]:
        return self.source, self.target, self.relation

    def key(self) -> tuple[Any, ...]:
        return ("edge", self.source, self.target, self.relation, self.field)

Attributes

field = PydanticField(..., description='Property name on that edge.') class-attribute instance-attribute

relation = PydanticField(default=None, description='Relation name; ``None`` matches the edge with no relation set.') class-attribute instance-attribute

semantics = PydanticField(default=None, description='Grounding for the property; ``None`` clears it.') class-attribute instance-attribute

source = PydanticField(..., description='Source vertex type name.') class-attribute instance-attribute

target = PydanticField(..., description='Target vertex type name.') class-attribute instance-attribute

Methods:

edge_id()

Source code in graflo/architecture/evolution/ops.py

def edge_id(self) -> tuple[str, str, str | None]:
    return self.source, self.target, self.relation

key()

Source code in graflo/architecture/evolution/ops.py

def key(self) -> tuple[Any, ...]:
    return ("edge", self.source, self.target, self.relation, self.field)

EdgeIdentitiesEntry

Bases: ConfigBaseModel

New uniqueness keys for one edge triple.

Source code in graflo/architecture/evolution/ops.py

class EdgeIdentitiesEntry(ConfigBaseModel):
    """New uniqueness keys for one edge triple."""

    source: str = PydanticField(..., description="Source vertex type name.")
    target: str = PydanticField(..., description="Target vertex type name.")
    relation: str | None = PydanticField(
        default=None,
        description="Relation name; ``None`` matches the edge with no relation set.",
    )
    identities: list[list[str]] = PydanticField(
        ...,
        description=(
            "Replacement uniqueness keys. Each key lists fields that, together with "
            "the resolved endpoints, must be unique; the ``source`` / ``target`` "
            "tokens stand for the endpoints themselves. An empty list clears them."
        ),
    )

    def edge_id(self) -> tuple[str, str, str | None]:
        return self.source, self.target, self.relation

Attributes

identities = PydanticField(..., description='Replacement uniqueness keys. Each key lists fields that, together with the resolved endpoints, must be unique; the ``source`` / ``target`` tokens stand for the endpoints themselves. An empty list clears them.') class-attribute instance-attribute

relation = PydanticField(default=None, description='Relation name; ``None`` matches the edge with no relation set.') class-attribute instance-attribute

source = PydanticField(..., description='Source vertex type name.') class-attribute instance-attribute

target = PydanticField(..., description='Target vertex type name.') class-attribute instance-attribute

Methods:

edge_id()

Source code in graflo/architecture/evolution/ops.py

def edge_id(self) -> tuple[str, str, str | None]:
    return self.source, self.target, self.relation

EdgeIndexEntry

Bases: ConfigBaseModel

Indexes for one edge physical spec.

Source code in graflo/architecture/evolution/ops.py

class EdgeIndexEntry(ConfigBaseModel):
    """Indexes for one edge physical spec."""

    source: str = PydanticField(..., description="Source vertex type name.")
    target: str = PydanticField(..., description="Target vertex type name.")
    relation: str | None = PydanticField(
        default=None,
        description="Relation name; ``None`` matches the edge with no relation set.",
    )
    purpose: str | None = PydanticField(
        default=None,
        description="Physical variant purpose; ``None`` addresses the base spec.",
    )
    indexes: list[Index] = PydanticField(
        default_factory=list,
        description="Indexes to add to this spec.",
    )
    fields: list[list[str]] = PydanticField(
        default_factory=list,
        description="Field lists identifying indexes to remove from this spec.",
    )

    def physical_key(self) -> tuple[str, str, str | None, str | None]:
        return self.source, self.target, self.relation, self.purpose

Attributes

fields = PydanticField(default_factory=list, description='Field lists identifying indexes to remove from this spec.') class-attribute instance-attribute

indexes = PydanticField(default_factory=list, description='Indexes to add to this spec.') class-attribute instance-attribute

purpose = PydanticField(default=None, description='Physical variant purpose; ``None`` addresses the base spec.') class-attribute instance-attribute

relation = PydanticField(default=None, description='Relation name; ``None`` matches the edge with no relation set.') class-attribute instance-attribute

source = PydanticField(..., description='Source vertex type name.') class-attribute instance-attribute

target = PydanticField(..., description='Target vertex type name.') class-attribute instance-attribute

Methods:

physical_key()

Source code in graflo/architecture/evolution/ops.py

def physical_key(self) -> tuple[str, str, str | None, str | None]:
    return self.source, self.target, self.relation, self.purpose

EdgeRetargetEntry

Bases: ConfigBaseModel

Repoint one edge triple at a different source and/or target vertex type.

Source code in graflo/architecture/evolution/ops.py

class EdgeRetargetEntry(ConfigBaseModel):
    """Repoint one edge triple at a different source and/or target vertex type."""

    source: str = PydanticField(..., description="Current source vertex type name.")
    target: str = PydanticField(..., description="Current target vertex type name.")
    relation: str | None = PydanticField(
        default=None,
        description="Relation name; ``None`` matches the edge with no relation set.",
    )
    new_source: str | None = PydanticField(
        default=None,
        description="Replacement source vertex type; omit to keep the current one.",
    )
    new_target: str | None = PydanticField(
        default=None,
        description="Replacement target vertex type; omit to keep the current one.",
    )

    def edge_id(self) -> tuple[str, str, str | None]:
        return self.source, self.target, self.relation

    def retargeted_edge_id(self) -> tuple[str, str, str | None]:
        return (
            self.new_source or self.source,
            self.new_target or self.target,
            self.relation,
        )

    @model_validator(mode="after")
    def _require_a_change(self) -> EdgeRetargetEntry:
        if self.new_source is None and self.new_target is None:
            raise ValueError(
                "retarget_edges requires at least one of new_source or new_target"
            )
        if self.retargeted_edge_id() == self.edge_id():
            raise ValueError(
                f"retarget_edges: edge {self.edge_id()} would not change endpoints"
            )
        return self

Attributes

new_source = PydanticField(default=None, description='Replacement source vertex type; omit to keep the current one.') class-attribute instance-attribute

new_target = PydanticField(default=None, description='Replacement target vertex type; omit to keep the current one.') class-attribute instance-attribute

relation = PydanticField(default=None, description='Relation name; ``None`` matches the edge with no relation set.') class-attribute instance-attribute

source = PydanticField(..., description='Current source vertex type name.') class-attribute instance-attribute

target = PydanticField(..., description='Current target vertex type name.') class-attribute instance-attribute

Methods:

edge_id()

Source code in graflo/architecture/evolution/ops.py

def edge_id(self) -> tuple[str, str, str | None]:
    return self.source, self.target, self.relation

retargeted_edge_id()

Source code in graflo/architecture/evolution/ops.py

def retargeted_edge_id(self) -> tuple[str, str, str | None]:
    return (
        self.new_source or self.source,
        self.new_target or self.target,
        self.relation,
    )

EdgeSelector

Bases: ConfigBaseModel

Schema edge triple selector matching :data:~graflo.architecture.graph_types.EdgeId.

Source code in graflo/architecture/evolution/ops.py

class EdgeSelector(ConfigBaseModel):
    """Schema edge triple selector matching :data:`~graflo.architecture.graph_types.EdgeId`."""

    source: str = PydanticField(..., description="Source vertex type name.")
    target: str = PydanticField(..., description="Target vertex type name.")
    relation: str | None = PydanticField(
        default=None,
        description="Relation name; ``None`` matches edges with no relation set.",
    )

    def edge_id(self) -> tuple[str, str, str | None]:
        return self.source, self.target, self.relation

Attributes

relation = PydanticField(default=None, description='Relation name; ``None`` matches edges with no relation set.') class-attribute instance-attribute

source = PydanticField(..., description='Source vertex type name.') class-attribute instance-attribute

target = PydanticField(..., description='Target vertex type name.') class-attribute instance-attribute

Methods:

edge_id()

Source code in graflo/architecture/evolution/ops.py

def edge_id(self) -> tuple[str, str, str | None]:
    return self.source, self.target, self.relation

EnsureExtractedFields

Bases: ConfigBaseModel

Fields that must survive extraction for one vertex type at one level.

Source code in graflo/architecture/evolution/ops.py

class EnsureExtractedFields(ConfigBaseModel):
    """Fields that must survive extraction for one vertex type at one level."""

    vertex: str = PydanticField(
        ...,
        description="The vertex type whose extraction must keep ``fields``.",
    )
    fields: list[str] = PydanticField(
        ...,
        min_length=1,
        description="Property names that must reach the extracted vertex document.",
    )
    at: list[int] = PydanticField(
        default_factory=list,
        description=(
            "Pipeline level holding the producing step, as ``descend`` step "
            "indices. Empty means the root level."
        ),
    )

Attributes

at = PydanticField(default_factory=list, description='Pipeline level holding the producing step, as ``descend`` step indices. Empty means the root level.') class-attribute instance-attribute

fields = PydanticField(..., min_length=1, description='Property names that must reach the extracted vertex document.') class-attribute instance-attribute

vertex = PydanticField(..., description='The vertex type whose extraction must keep ``fields``.') class-attribute instance-attribute

EnsureExtractedFieldsOp

Bases: ConfigBaseModel

Widen a producing step's projection so named fields are not dropped.

Needed because a vertex_router delivers differently from a vertex step. The router builds its child VertexActor at lindex.extend((role, 0)), where the transform buffer is empty, so derived fields reach the child only through the merged observation — that is, through passthrough or from. A plain vertex step instead reads the buffer directly, which bypasses keep_fields and extraction_scope entirely.

So on a router, extraction_scope: mapped_only or a keep_fields list that does not name the fields drops them silently. This op restores them: keep_fields gains the names, and under mapped_only the per-type vertex_from_map entry gains identity mappings — seeded from the router-level from when the entry does not exist yet, since creating it otherwise replaces the author's projection rather than extending it.

A plain vertex step, or a router that restricts nothing, is a no-op: the fields already survive.

Source code in graflo/architecture/evolution/ops.py

class EnsureExtractedFieldsOp(ConfigBaseModel):
    """Widen a producing step's projection so named fields are not dropped.

    Needed because a ``vertex_router`` delivers differently from a ``vertex``
    step. The router builds its child ``VertexActor`` at
    ``lindex.extend((role, 0))``, where the transform buffer is empty, so
    derived fields reach the child only through the merged observation — that
    is, through passthrough or ``from``. A plain ``vertex`` step instead reads
    the buffer directly, which bypasses ``keep_fields`` and
    ``extraction_scope`` entirely.

    So on a router, ``extraction_scope: mapped_only`` or a ``keep_fields`` list
    that does not name the fields drops them silently. This op restores them:
    ``keep_fields`` gains the names, and under ``mapped_only`` the per-type
    ``vertex_from_map`` entry gains identity mappings — seeded from the
    router-level ``from`` when the entry does not exist yet, since creating it
    otherwise replaces the author's projection rather than extending it.

    A plain ``vertex`` step, or a router that restricts nothing, is a no-op:
    the fields already survive.
    """

    op: Literal["ensure_extracted_fields"] = "ensure_extracted_fields"
    additions: dict[str, list[EnsureExtractedFields]] = PydanticField(
        ...,
        description=(
            "Per-resource extraction guarantees: "
            "``{resource_name: [{vertex, fields, at}, ...]}``."
        ),
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_entries(self) -> EnsureExtractedFieldsOp:
        for resource_name, entries in self.additions.items():
            if not entries:
                raise ValueError(
                    f"ensure_extracted_fields: empty entry list for resource "
                    f"{resource_name!r}"
                )
            for entry in entries:
                if any(index < 0 for index in entry.at):
                    raise ValueError(
                        f"ensure_extracted_fields: `at` path for resource "
                        f"{resource_name!r} has a negative index: {entry.at}"
                    )
        return self

Attributes

additions = PydanticField(..., description='Per-resource extraction guarantees: ``{resource_name: [{vertex, fields, at}, ...]}``.', min_length=1) class-attribute instance-attribute

op = 'ensure_extracted_fields' class-attribute instance-attribute

FieldSemanticsTarget

Bases: ConfigBaseModel

One property of one vertex, and the grounding to put on it.

Source code in graflo/architecture/evolution/ops.py

class FieldSemanticsTarget(ConfigBaseModel):
    """One property of one vertex, and the grounding to put on it."""

    vertex: str = PydanticField(..., description="Vertex type name.")
    field: str = PydanticField(..., description="Property name on that vertex.")
    semantics: FieldSemantics | None = PydanticField(
        default=None,
        description="Grounding for the property; ``None`` clears it.",
    )

    def key(self) -> tuple[Any, ...]:
        return ("vertex", self.vertex, self.field)

Attributes

field = PydanticField(..., description='Property name on that vertex.') class-attribute instance-attribute

semantics = PydanticField(default=None, description='Grounding for the property; ``None`` clears it.') class-attribute instance-attribute

vertex = PydanticField(..., description='Vertex type name.') class-attribute instance-attribute

Methods:

key()

Source code in graflo/architecture/evolution/ops.py

def key(self) -> tuple[Any, ...]:
    return ("vertex", self.vertex, self.field)

FieldTypeSpec

Bases: ConfigBaseModel

Target logical type for one property.

Source code in graflo/architecture/evolution/ops.py

class FieldTypeSpec(ConfigBaseModel):
    """Target logical type for one property."""

    type: FieldType | None = PydanticField(
        ...,
        description="New logical field type; ``None`` clears the declared type.",
    )
    item_type: FieldType | None = PydanticField(
        default=None,
        description="Element type, required when ``type`` is ``LIST``.",
    )

    @model_validator(mode="after")
    def _validate_item_type(self) -> FieldTypeSpec:
        if self.type == FieldType.LIST and self.item_type is None:
            raise ValueError("a LIST field type requires item_type")
        if self.type != FieldType.LIST and self.item_type is not None:
            raise ValueError("item_type is only meaningful for a LIST field type")
        return self

Attributes

item_type = PydanticField(default=None, description='Element type, required when ``type`` is ``LIST``.') class-attribute instance-attribute

type = PydanticField(..., description='New logical field type; ``None`` clears the declared type.') class-attribute instance-attribute

FunnelIdentityTarget

Bases: ConfigBaseModel

Target an identity funnel: fallback branches digested into digest_field.

The general form of :class:HashIdentityTarget — a flat hash key is a funnel with one branch. Both resolve to identity mode hash.

Source code in graflo/architecture/evolution/ops.py

class FunnelIdentityTarget(ConfigBaseModel):
    """Target an identity funnel: fallback branches digested into ``digest_field``.

    The general form of :class:`HashIdentityTarget` — a flat hash key is a funnel
    with one branch. Both resolve to identity mode ``hash``.
    """

    mode: Literal["funnel"] = "funnel"
    funnel: IdentityFunnel = PydanticField(
        ...,
        description="Ordered fallback branches; the first complete one wins.",
    )
    digest_field: str = PydanticField(
        default="id",
        min_length=1,
        description="Property the digest is stored in; the vertex's identity field.",
    )

    @model_validator(mode="after")
    def _digest_field_is_no_branch_field(self) -> FunnelIdentityTarget:
        # The cast drops a digest vertex's identity field from each record, so
        # a branch reading the same field could never complete.
        if self.digest_field in self.funnel.field_names:
            raise ValueError(
                f"funnel identity: digest_field {self.digest_field!r} is also a "
                "branch field; the digest would replace that branch's input"
            )
        return self

Attributes

digest_field = PydanticField(default='id', min_length=1, description="Property the digest is stored in; the vertex's identity field.") class-attribute instance-attribute

funnel = PydanticField(..., description='Ordered fallback branches; the first complete one wins.') class-attribute instance-attribute

mode = 'funnel' class-attribute instance-attribute

HashIdentityTarget

Bases: ConfigBaseModel

Target a hash identity: a deterministic synthetic id digested from fields.

Source code in graflo/architecture/evolution/ops.py

class HashIdentityTarget(ConfigBaseModel):
    """Target a hash identity: a deterministic synthetic ``id`` digested from fields."""

    mode: Literal["hash"] = "hash"
    hash_from: list[str] = PydanticField(
        ...,
        description="Source property names whose values are digested into ``id``.",
        min_length=1,
    )

Attributes

hash_from = PydanticField(..., description='Source property names whose values are digested into ``id``.', min_length=1) class-attribute instance-attribute

mode = 'hash' class-attribute instance-attribute

IdentityReplacement

Bases: ConfigBaseModel

New identity policy for one vertex, plus what becomes of the old one.

Source code in graflo/architecture/evolution/ops.py

class IdentityReplacement(ConfigBaseModel):
    """New identity policy for one vertex, plus what becomes of the old one."""

    to: IdentityTarget = PydanticField(
        ...,
        description="The identity policy this vertex should have after the op.",
    )
    retire: Literal["demote", "keep", "drop"] = PydanticField(
        default="demote",
        description=(
            "What happens to the old identity field-set. ``demote`` turns it into a "
            "secondary identity (lookup index follows automatically), ``keep`` leaves "
            "the fields as plain properties, ``drop`` removes them. Demotion is "
            "downgraded to ``keep`` when the old identity was synthetic (hash / "
            "assigned / blank) or already equals the new one."
        ),
    )
    retire_as: str | None = PydanticField(
        default=None,
        description=(
            "Name for the demoted secondary identity. Defaults to "
            "``retired_identity``. Only meaningful with ``retire: demote``."
        ),
    )
    endpoints: Literal["follow_new", "pin_to_retired"] = PydanticField(
        default="follow_new",
        description=(
            "How edge steps that match this vertex on its primary identity behave "
            "afterwards. ``follow_new`` (default) leaves them on the primary, so they "
            "match the new identity. ``pin_to_retired`` rewrites them to select the "
            "demoted secondary identity, preserving the previous matching behaviour "
            "for sources that only carry the old key. Requires ``retire: demote``."
        ),
    )

    @model_validator(mode="after")
    def _validate_endpoint_policy(self) -> IdentityReplacement:
        if self.endpoints == "pin_to_retired" and self.retire != "demote":
            raise ValueError(
                "endpoints: pin_to_retired requires retire: demote — there is no "
                "retired secondary identity to pin to otherwise"
            )
        if self.retire_as is not None and self.retire != "demote":
            raise ValueError("retire_as is only meaningful with retire: demote")
        return self

Attributes

endpoints = PydanticField(default='follow_new', description='How edge steps that match this vertex on its primary identity behave afterwards. ``follow_new`` (default) leaves them on the primary, so they match the new identity. ``pin_to_retired`` rewrites them to select the demoted secondary identity, preserving the previous matching behaviour for sources that only carry the old key. Requires ``retire: demote``.') class-attribute instance-attribute

retire = PydanticField(default='demote', description='What happens to the old identity field-set. ``demote`` turns it into a secondary identity (lookup index follows automatically), ``keep`` leaves the fields as plain properties, ``drop`` removes them. Demotion is downgraded to ``keep`` when the old identity was synthetic (hash / assigned / blank) or already equals the new one.') class-attribute instance-attribute

retire_as = PydanticField(default=None, description='Name for the demoted secondary identity. Defaults to ``retired_identity``. Only meaningful with ``retire: demote``.') class-attribute instance-attribute

to = PydanticField(..., description='The identity policy this vertex should have after the op.') class-attribute instance-attribute

LocalKeyBranch

Bases: ConfigBaseModel

The last funnel branch: each source's own key behind a namespace tag.

A record that completes no earlier branch keys on this one, so it is still written — as its own vertex, not joined with records of another source. local_key maps each resource to the column carrying its own key, or, like :attr:DerivedBranch.sources, to one such source per member class.

Source code in graflo/architecture/evolution/ops.py

class LocalKeyBranch(ConfigBaseModel):
    """The last funnel branch: each source's own key behind a namespace tag.

    A record that completes no earlier branch keys on this one, so it is still
    written — as its own vertex, not joined with records of another source.
    ``local_key`` maps each resource to the column carrying its own key, or,
    like :attr:`DerivedBranch.sources`, to one such source per member class.
    """

    local_key: dict[str, LocalKeySource | dict[str, LocalKeySource]] = PydanticField(
        ...,
        min_length=1,
        description=(
            "Per-resource local-key wiring: ``{resource: source}``, or "
            "``{resource: {member_class: source}}`` when the member decides."
        ),
    )
    name: str = PydanticField(
        default="local_key",
        description="Canonical fallback property name on the class.",
    )
    sep: str = PydanticField(
        default=":",
        description="Separator between tag and key.",
    )

    @model_validator(mode="after")
    def _validate_sources(self) -> LocalKeyBranch:
        _refuse_member_keyed_guards("local_key branch", self.local_key, self.name)
        return self

    @property
    def sources(self) -> dict[str, LocalKeySource | dict[str, LocalKeySource]]:
        """The per-resource wiring, under the name :class:`DerivedBranch` uses."""
        return self.local_key

    def sources_for(self, resource: str) -> list[LocalKeySource]:
        """Local-key sources *resource* contributes, in order."""
        source = self.local_key.get(resource)
        if source is None:
            return []
        return [source] if isinstance(source, LocalKeySource) else list(source.values())

    def members_for(self, resource: str) -> list[str] | None:
        """Member classes keying *resource*'s sources, or ``None`` if unkeyed."""
        source = self.local_key.get(resource)
        return list(source) if isinstance(source, dict) else None

Attributes

local_key = PydanticField(..., min_length=1, description='Per-resource local-key wiring: ``{resource: source}``, or ``{resource: {member_class: source}}`` when the member decides.') class-attribute instance-attribute

name = PydanticField(default='local_key', description='Canonical fallback property name on the class.') class-attribute instance-attribute

sep = PydanticField(default=':', description='Separator between tag and key.') class-attribute instance-attribute

sources property

The per-resource wiring, under the name :class:DerivedBranch uses.

Methods:

members_for(resource)

Member classes keying resource's sources, or None if unkeyed.

Source code in graflo/architecture/evolution/ops.py

def members_for(self, resource: str) -> list[str] | None:
    """Member classes keying *resource*'s sources, or ``None`` if unkeyed."""
    source = self.local_key.get(resource)
    return list(source) if isinstance(source, dict) else None

sources_for(resource)

Local-key sources resource contributes, in order.

Source code in graflo/architecture/evolution/ops.py

def sources_for(self, resource: str) -> list[LocalKeySource]:
    """Local-key sources *resource* contributes, in order."""
    source = self.local_key.get(resource)
    if source is None:
        return []
    return [source] if isinstance(source, LocalKeySource) else list(source.values())

LocalKeySource

Bases: ConfigBaseModel

Where one resource's side-local key comes from, and its namespace tag.

The tag is what keeps records of different sources apart once they fail to fuse: f2 from one source and f2 from another are different entities, and a:f2 / b:f2 say so. It is required so that opting out is a statement, not an omission: tag=None (stored as "", the neutral element, so it survives serialization) keeps the raw value as the local key with no separator — the author's claim that the values are already unique across every source of the class (UUIDs, IRIs, ids the source itself prefixes).

Source code in graflo/architecture/evolution/ops.py

class LocalKeySource(ConfigBaseModel):
    """Where one resource's side-local key comes from, and its namespace tag.

    The tag is what keeps records of different sources apart once they fail to
    fuse: ``f2`` from one source and ``f2`` from another are different
    entities, and ``a:f2`` / ``b:f2`` say so. It is required so that opting
    out is a statement, not an omission: ``tag=None`` (stored as ``""``, the
    neutral element, so it survives serialization) keeps the raw value as the
    local key with no separator — the author's claim that the values are
    already unique across every source of the class (UUIDs, IRIs, ids the
    source itself prefixes).
    """

    field: str = PydanticField(
        ...,
        description="RAW doc field carrying the side-local key.",
    )
    tag: str = PydanticField(
        ...,
        description=(
            "Namespace tag: tag 'a' turns 'f2' into 'a:f2'. ``None`` or ``\"\"`` "
            "keeps the raw value, no separator — only for values already "
            "unique across every source of the class."
        ),
    )
    when: TransformGuardConfig | None = PydanticField(
        default=None,
        description=(
            "Same as :attr:`DerivationSpec.when`: an explicit guard replacing "
            "the one the merge derives. Not allowed on a source keyed by member."
        ),
    )

    @field_validator("tag", mode="before")
    @classmethod
    def _none_is_the_empty_tag(cls, value: Any) -> Any:
        return "" if value is None else value

Attributes

field = PydanticField(..., description='RAW doc field carrying the side-local key.') class-attribute instance-attribute

tag = PydanticField(..., description='Namespace tag: tag \'a\' turns \'f2\' into \'a:f2\'. ``None`` or ``""`` keeps the raw value, no separator — only for values already unique across every source of the class.') class-attribute instance-attribute

when = PydanticField(default=None, description='Same as :attr:`DerivationSpec.when`: an explicit guard replacing the one the merge derives. Not allowed on a source keyed by member.') class-attribute instance-attribute

MergeEdgesOp

Bases: ConfigBaseModel

Merge source relation names into a canonical relation name.

Source code in graflo/architecture/evolution/ops.py

class MergeEdgesOp(ConfigBaseModel):
    """Merge source relation names into a canonical relation name."""

    op: Literal["merge_edges"] = "merge_edges"
    sources: list[str] = PydanticField(
        ...,
        description="Relation names to merge away. Must not include ``into``.",
        min_length=1,
    )
    into: str = PydanticField(
        ...,
        description="Canonical relation name that receives all source relations.",
    )

    @model_validator(mode="after")
    def _validate_sources(self) -> MergeEdgesOp:
        validate_merge_sources(self.sources, self.into, kind="merge_edges")
        return self

Attributes

into = PydanticField(..., description='Canonical relation name that receives all source relations.') class-attribute instance-attribute

op = 'merge_edges' class-attribute instance-attribute

sources = PydanticField(..., description='Relation names to merge away. Must not include ``into``.', min_length=1) class-attribute instance-attribute

MergeFieldTypes

Bases: ConfigBaseModel

The type of a merged property, declared on the merge by its merged names.

Keyed by the class or relation name and the property name the merged manifest carries. The merge retypes every member, on either side, that carries the property (under whatever spelling it is renamed from) before folding it, so members that disagree on a type merge into the declared one. Members that do not carry the property are left alone.

Source code in graflo/architecture/evolution/ops.py

class MergeFieldTypes(ConfigBaseModel):
    """The type of a merged property, declared on the merge by its merged names.

    Keyed by the class or relation name and the property name the **merged**
    manifest carries. The merge retypes every member, on either side, that
    carries the property (under whatever spelling it is renamed from) before
    folding it, so members that disagree on a type merge into the declared one.
    Members that do not carry the property are left alone.
    """

    vertices: dict[str, dict[str, FieldTypeSpec]] = PydanticField(
        default_factory=dict,
        description="``{merged_vertex: {merged_property: {type, item_type}}}``.",
    )
    edges: dict[str, dict[str, FieldTypeSpec]] = PydanticField(
        default_factory=dict,
        description="``{merged_relation: {property: {type, item_type}}}``.",
    )

    @model_validator(mode="after")
    def _require_a_target(self) -> MergeFieldTypes:
        if not self.vertices and not self.edges:
            raise ValueError("field_types requires at least one of vertices or edges")
        return self

Attributes

edges = PydanticField(default_factory=dict, description='``{merged_relation: {property: {type, item_type}}}``.') class-attribute instance-attribute

vertices = PydanticField(default_factory=dict, description='``{merged_vertex: {merged_property: {type, item_type}}}``.') class-attribute instance-attribute

MergeManifestsOp

Bases: ConfigBaseModel

Merge two full GraphManifests using explicit equivalence maps.

Binary only — apply via :func:~graflo.architecture.evolution.merge.merge_manifests. Unary :func:~graflo.architecture.evolution.apply.apply_evolution rejects this op.

Every name in the op is a name the input manifests declare. The vocabulary (canonical_maps), the equivalences and renames are resolved together, in one pass over those names, so nothing has to be written in an intermediate vocabulary. A group is everything an equivalence links, including every class a vocabulary merges with one of its members; its merged name is into, else the vocabulary's name, else the one spelling its members share. Empty equivalences yield a disjoint union, subject to name_conflict.

A vertex equivalence's identity with a derived or local_key branch is applied to the merged union before return (canonical attributes → resource derivations → priority funnel), then the members' own keys are demoted to secondary identities.

Source code in graflo/architecture/evolution/ops.py

class MergeManifestsOp(ConfigBaseModel):
    """Merge two full ``GraphManifest``s using explicit equivalence maps.

    Binary only — apply via :func:`~graflo.architecture.evolution.merge.merge_manifests`.
    Unary :func:`~graflo.architecture.evolution.apply.apply_evolution` rejects this op.

    Every name in the op is a name the input manifests declare. The
    vocabulary (``canonical_maps``), the equivalences and ``renames`` are
    resolved together, in one pass over those names, so nothing has to be
    written in an intermediate vocabulary. A group is everything an
    equivalence links, including every class a vocabulary merges with one of
    its members; its merged name is ``into``, else the vocabulary's name, else
    the one spelling its members share. Empty equivalences yield a disjoint
    union, subject to ``name_conflict``.

    A vertex equivalence's ``identity`` with a derived or ``local_key`` branch
    is applied to the merged union before return (canonical attributes →
    resource derivations → priority funnel), then the members' own keys are
    demoted to secondary identities.
    """

    op: Literal["merge_manifests"] = "merge_manifests"
    vertex_equivalences: list[VertexEquivalence] = PydanticField(
        default_factory=list,
        description="Vertex equivalences across the two input manifests.",
    )
    relation_equivalences: list[RelationEquivalence] = PydanticField(
        default_factory=list,
        description="Relation equivalences across the two input manifests.",
    )
    renames: MergeRenames = PydanticField(
        default_factory=MergeRenames,
        description=(
            "Per-side renames of classes, relations, attributes and resources "
            "that no equivalence groups."
        ),
    )
    name: str | None = PydanticField(
        default=None,
        description=(
            "Label for the merged manifest and its schema. Unset, the two "
            "sides' names are folded into ``left+right``."
        ),
    )
    target_namespace: str | None = PydanticField(
        default=None,
        description=(
            "Database / graph / space the merged schema deploys into. "
            "Supersedes both sides' ``db_profile.target_namespace`` (so it "
            "also resolves a disagreement between them) and is validated "
            "against the merged ``db_flavor``. Unset, the namespace is "
            "derived from the schema name when deployed."
        ),
    )
    name_conflict: Literal["error", "prefix_right", "union_right"] = PydanticField(
        default="error",
        description=(
            "How to handle a name both sides arrive at that no equivalence "
            "covers (vertices, relations, resources, connectors). ``error`` "
            "refuses and names the equivalences to declare; ``prefix_right`` "
            "keeps them apart under ``r_`` names; ``union_right`` unions "
            "vertices and relations of exactly the same name, each pair "
            "becoming a synthesized 1-1 equivalence, so identity and property "
            "reconciliation apply as to a declared one (resources and "
            "connectors are addresses, not concepts, so it behaves as "
            "``error`` for them). Two spellings of one concept "
            "(``OrderLine`` / ``order_line``) are never unioned: ``error`` and "
            "``union_right`` refuse them, ``prefix_right`` keeps them apart."
        ),
    )
    router_scope: Literal["side", "union"] = PydanticField(
        default="side",
        description=(
            "What a ``vertex_router`` may route a discriminator value missing "
            "from its ``type_map`` to, after the merge. ``side`` closes each "
            "router over its own side's classes: merge writes every class of "
            "that side into the table, under its merged name, and sets "
            "``type_map_only``, so a value the side never modeled is skipped "
            "as it was before the merge. ``union`` leaves routers open: such a "
            "value can name any class of the merged schema, the other side's "
            "included -- for sources that share type names and ids."
        ),
    )
    allow_dangling_entries: bool = PydanticField(
        default=False,
        description=(
            "Accept canonical map entries that name nothing on the side they "
            "are scoped to, dropping and logging each one instead of refusing "
            "with the list. Set it on a map itself to say the map is broader "
            "than this merge; set it here to say so for both maps at once."
        ),
    )
    canonical_maps: dict[Literal["left", "right", "both"], CanonicalMap] = (
        PydanticField(
            default_factory=dict,
            description=(
                "The vocabulary per side. ``left`` / ``right`` apply to that "
                "manifest's own names, ``both`` to either. A vocabulary is the "
                "default name of a class; an equivalence's ``into`` overrides "
                "it for the whole group."
            ),
        )
    )
    field_types: MergeFieldTypes | None = PydanticField(
        default=None,
        description=(
            "Merged property types, keyed by merged names: every member that "
            "carries the property is retyped before the fold, so members that "
            "disagree on a type merge into this one."
        ),
    )

    @model_validator(mode="before")
    @classmethod
    def _refuse_removed_keys(cls, data: Any) -> Any:
        if not isinstance(data, Mapping):
            return data
        removed = sorted(key for key in data if key in _REMOVED_MERGE_KEYS)
        if data.get("name_conflict") == "fuse_right":
            raise ValueError(
                "merge_manifests: name_conflict 'fuse_right' was removed; spell "
                "it 'union_right'"
            )
        if removed:
            raise ValueError(
                "merge_manifests: "
                + "; ".join(
                    f"`{key}` was removed: {_REMOVED_MERGE_KEYS[key]}"
                    for key in removed
                )
            )
        return data

Attributes

allow_dangling_entries = PydanticField(default=False, description='Accept canonical map entries that name nothing on the side they are scoped to, dropping and logging each one instead of refusing with the list. Set it on a map itself to say the map is broader than this merge; set it here to say so for both maps at once.') class-attribute instance-attribute

canonical_maps = PydanticField(default_factory=dict, description="The vocabulary per side. ``left`` / ``right`` apply to that manifest's own names, ``both`` to either. A vocabulary is the default name of a class; an equivalence's ``into`` overrides it for the whole group.") class-attribute instance-attribute

field_types = PydanticField(default=None, description='Merged property types, keyed by merged names: every member that carries the property is retyped before the fold, so members that disagree on a type merge into this one.') class-attribute instance-attribute

name = PydanticField(default=None, description="Label for the merged manifest and its schema. Unset, the two sides' names are folded into ``left+right``.") class-attribute instance-attribute

name_conflict = PydanticField(default='error', description='How to handle a name both sides arrive at that no equivalence covers (vertices, relations, resources, connectors). ``error`` refuses and names the equivalences to declare; ``prefix_right`` keeps them apart under ``r_`` names; ``union_right`` unions vertices and relations of exactly the same name, each pair becoming a synthesized 1-1 equivalence, so identity and property reconciliation apply as to a declared one (resources and connectors are addresses, not concepts, so it behaves as ``error`` for them). Two spellings of one concept (``OrderLine`` / ``order_line``) are never unioned: ``error`` and ``union_right`` refuse them, ``prefix_right`` keeps them apart.') class-attribute instance-attribute

op = 'merge_manifests' class-attribute instance-attribute

relation_equivalences = PydanticField(default_factory=list, description='Relation equivalences across the two input manifests.') class-attribute instance-attribute

renames = PydanticField(default_factory=MergeRenames, description='Per-side renames of classes, relations, attributes and resources that no equivalence groups.') class-attribute instance-attribute

router_scope = PydanticField(default='side', description="What a ``vertex_router`` may route a discriminator value missing from its ``type_map`` to, after the merge. ``side`` closes each router over its own side's classes: merge writes every class of that side into the table, under its merged name, and sets ``type_map_only``, so a value the side never modeled is skipped as it was before the merge. ``union`` leaves routers open: such a value can name any class of the merged schema, the other side's included -- for sources that share type names and ids.") class-attribute instance-attribute

target_namespace = PydanticField(default=None, description="Database / graph / space the merged schema deploys into. Supersedes both sides' ``db_profile.target_namespace`` (so it also resolves a disagreement between them) and is validated against the merged ``db_flavor``. Unset, the namespace is derived from the schema name when deployed.") class-attribute instance-attribute

vertex_equivalences = PydanticField(default_factory=list, description='Vertex equivalences across the two input manifests.') class-attribute instance-attribute

MergeRenames

Bases: ConfigBaseModel

Per-side renames of a merge, in each side's own names.

Source code in graflo/architecture/evolution/ops.py

class MergeRenames(ConfigBaseModel):
    """Per-side renames of a merge, in each side's own names."""

    left: SideRenames = PydanticField(default_factory=SideRenames)
    right: SideRenames = PydanticField(default_factory=SideRenames)

    def __getitem__(self, side: str) -> SideRenames:
        return self.left if side == "left" else self.right

Attributes

left = PydanticField(default_factory=SideRenames) class-attribute instance-attribute

right = PydanticField(default_factory=SideRenames) class-attribute instance-attribute

Methods:

__getitem__(side)

Source code in graflo/architecture/evolution/ops.py

def __getitem__(self, side: str) -> SideRenames:
    return self.left if side == "left" else self.right

MergeVerticesOp

Bases: ConfigBaseModel

Merge source vertices into a single logical name (schema, edges, ingestion).

Source code in graflo/architecture/evolution/ops.py

class MergeVerticesOp(ConfigBaseModel):
    """Merge source vertices into a single logical name (schema, edges, ingestion)."""

    op: Literal["merge_vertices"] = "merge_vertices"
    sources: list[str] = PydanticField(
        ...,
        description=(
            "Vertex type names to merge away. Must not include ``into``. "
            "Each name must exist in the schema before the merge."
        ),
        min_length=1,
    )
    into: str = PydanticField(
        ...,
        description=(
            "Resulting vertex type name. If it already exists, source vertices are "
            "merged into it. If it does not exist, a new vertex is built from all sources."
        ),
    )
    allow_self_relations: bool = PydanticField(
        default=False,
        description=(
            "Accept edges whose endpoints both land on ``into``. A self-relation makes "
            "both endpoints share one accumulator slot, so assembly merges observations "
            "that were previously separate nodes. Rejected unless set."
        ),
    )
    allow_observation_fusion: bool = PydanticField(
        default=False,
        validation_alias=AliasChoices("allow_observation_fusion", "allow_row_fusion"),
        description=(
            "Accept resource pipelines whose sources land in one accumulator slot — "
            "the same level and the same ``role``, or both bare. Those steps then "
            "write to one slot, fusing into one node what a single source document "
            "emitted as two vertex observations. Same-level steps with distinct "
            "``role``s never share a slot and need no acknowledgement. Rejected "
            "unless set. ``allow_row_fusion`` is accepted as a legacy alias."
        ),
    )

    @model_validator(mode="after")
    def _validate_sources(self) -> MergeVerticesOp:
        validate_merge_sources(self.sources, self.into, kind="merge_vertices")
        return self

Attributes

allow_observation_fusion = PydanticField(default=False, validation_alias=AliasChoices('allow_observation_fusion', 'allow_row_fusion'), description='Accept resource pipelines whose sources land in one accumulator slot — the same level and the same ``role``, or both bare. Those steps then write to one slot, fusing into one node what a single source document emitted as two vertex observations. Same-level steps with distinct ``role``s never share a slot and need no acknowledgement. Rejected unless set. ``allow_row_fusion`` is accepted as a legacy alias.') class-attribute instance-attribute

allow_self_relations = PydanticField(default=False, description='Accept edges whose endpoints both land on ``into``. A self-relation makes both endpoints share one accumulator slot, so assembly merges observations that were previously separate nodes. Rejected unless set.') class-attribute instance-attribute

into = PydanticField(..., description='Resulting vertex type name. If it already exists, source vertices are merged into it. If it does not exist, a new vertex is built from all sources.') class-attribute instance-attribute

op = 'merge_vertices' class-attribute instance-attribute

sources = PydanticField(..., description='Vertex type names to merge away. Must not include ``into``. Each name must exist in the schema before the merge.', min_length=1) class-attribute instance-attribute

NaturalIdentityTarget

Bases: ConfigBaseModel

Target a natural key: the named properties identify the vertex directly.

Source code in graflo/architecture/evolution/ops.py

class NaturalIdentityTarget(ConfigBaseModel):
    """Target a natural key: the named properties identify the vertex directly."""

    mode: Literal["natural"] = "natural"
    identity: list[str] = PydanticField(
        ...,
        description="Property names forming the new primary identity.",
        min_length=1,
    )

Attributes

identity = PydanticField(..., description='Property names forming the new primary identity.', min_length=1) class-attribute instance-attribute

mode = 'natural' class-attribute instance-attribute

ProjectManifestOp

Bases: ConfigBaseModel

Project a manifest to a vertex/edge subgraph with consistent cascade.

Keeps only the requested logical vertices and edges (and optionally resources). All schema, db_profile, ingestion, and bindings references to removed entities are pruned. Inverse edges are not kept by default; list them in keep_edges, or set keep_inverse_edges to keep the declared mirror of every edge that is kept.

With connectivity="induced_prune" (v1 default), when keep_vertices is set, vertex types from that list with no incident surviving edge are dropped.

depth turns keep_vertices from a literal list into seeds for a neighbourhood walk. One rule covers every combination: let E be keep_edges when given and every declared edge otherwise; the survivors are the vertex types within depth hops of a seed along E under direction, and then E restricted to surviving endpoints. So the result is the induced subgraph on the hop ball — an edge between two neighbours survives even though no walk needed it — and keep_edges bounds the walk rather than being overridden by it. Pruning is unchanged: a seed left with no surviving edge is still dropped, whatever the depth.

Edge.by (the third vertex type on an EdgeType.INDIRECT edge) is not part of schema adjacency, so a walk never pulls it in — the same blind spot the flat selection already has.

Source code in graflo/architecture/evolution/ops.py

class ProjectManifestOp(ConfigBaseModel):
    """Project a manifest to a vertex/edge subgraph with consistent cascade.

    Keeps only the requested logical vertices and edges (and optionally resources).
    All schema, ``db_profile``, ingestion, and bindings references to removed
    entities are pruned. Inverse edges are **not** kept by default; list them in
    ``keep_edges``, or set ``keep_inverse_edges`` to keep the declared mirror of
    every edge that is kept.

    With ``connectivity=\"induced_prune\"`` (v1 default), when ``keep_vertices`` is
    set, vertex types from that list with no incident surviving edge are dropped.

    ``depth`` turns ``keep_vertices`` from a literal list into seeds for a
    neighbourhood walk. One rule covers every combination: let ``E`` be
    ``keep_edges`` when given and every declared edge otherwise; the survivors are
    the vertex types within ``depth`` hops of a seed along ``E`` under
    ``direction``, and then ``E`` restricted to surviving endpoints. So the result
    is the *induced* subgraph on the hop ball — an edge between two neighbours
    survives even though no walk needed it — and ``keep_edges`` bounds the walk
    rather than being overridden by it. Pruning is unchanged: a seed left with no
    surviving edge is still dropped, whatever the depth.

    ``Edge.by`` (the third vertex type on an ``EdgeType.INDIRECT`` edge) is not part
    of schema adjacency, so a walk never pulls it in — the same blind spot the flat
    selection already has.
    """

    op: Literal["project_manifest"] = "project_manifest"
    keep_vertices: list[str] | None = PydanticField(
        default=None,
        description="Vertex type names to retain (after induced connectivity pruning).",
    )
    keep_edges: list[EdgeSelector] | None = PydanticField(
        default=None,
        description="Edge triples ``(source, target, relation)`` to retain.",
    )
    connectivity: Literal["induced_prune"] = PydanticField(
        default="induced_prune",
        description="How to interpret ``keep_vertices`` relative to surviving edges.",
    )
    depth: int = PydanticField(
        default=0,
        ge=0,
        description=(
            "Hops to expand ``keep_vertices`` by before the induced slice. "
            "``0`` (default) keeps the literal list."
        ),
    )
    direction: EdgeDirection = PydanticField(
        default=EdgeDirection.ANY,
        description=(
            "Orientation followed when expanding by ``depth``. Edges declared "
            "``directed: false`` are followed both ways regardless. Ignored when "
            "``depth`` is 0."
        ),
    )
    keep_resources: list[str] | None = PydanticField(
        default=None,
        description="Optional ingestion resource names to retain after graph slice.",
    )
    keep_inverse_edges: bool = PydanticField(
        default=False,
        description=(
            "With ``keep_edges``: also keep the declared mirror ``(T, S, inv)`` of "
            "each kept ``(S, T, r)``, so a materialized pair survives as a pair. "
            "Without ``keep_edges`` every edge between surviving vertices is kept "
            "already."
        ),
    )
    strict: bool = PydanticField(
        default=True,
        description="When True, unknown vertex/edge selectors raise ``ValueError``.",
    )
    partial_resources: Literal["trim", "drop"] = PydanticField(
        default="trim",
        description=(
            "A resource whose pipeline the projection shortens: ``trim`` keeps "
            "what survives and logs a warning naming each one; ``drop`` removes "
            "it with the bindings that served it, keeping only resources the "
            "projection leaves whole."
        ),
    )

    @model_validator(mode="after")
    def _validate_projection_selectors(self) -> ProjectManifestOp:
        if not self.keep_vertices and not self.keep_edges:
            raise ValueError(
                "project_manifest requires at least one of keep_vertices or keep_edges"
            )
        if self.keep_vertices and len(self.keep_vertices) != len(
            set(self.keep_vertices)
        ):
            raise ValueError("keep_vertices entries must be unique")
        if self.keep_edges:
            edge_ids = [selector.edge_id() for selector in self.keep_edges]
            if len(edge_ids) != len(set(edge_ids)):
                raise ValueError(
                    "keep_edges entries must be unique by (source, target, relation)"
                )
        if self.depth > 0 and not self.keep_vertices:
            # Without seeds `keep_vertices=None` already means "every vertex type",
            # so there is nothing for a walk to expand — the request is a mistake
            # rather than a no-op, and saying so beats silently ignoring `depth`.
            raise ValueError("project_manifest: depth > 0 requires keep_vertices")
        return self

Attributes

connectivity = PydanticField(default='induced_prune', description='How to interpret ``keep_vertices`` relative to surviving edges.') class-attribute instance-attribute

depth = PydanticField(default=0, ge=0, description='Hops to expand ``keep_vertices`` by before the induced slice. ``0`` (default) keeps the literal list.') class-attribute instance-attribute

direction = PydanticField(default=EdgeDirection.ANY, description='Orientation followed when expanding by ``depth``. Edges declared ``directed: false`` are followed both ways regardless. Ignored when ``depth`` is 0.') class-attribute instance-attribute

keep_edges = PydanticField(default=None, description='Edge triples ``(source, target, relation)`` to retain.') class-attribute instance-attribute

keep_inverse_edges = PydanticField(default=False, description='With ``keep_edges``: also keep the declared mirror ``(T, S, inv)`` of each kept ``(S, T, r)``, so a materialized pair survives as a pair. Without ``keep_edges`` every edge between surviving vertices is kept already.') class-attribute instance-attribute

keep_resources = PydanticField(default=None, description='Optional ingestion resource names to retain after graph slice.') class-attribute instance-attribute

keep_vertices = PydanticField(default=None, description='Vertex type names to retain (after induced connectivity pruning).') class-attribute instance-attribute

op = 'project_manifest' class-attribute instance-attribute

partial_resources = PydanticField(default='trim', description='A resource whose pipeline the projection shortens: ``trim`` keeps what survives and logs a warning naming each one; ``drop`` removes it with the bindings that served it, keeping only resources the projection leaves whole.') class-attribute instance-attribute

strict = PydanticField(default=True, description='When True, unknown vertex/edge selectors raise ``ValueError``.') class-attribute instance-attribute

PropertyEquivalence

Bases: ConfigBaseModel

Align a property from the left and/or right member(s) onto a canonical name.

At least one of left / right must be set. A bare string applies to every member declared on that side of the owning :class:VertexEquivalence; a {member: field} dict maps per member, for when members are not aligned under the same source field name.

Exact-name matches do not need a :class:PropertyEquivalence: after boundary rename, merge_vertex_models unions fields by spelling, so a property present under the same name on every member fuses for free. Declare an equivalence only to rename or to pick a different into; the merged key is declared on the :class:VertexEquivalence, the merged type in :attr:MergeManifestsOp.field_types.

Source code in graflo/architecture/evolution/ops.py

class PropertyEquivalence(ConfigBaseModel):
    """Align a property from the left and/or right member(s) onto a canonical name.

    At least one of ``left`` / ``right`` must be set. A bare string applies to
    every member declared on that side of the owning
    :class:`VertexEquivalence`; a ``{member: field}`` dict maps per member,
    for when members are not aligned under the same source field name.

    Exact-name matches do **not** need a :class:`PropertyEquivalence`: after
    boundary rename, ``merge_vertex_models`` unions fields by spelling, so a
    property present under the same name on every member fuses for free.
    Declare an equivalence only to rename or to pick a different ``into``; the
    merged key is declared on the :class:`VertexEquivalence`, the merged type
    in :attr:`MergeManifestsOp.field_types`.
    """

    left: str | dict[str, str] | None = PydanticField(
        default=None,
        description=(
            "Field name on the left member(s): a bare string applies to every "
            "left member of the owning equivalence, a ``{member: field}`` dict "
            "maps per member."
        ),
    )
    right: str | dict[str, str] | None = PydanticField(
        default=None,
        description="Same shape as ``left``, for the right member(s).",
    )
    into: str = PydanticField(
        ...,
        description="Canonical property name on the merged vertex.",
    )

    @model_validator(mode="after")
    def _require_side(self) -> PropertyEquivalence:
        if self.left is None and self.right is None:
            raise ValueError(
                "PropertyEquivalence requires at least one of left or right"
            )
        for side, spec in (("left", self.left), ("right", self.right)):
            if isinstance(spec, dict) and not spec:
                raise ValueError(
                    f"PropertyEquivalence: {side} is an empty per-member map"
                )
        return self

Attributes

into = PydanticField(..., description='Canonical property name on the merged vertex.') class-attribute instance-attribute

left = PydanticField(default=None, description='Field name on the left member(s): a bare string applies to every left member of the owning equivalence, a ``{member: field}`` dict maps per member.') class-attribute instance-attribute

right = PydanticField(default=None, description='Same shape as ``left``, for the right member(s).') class-attribute instance-attribute

RelationEquivalence

Bases: ConfigBaseModel

Collapse one or more left relations and one or more right relations onto one name.

Shares the left / right n-ary shape and the naming rules of :class:VertexEquivalence.

Source code in graflo/architecture/evolution/ops.py

class RelationEquivalence(ConfigBaseModel):
    """Collapse one or more left relations and one or more right relations onto one name.

    Shares the ``left`` / ``right`` n-ary shape and the naming rules of
    :class:`VertexEquivalence`.
    """

    left: str | list[str] = PydanticField(
        ..., description="One or more left relation names."
    )
    right: str | list[str] = PydanticField(
        ..., description="One or more right relation names."
    )
    into: str | None = PydanticField(
        default=None,
        description=(
            "Merged relation name, never translated by a vocabulary. Omitted, "
            "the name comes from the vocabulary, or from the one spelling "
            "every member shares."
        ),
    )

    @property
    def left_members(self) -> list[str]:
        return _member_list(self.left)

    @property
    def right_members(self) -> list[str]:
        return _member_list(self.right)

    def members(self, side: Literal["left", "right"]) -> list[str]:
        return self.left_members if side == "left" else self.right_members

    @model_validator(mode="after")
    def _validate_members(self) -> RelationEquivalence:
        for side, members in (
            ("left", self.left_members),
            ("right", self.right_members),
        ):
            if not members:
                raise ValueError(
                    f"RelationEquivalence: {side} must name at least one relation"
                )
            if len(members) != len(set(members)):
                raise ValueError(
                    f"RelationEquivalence: {side} lists a relation more than once: {members}"
                )
        return self

Attributes

into = PydanticField(default=None, description='Merged relation name, never translated by a vocabulary. Omitted, the name comes from the vocabulary, or from the one spelling every member shares.') class-attribute instance-attribute

left = PydanticField(..., description='One or more left relation names.') class-attribute instance-attribute

left_members property

right = PydanticField(..., description='One or more right relation names.') class-attribute instance-attribute

right_members property

Methods:

members(side)

Source code in graflo/architecture/evolution/ops.py

def members(self, side: Literal["left", "right"]) -> list[str]:
    return self.left_members if side == "left" else self.right_members

RemoveEdgeIndexesOp

Bases: ConfigBaseModel

Withdraw authored indexes from edge physical specs, addressed by field list.

Source code in graflo/architecture/evolution/ops.py

class RemoveEdgeIndexesOp(ConfigBaseModel):
    """Withdraw authored indexes from edge physical specs, addressed by field list."""

    op: Literal["remove_edge_indexes"] = "remove_edge_indexes"
    edges: list[EdgeIndexEntry] = PydanticField(
        ...,
        description="Per-spec indexes to remove (``fields`` on each entry).",
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_entries(self) -> RemoveEdgeIndexesOp:
        _validate_edge_index_entries(
            self.edges, kind="remove_edge_indexes", carries="fields"
        )
        return self

Attributes

edges = PydanticField(..., description='Per-spec indexes to remove (``fields`` on each entry).', min_length=1) class-attribute instance-attribute

op = 'remove_edge_indexes' class-attribute instance-attribute

RemoveEdgePropertiesOp

Bases: ConfigBaseModel

Remove edge properties for each relation across schema/profile/ingestion.

Source code in graflo/architecture/evolution/ops.py

class RemoveEdgePropertiesOp(ConfigBaseModel):
    """Remove edge properties for each relation across schema/profile/ingestion."""

    op: Literal["remove_edge_properties"] = "remove_edge_properties"
    removals: dict[str, list[str]] = PydanticField(
        ...,
        description=(
            "Per-relation edge field removals: ``{relation_name: [field_name, ...]}``."
        ),
        min_length=1,
    )

Attributes

op = 'remove_edge_properties' class-attribute instance-attribute

removals = PydanticField(..., description='Per-relation edge field removals: ``{relation_name: [field_name, ...]}``.', min_length=1) class-attribute instance-attribute

RemoveEdgesOp

Bases: ConfigBaseModel

Remove logical edges from schema, profile, and ingestion selectors.

Two addressing forms, combinable in one op. relations removes a relation on every endpoint pair it occurs on. edges removes exactly the named triples, which is the only way to remove one pair of several sharing a relation, or an edge with no relation set at all.

Source code in graflo/architecture/evolution/ops.py

class RemoveEdgesOp(ConfigBaseModel):
    """Remove logical edges from schema, profile, and ingestion selectors.

    Two addressing forms, combinable in one op. ``relations`` removes a relation
    on every endpoint pair it occurs on. ``edges`` removes exactly the named
    triples, which is the only way to remove one pair of several sharing a
    relation, or an edge with no relation set at all.
    """

    op: Literal["remove_edges"] = "remove_edges"
    relations: list[str] = PydanticField(
        default_factory=list,
        description="Relation names to remove from edge definitions and references.",
    )
    edges: list[EdgeSelector] = PydanticField(
        default_factory=list,
        description="Edge triples ``(source, target, relation)`` to remove.",
    )

    @model_validator(mode="after")
    def _validate_targets(self) -> RemoveEdgesOp:
        if not self.relations and not self.edges:
            raise ValueError("remove_edges requires at least one of relations or edges")
        _validate_unique_edge_selectors(self.edges, kind="remove_edges")
        return self

Attributes

edges = PydanticField(default_factory=list, description='Edge triples ``(source, target, relation)`` to remove.') class-attribute instance-attribute

op = 'remove_edges' class-attribute instance-attribute

relations = PydanticField(default_factory=list, description='Relation names to remove from edge definitions and references.') class-attribute instance-attribute

RemoveResourcesOp

Bases: ConfigBaseModel

Remove ingestion resources and the bindings that wired them.

Source code in graflo/architecture/evolution/ops.py

class RemoveResourcesOp(ConfigBaseModel):
    """Remove ingestion resources and the bindings that wired them."""

    op: Literal["remove_resources"] = "remove_resources"
    names: list[str] = PydanticField(
        ...,
        description="Resource names to remove.",
        min_length=1,
    )

Attributes

names = PydanticField(..., description='Resource names to remove.', min_length=1) class-attribute instance-attribute

op = 'remove_resources' class-attribute instance-attribute

RemoveSecondaryIdentitiesOp

Bases: ConfigBaseModel

Withdraw alternate lookup keys, dropping their derived indexes.

Rejected when a surviving edge step still selects the removed field-set — that step would have no way to resolve its endpoint.

Source code in graflo/architecture/evolution/ops.py

class RemoveSecondaryIdentitiesOp(ConfigBaseModel):
    """Withdraw alternate lookup keys, dropping their derived indexes.

    Rejected when a surviving edge step still selects the removed field-set — that
    step would have no way to resolve its endpoint.
    """

    op: Literal["remove_secondary_identities"] = "remove_secondary_identities"
    removals: dict[str, list[str | list[str]]] = PydanticField(
        ...,
        description=(
            "Per-vertex secondary identities to withdraw, addressed by name or by "
            "field list: ``{vertex_name: [name | [field, ...], ...]}``."
        ),
        min_length=1,
    )

Attributes

op = 'remove_secondary_identities' class-attribute instance-attribute

removals = PydanticField(..., description='Per-vertex secondary identities to withdraw, addressed by name or by field list: ``{vertex_name: [name | [field, ...], ...]}``.', min_length=1) class-attribute instance-attribute

RemoveVertexIndexesOp

Bases: ConfigBaseModel

Withdraw authored vertex indexes, addressed by field list.

Indexes derived from secondary_identities are not removable here — they would be re-registered by the next finish_init. Use :class:RemoveSecondaryIdentitiesOp for those.

Source code in graflo/architecture/evolution/ops.py

class RemoveVertexIndexesOp(ConfigBaseModel):
    """Withdraw authored vertex indexes, addressed by field list.

    Indexes derived from ``secondary_identities`` are not removable here — they would
    be re-registered by the next ``finish_init``. Use
    :class:`RemoveSecondaryIdentitiesOp` for those.
    """

    op: Literal["remove_vertex_indexes"] = "remove_vertex_indexes"
    indexes: dict[
        str,
        Annotated[
            list[Annotated[list[str], PydanticField(min_length=1)]],
            PydanticField(min_length=1),
        ],
    ] = PydanticField(
        ...,
        description="``{vertex_name: [[field, ...], ...]}``.",
        min_length=1,
    )

Attributes

indexes = PydanticField(..., description='``{vertex_name: [[field, ...], ...]}``.', min_length=1) class-attribute instance-attribute

op = 'remove_vertex_indexes' class-attribute instance-attribute

RemoveVertexPropertiesOp

Bases: ConfigBaseModel

Remove vertex properties and propagate pruning to ingestion/db profile references.

Source code in graflo/architecture/evolution/ops.py

class RemoveVertexPropertiesOp(ConfigBaseModel):
    """Remove vertex properties and propagate pruning to ingestion/db profile references."""

    op: Literal["remove_vertex_properties"] = "remove_vertex_properties"
    removals: dict[str, list[str]] = PydanticField(
        ...,
        description=(
            "Per-vertex field removal map: ``{vertex_name: [field_name, ...]}``."
        ),
        min_length=1,
    )

Attributes

op = 'remove_vertex_properties' class-attribute instance-attribute

removals = PydanticField(..., description='Per-vertex field removal map: ``{vertex_name: [field_name, ...]}``.', min_length=1) class-attribute instance-attribute

RemoveVerticesOp

Bases: ConfigBaseModel

Remove logical vertices and cascade: edges, ingestion resources, bindings.

Source code in graflo/architecture/evolution/ops.py

class RemoveVerticesOp(ConfigBaseModel):
    """Remove logical vertices and cascade: edges, ingestion resources, bindings."""

    op: Literal["remove_vertices"] = "remove_vertices"
    names: list[str] = PydanticField(
        ...,
        description="Vertex type names to remove from the schema.",
        min_length=1,
    )

Attributes

names = PydanticField(..., description='Vertex type names to remove from the schema.', min_length=1) class-attribute instance-attribute

op = 'remove_vertices' class-attribute instance-attribute

RenameEdgePropertiesOp

Bases: ConfigBaseModel

Rename edge properties for each relation across schema/profile/ingestion.

Source code in graflo/architecture/evolution/ops.py

class RenameEdgePropertiesOp(ConfigBaseModel):
    """Rename edge properties for each relation across schema/profile/ingestion."""

    op: Literal["rename_edge_properties"] = "rename_edge_properties"
    renames: dict[str, dict[str, str]] = PydanticField(
        ...,
        description=(
            "Per-relation edge field rename map: "
            "``{relation_name: {old_field: new_field}}``."
        ),
        min_length=1,
    )

Attributes

op = 'rename_edge_properties' class-attribute instance-attribute

renames = PydanticField(..., description='Per-relation edge field rename map: ``{relation_name: {old_field: new_field}}``.', min_length=1) class-attribute instance-attribute

RenameRelationsOp

Bases: ConfigBaseModel

Rename logical edge relation names across schema and ingestion.

Source code in graflo/architecture/evolution/ops.py

class RenameRelationsOp(ConfigBaseModel):
    """Rename logical edge relation names across schema and ingestion."""

    op: Literal["rename_relations"] = "rename_relations"
    renames: dict[str, str] = PydanticField(
        ...,
        validation_alias=AliasChoices("renames", "relations"),
        description=(
            "Relation rename map: ``{old_relation: new_relation}``. Must be "
            "injective. ``relations`` is accepted as a legacy alias."
        ),
        min_length=1,
    )

    @model_validator(mode="after")
    def _reject_collapsing_map(self) -> RenameRelationsOp:
        validate_rename_map_is_injective(
            self.renames,
            kind="rename_relations",
            merge_hint="MergeEdgesOp(sources=[...], into=...)",
        )
        return self

Attributes

op = 'rename_relations' class-attribute instance-attribute

renames = PydanticField(..., validation_alias=AliasChoices('renames', 'relations'), description='Relation rename map: ``{old_relation: new_relation}``. Must be injective. ``relations`` is accepted as a legacy alias.', min_length=1) class-attribute instance-attribute

RenameResourcesOp

Bases: ConfigBaseModel

Rename ingestion resource names and bindings references.

Source code in graflo/architecture/evolution/ops.py

class RenameResourcesOp(ConfigBaseModel):
    """Rename ingestion resource names and bindings references."""

    op: Literal["rename_resources"] = "rename_resources"
    renames: dict[str, str] = PydanticField(
        ...,
        validation_alias=AliasChoices("renames", "resources"),
        description=(
            "Ingestion resource rename map: ``{old_resource: new_resource}``. Must "
            "be injective. ``resources`` is accepted as a legacy alias."
        ),
        min_length=1,
    )

    @model_validator(mode="after")
    def _reject_collapsing_map(self) -> RenameResourcesOp:
        # IngestionModel already rejects duplicate resource names, so a collapsing
        # map fails downstream anyway — but with a message about the model rather
        # than about the op the author actually wrote.
        validate_rename_map_is_injective(
            self.renames,
            kind="rename_resources",
            merge_hint="MergeManifestsOp renames.<side>.resources",
        )
        return self

Attributes

op = 'rename_resources' class-attribute instance-attribute

renames = PydanticField(..., validation_alias=AliasChoices('renames', 'resources'), description='Ingestion resource rename map: ``{old_resource: new_resource}``. Must be injective. ``resources`` is accepted as a legacy alias.', min_length=1) class-attribute instance-attribute

RenameVertexPropertiesOp

Bases: ConfigBaseModel

Rename vertex properties (and identity references) and propagate to ingestion.

renames maps each vertex name to a per-vertex {old_field: new_field} map. Schema-side: rewrites Field.name, vertex.identity, and any DB profile structures that reference field names (vertex_indexes, edge_specs.indexes). Ingestion-side: rewrites VertexActor.from so the doc still uses the OLD field name (injecting {new_field: old_field} when missing), rewrites TransformActor.rename values that target a renamed vertex field, and updates Resource.extra_weights / edge.vertex_weights (:class:~graflo.architecture.graph_types.Weight fields, map, and filter keys that address vertex observation columns).

Source code in graflo/architecture/evolution/ops.py

class RenameVertexPropertiesOp(ConfigBaseModel):
    """Rename vertex properties (and identity references) and propagate to ingestion.

    ``renames`` maps each vertex name to a per-vertex ``{old_field: new_field}`` map.
    Schema-side: rewrites ``Field.name``, ``vertex.identity``, and any DB profile
    structures that reference field names (``vertex_indexes``, ``edge_specs.indexes``).
    Ingestion-side: rewrites ``VertexActor.from`` so the doc still uses the OLD field
    name (injecting ``{new_field: old_field}`` when missing), rewrites
    ``TransformActor.rename`` values that target a renamed vertex field, and updates
    ``Resource.extra_weights`` / ``edge.vertex_weights`` (:class:`~graflo.architecture.graph_types.Weight`
    ``fields``, ``map``, and ``filter`` keys that address vertex observation columns).
    """

    op: Literal["rename_vertex_properties"] = "rename_vertex_properties"
    renames: dict[str, dict[str, str]] = PydanticField(
        ...,
        description=(
            "Per-vertex field rename map: ``{vertex_name: {old_field: new_field}}``."
        ),
        min_length=1,
    )

Attributes

op = 'rename_vertex_properties' class-attribute instance-attribute

renames = PydanticField(..., description='Per-vertex field rename map: ``{vertex_name: {old_field: new_field}}``.', min_length=1) class-attribute instance-attribute

RenameVerticesOp

Bases: ConfigBaseModel

Rename logical vertex names across schema, ingestion, and bindings.

Source code in graflo/architecture/evolution/ops.py

class RenameVerticesOp(ConfigBaseModel):
    """Rename logical vertex names across schema, ingestion, and bindings."""

    op: Literal["rename_vertices"] = "rename_vertices"
    renames: dict[str, str] = PydanticField(
        ...,
        validation_alias=AliasChoices("renames", "vertices"),
        description=(
            "Vertex rename map: ``{old_vertex: new_vertex}``. Must be injective. "
            "``vertices`` is accepted as a legacy alias."
        ),
        min_length=1,
    )

    @model_validator(mode="after")
    def _reject_collapsing_map(self) -> RenameVerticesOp:
        validate_rename_map_is_injective(
            self.renames,
            kind="rename_vertices",
            merge_hint="MergeVerticesOp(sources=[...], into=...)",
        )
        return self

Attributes

op = 'rename_vertices' class-attribute instance-attribute

renames = PydanticField(..., validation_alias=AliasChoices('renames', 'vertices'), description='Vertex rename map: ``{old_vertex: new_vertex}``. Must be injective. ``vertices`` is accepted as a legacy alias.', min_length=1) class-attribute instance-attribute

ReplaceEdgeIdentitiesOp

Bases: ConfigBaseModel

Replace the uniqueness keys of logical edges.

The edge-side counterpart of :class:ReplaceIdentityOp. There is no retire policy: edge identities have no lookup plane to demote into. Non-endpoint tokens are merged into edge properties by Edge.finish_init, as with authored identities.

Source code in graflo/architecture/evolution/ops.py

class ReplaceEdgeIdentitiesOp(ConfigBaseModel):
    """Replace the uniqueness keys of logical edges.

    The edge-side counterpart of :class:`ReplaceIdentityOp`. There is no retire policy:
    edge identities have no lookup plane to demote into. Non-endpoint tokens are merged
    into edge ``properties`` by ``Edge.finish_init``, as with authored identities.
    """

    op: Literal["replace_edge_identities"] = "replace_edge_identities"
    edges: list[EdgeIdentitiesEntry] = PydanticField(
        ...,
        description="Per-edge replacement uniqueness keys.",
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_unique_selectors(self) -> ReplaceEdgeIdentitiesOp:
        edge_ids = [entry.edge_id() for entry in self.edges]
        if len(edge_ids) != len(set(edge_ids)):
            raise ValueError(
                "replace_edge_identities entries must be unique by "
                "(source, target, relation)"
            )
        return self

Attributes

edges = PydanticField(..., description='Per-edge replacement uniqueness keys.', min_length=1) class-attribute instance-attribute

op = 'replace_edge_identities' class-attribute instance-attribute

ReplaceIdentityOp

Bases: ConfigBaseModel

Replace the identity policy of one or more vertices.

Covers both a change of identity fields and a change of identity mode (natural / hash / assigned / blank), because the cascade is the same in either case: the field-set that upserts changes, and everything that referenced the old one must be repointed or retired.

Not covered: blank vertices cannot retire by demotion (they cannot declare secondary identities at all).

Source code in graflo/architecture/evolution/ops.py

class ReplaceIdentityOp(ConfigBaseModel):
    """Replace the identity policy of one or more vertices.

    Covers both a change of identity *fields* and a change of identity *mode*
    (``natural`` / ``hash`` / ``assigned`` / ``blank``), because the cascade is the
    same in either case: the field-set that upserts changes, and everything that
    referenced the old one must be repointed or retired.

    Not covered: ``blank`` vertices cannot retire by demotion (they cannot declare
    secondary identities at all).
    """

    op: Literal["replace_identity"] = "replace_identity"
    replacements: dict[str, IdentityReplacement] = PydanticField(
        ...,
        validation_alias=AliasChoices("replacements", "vertices"),
        description=(
            "Per-vertex identity replacement: ``{vertex_name: replacement}``. "
            "``vertices`` is accepted as a legacy alias."
        ),
        min_length=1,
    )

Attributes

op = 'replace_identity' class-attribute instance-attribute

replacements = PydanticField(..., validation_alias=AliasChoices('replacements', 'vertices'), description='Per-vertex identity replacement: ``{vertex_name: replacement}``. ``vertices`` is accepted as a legacy alias.', min_length=1) class-attribute instance-attribute

ReplaceResourcesOp

Bases: ConfigBaseModel

Replace the definitions of existing resources, matched by name.

Each resource keeps its position in ingestion_model.resources and its bindings; only its definition changes. This is how an edited pipeline is expressed, since a pipeline is an ordered program no finer op can patch.

Source code in graflo/architecture/evolution/ops.py

class ReplaceResourcesOp(ConfigBaseModel):
    """Replace the definitions of existing resources, matched by name.

    Each resource keeps its position in ``ingestion_model.resources`` and its
    bindings; only its definition changes. This is how an edited pipeline is
    expressed, since a pipeline is an ordered program no finer op can patch.
    """

    op: Literal["replace_resources"] = "replace_resources"
    resources: list[ResourceConfig] = PydanticField(
        ...,
        description="Full new definitions; each name must already exist.",
        min_length=1,
    )
    transforms: list[ProtoTransform] = PydanticField(
        default_factory=list,
        description=(
            "Named transforms the new definitions reference via ``call.use``, "
            "registered as ``add_resources`` registers them."
        ),
    )

    @model_validator(mode="after")
    def _validate_unique_names(self) -> ReplaceResourcesOp:
        names = [resource.name for resource in self.resources]
        if len(names) != len(set(names)):
            raise ValueError("replace_resources entries must be unique by name")
        return self

Attributes

op = 'replace_resources' class-attribute instance-attribute

resources = PydanticField(..., description='Full new definitions; each name must already exist.', min_length=1) class-attribute instance-attribute

transforms = PydanticField(default_factory=list, description='Named transforms the new definitions reference via ``call.use``, registered as ``add_resources`` registers them.') class-attribute instance-attribute

RetargetEdgesOp

Bases: ConfigBaseModel

Change which vertex types an edge connects, preserving everything else.

Remove-plus-add would lose the edge's properties, uniqueness keys, directed flag, and its db_profile physical spec. Retargeting rewrites the EdgeId everywhere it is keyed instead: edge config, physical specs, and pipeline edge steps.

Source code in graflo/architecture/evolution/ops.py

class RetargetEdgesOp(ConfigBaseModel):
    """Change which vertex types an edge connects, preserving everything else.

    Remove-plus-add would lose the edge's properties, uniqueness keys, ``directed``
    flag, and its ``db_profile`` physical spec. Retargeting rewrites the ``EdgeId``
    everywhere it is keyed instead: edge config, physical specs, and pipeline edge steps.
    """

    op: Literal["retarget_edges"] = "retarget_edges"
    edges: list[EdgeRetargetEntry] = PydanticField(
        ...,
        description="Per-edge endpoint retargeting.",
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_unique_selectors(self) -> RetargetEdgesOp:
        edge_ids = [entry.edge_id() for entry in self.edges]
        if len(edge_ids) != len(set(edge_ids)):
            raise ValueError(
                "retarget_edges entries must be unique by (source, target, relation)"
            )
        retargeted = [entry.retargeted_edge_id() for entry in self.edges]
        if len(retargeted) != len(set(retargeted)):
            raise ValueError(
                "retarget_edges entries must not collide after retargeting"
            )
        return self

Attributes

edges = PydanticField(..., description='Per-edge endpoint retargeting.', min_length=1) class-attribute instance-attribute

op = 'retarget_edges' class-attribute instance-attribute

RetractEdgeInversesOp

Bases: ConfigBaseModel

Withdraw declared inverses, addressed by relation name.

A paired relation retracts its whole pair (either side names it); a symmetric relation retracts its own declaration. Refused while a native inverse still realizes a pair: that would leave the database maintaining a type the schema no longer names. Explicit inverse edges are ordinary edges and survive a retraction; so does directed: false.

Source code in graflo/architecture/evolution/ops.py

class RetractEdgeInversesOp(ConfigBaseModel):
    """Withdraw declared inverses, addressed by relation name.

    A paired relation retracts its whole pair (either side names it); a
    symmetric relation retracts its own declaration. Refused while a native
    inverse still realizes a pair: that would leave the database maintaining a
    type the schema no longer names. Explicit inverse edges are ordinary edges
    and survive a retraction; so does ``directed: false``.
    """

    op: Literal["retract_edge_inverses"] = "retract_edge_inverses"
    relations: list[str] = PydanticField(
        ...,
        description="Relations whose declaration is withdrawn (either side of a pair).",
        min_length=1,
    )

Attributes

op = 'retract_edge_inverses' class-attribute instance-attribute

relations = PydanticField(..., description='Relations whose declaration is withdrawn (either side of a pair).', min_length=1) class-attribute instance-attribute

SanitizeOp

Bases: ConfigBaseModel

Record the physical names a target flavor needs in DatabaseProfile.

Vertex storage names, relation names and vertex/edge property names that the flavor cannot store (reserved word, invalid character, forbidden prefix) get a stored name in the profile, deduplicated within each database namespace. The logical schema and the ingestion model are left untouched: a backend naming constraint is a physical fact, so it never renames a logical property. Idempotent.

Source code in graflo/architecture/evolution/ops.py

class SanitizeOp(ConfigBaseModel):
    """Record the physical names a target flavor needs in ``DatabaseProfile``.

    Vertex storage names, relation names and vertex/edge property names that the
    flavor cannot store (reserved word, invalid character, forbidden prefix) get
    a stored name in the profile, deduplicated within each database namespace.
    The logical schema and the ingestion model are left untouched: a backend
    naming constraint is a physical fact, so it never renames a logical property.
    Idempotent.
    """

    op: Literal["sanitize"] = "sanitize"
    db_flavor: DBType = PydanticField(
        ...,
        description="Target database flavor whose reserved words/constraints drive the sanitization.",
    )
    reserved_words: list[str] | None = PydanticField(
        default=None,
        description=(
            "Optional override for the flavor's reserved words. "
            "When unset, ``graflo.db.util.load_reserved_words(db_flavor)`` is used."
        ),
    )

Attributes

db_flavor = PydanticField(..., description='Target database flavor whose reserved words/constraints drive the sanitization.') class-attribute instance-attribute

op = 'sanitize' class-attribute instance-attribute

reserved_words = PydanticField(default=None, description="Optional override for the flavor's reserved words. When unset, ``graflo.db.util.load_reserved_words(db_flavor)`` is used.") class-attribute instance-attribute

SetEdgeDirectedOp

Bases: ConfigBaseModel

Set the directed flag on logical edges.

Small, but load-bearing for replay: directed decides what :class:AddInverseEdgesOp is allowed to duplicate, so an un-authorable flag makes inverse-edge change sets non-replayable.

Source code in graflo/architecture/evolution/ops.py

class SetEdgeDirectedOp(ConfigBaseModel):
    """Set the ``directed`` flag on logical edges.

    Small, but load-bearing for replay: ``directed`` decides what
    :class:`AddInverseEdgesOp` is allowed to duplicate, so an un-authorable flag makes
    inverse-edge change sets non-replayable.
    """

    op: Literal["set_edge_directed"] = "set_edge_directed"
    edges: list[EdgeSelector] = PydanticField(
        ...,
        description="Edge triples whose ``directed`` flag changes.",
        min_length=1,
    )
    directed: bool = PydanticField(
        ...,
        description="Value applied to every selected edge.",
    )

    @model_validator(mode="after")
    def _validate_unique_selectors(self) -> SetEdgeDirectedOp:
        _validate_unique_edge_selectors(self.edges, kind="set_edge_directed")
        return self

Attributes

directed = PydanticField(..., description='Value applied to every selected edge.') class-attribute instance-attribute

edges = PydanticField(..., description='Edge triples whose ``directed`` flag changes.', min_length=1) class-attribute instance-attribute

op = 'set_edge_directed' class-attribute instance-attribute

SetEdgeSemanticsOp

Bases: ConfigBaseModel

Ground edge relations in an external vocabulary.

The vertex op's counterpart. Relations carry as much meaning as types -- wasDerivedFrom and dependsOn are not interchangeable -- and a conformance profile that asks whether types are grounded has to be able to ask it of edges too.

Source code in graflo/architecture/evolution/ops.py

class SetEdgeSemanticsOp(ConfigBaseModel):
    """Ground edge relations in an external vocabulary.

    The vertex op's counterpart. Relations carry as much meaning as types --
    ``wasDerivedFrom`` and ``dependsOn`` are not interchangeable -- and a
    conformance profile that asks whether types are grounded has to be able to
    ask it of edges too.
    """

    op: Literal["set_edge_semantics"] = "set_edge_semantics"
    edges: list[EdgeSelector] = PydanticField(
        ...,
        description="Edge triples whose grounding changes.",
        min_length=1,
    )
    semantics: Semantics | None = PydanticField(
        default=None,
        description=(
            "Grounding applied to every selected edge; ``None`` clears it. Not "
            "``FieldSemantics``: a unit on an edge is meaningless, and the model "
            "split is what makes ``unit:`` here a validation error."
        ),
    )

    @model_validator(mode="after")
    def _validate_unique_selectors(self) -> SetEdgeSemanticsOp:
        _validate_unique_edge_selectors(self.edges, kind="set_edge_semantics")
        return self

Attributes

edges = PydanticField(..., description='Edge triples whose grounding changes.', min_length=1) class-attribute instance-attribute

op = 'set_edge_semantics' class-attribute instance-attribute

semantics = PydanticField(default=None, description='Grounding applied to every selected edge; ``None`` clears it. Not ``FieldSemantics``: a unit on an edge is meaningless, and the model split is what makes ``unit:`` here a validation error.') class-attribute instance-attribute

SetFieldSemanticsOp

Bases: ConfigBaseModel

Ground vertex and edge properties, including their unit of measure.

Takes :class:~graflo.architecture.schema.semantics.FieldSemantics rather than :class:~graflo.architecture.schema.semantics.Semantics, which is the entire reason this is a third op rather than a mode of the vertex one: only a property may carry unit, and the two models are kept apart so that unit: on a type is a validation error rather than a silent no-op.

A target names either a vertex property (vertex + field) or an edge property (source / target / relation + field).

Source code in graflo/architecture/evolution/ops.py

class SetFieldSemanticsOp(ConfigBaseModel):
    """Ground vertex and edge properties, including their unit of measure.

    Takes :class:`~graflo.architecture.schema.semantics.FieldSemantics` rather
    than :class:`~graflo.architecture.schema.semantics.Semantics`, which is the
    entire reason this is a third op rather than a mode of the vertex one: only
    a property may carry ``unit``, and the two models are kept apart so that
    ``unit:`` on a type is a validation error rather than a silent no-op.

    A target names either a vertex property (``vertex`` + ``field``) or an
    edge property (``source`` / ``target`` / ``relation`` + ``field``).
    """

    op: Literal["set_field_semantics"] = "set_field_semantics"
    targets: list[FieldSemanticsTarget | EdgeFieldSemanticsTarget] = PydanticField(
        ...,
        description="Properties whose grounding changes.",
        min_length=1,
    )

    @model_validator(mode="after")
    def _validate_unique_targets(self) -> SetFieldSemanticsOp:
        keys = [t.key() for t in self.targets]
        if len(keys) != len(set(keys)):
            raise ValueError(
                "set_field_semantics targets must be unique by (vertex, field) or "
                "(source, target, relation, field)"
            )
        return self

Attributes

op = 'set_field_semantics' class-attribute instance-attribute

targets = PydanticField(..., description='Properties whose grounding changes.', min_length=1) class-attribute instance-attribute

SetInverseEmissionOp

Bases: ConfigBaseModel

Set or clear emit_inverse on edge steps, addressed by position.

The ingestion half of a materialized inverse, as a primitive: which steps mirror the edges they write into the declared inverse. A step is addressed by resource and :class:EdgeStepRef (at / step / link), so the op says exactly which steps change and nothing is inferred at replay time.

Enabling is refused where the flag could never write anything -- a step naming exactly one edge whose relation has no declared pair, is symmetric, or has no declared inverse edge. A step whose relation comes from the data is accepted; it mirrors per document what has a materialized inverse.

Source code in graflo/architecture/evolution/ops.py

class SetInverseEmissionOp(ConfigBaseModel):
    """Set or clear ``emit_inverse`` on edge steps, addressed by position.

    The ingestion half of a materialized inverse, as a primitive: which steps
    mirror the edges they write into the declared inverse. A step is addressed
    by resource and :class:`EdgeStepRef` (``at`` / ``step`` / ``link``), so the
    op says exactly which steps change and nothing is inferred at replay time.

    Enabling is refused where the flag could never write anything -- a step
    naming exactly one edge whose relation has no declared pair, is symmetric,
    or has no declared inverse edge. A step whose relation comes from the data
    is accepted; it mirrors per document what has a materialized inverse.
    """

    op: Literal["set_inverse_emission"] = "set_inverse_emission"
    steps: dict[str, list[EdgeStepRef]] = PydanticField(
        ...,
        description="Edge steps whose flag changes: ``{resource: [step ref, ...]}``.",
        min_length=1,
    )
    enabled: bool = PydanticField(
        default=True,
        description="``False`` clears the flag.",
    )

    @model_validator(mode="after")
    def _validate_steps(self) -> SetInverseEmissionOp:
        empty = sorted(name for name, refs in self.steps.items() if not refs)
        if empty:
            raise ValueError(f"set_inverse_emission: no steps given for {empty}")
        for name, refs in self.steps.items():
            keys = [ref.sort_key for ref in refs]
            if len(set(keys)) != len(keys):
                raise ValueError(
                    f"set_inverse_emission: steps of {name!r} must be unique"
                )
        return self

Attributes

enabled = PydanticField(default=True, description='``False`` clears the flag.') class-attribute instance-attribute

op = 'set_inverse_emission' class-attribute instance-attribute

steps = PydanticField(..., description='Edge steps whose flag changes: ``{resource: [step ref, ...]}``.', min_length=1) class-attribute instance-attribute

SetNativeInversesOp

Bases: ConfigBaseModel

Have the database maintain declared inverses (TigerGraph WITH REVERSE_EDGE).

Physical, not logical: adds relations to, or removes them from, db_profile.native_inverses. Keyed by relation, as TigerGraph is: the reverse type belongs to the edge type, which spans every (S, T) pair of the relation. The reverse type is named by the declared pair, so the pair must be declared (:class:DeclareEdgeInversesOp). Refused for symmetric relations, where explicit inverse edges exist, on undirected edges, for a relation stored under several physical names, and on non-TigerGraph profiles.

Source code in graflo/architecture/evolution/ops.py

class SetNativeInversesOp(ConfigBaseModel):
    """Have the database maintain declared inverses (TigerGraph ``WITH REVERSE_EDGE``).

    Physical, not logical: adds relations to, or removes them from,
    ``db_profile.native_inverses``. Keyed by relation, as TigerGraph is: the
    reverse type belongs to the edge type, which spans every ``(S, T)`` pair of
    the relation. The reverse type is named by the declared pair, so the pair
    must be declared (:class:`DeclareEdgeInversesOp`). Refused for symmetric
    relations, where explicit inverse edges exist, on undirected edges, for a
    relation stored under several physical names, and on non-TigerGraph
    profiles.
    """

    op: Literal["set_native_inverses"] = "set_native_inverses"
    relations: list[str] = PydanticField(
        ...,
        description="Paired relations whose declared inverse the database maintains.",
        min_length=1,
    )
    enabled: bool = PydanticField(
        default=True,
        description="``False`` withdraws the native inverse.",
    )

    @model_validator(mode="after")
    def _validate_unique(self) -> SetNativeInversesOp:
        if len(set(self.relations)) != len(self.relations):
            raise ValueError("set_native_inverses: relations must be unique")
        return self

Attributes

enabled = PydanticField(default=True, description='``False`` withdraws the native inverse.') class-attribute instance-attribute

op = 'set_native_inverses' class-attribute instance-attribute

relations = PydanticField(..., description='Paired relations whose declared inverse the database maintains.', min_length=1) class-attribute instance-attribute

SetVertexDescriptionsOp

Bases: ConfigBaseModel

Set or clear the human-readable description of existing vertex types.

A description was authorable only when a type was first written, so a change to one -- a merge folding two types together and joining what each side said about it, or a correction to what an inferred type means -- had no operation and could only be diffed as inexpressible. Like grounding, a description is never consulted at execution time: this op cannot change how anything ingests or stores.

Source code in graflo/architecture/evolution/ops.py

class SetVertexDescriptionsOp(ConfigBaseModel):
    """Set or clear the human-readable description of existing vertex types.

    A description was authorable only when a type was first written, so a
    change to one -- a merge folding two types together and joining what each
    side said about it, or a correction to what an inferred type means -- had no
    operation and could only be diffed as inexpressible. Like grounding, a
    description is never consulted at execution time: this op cannot change how
    anything ingests or stores.
    """

    op: Literal["set_vertex_descriptions"] = "set_vertex_descriptions"
    descriptions: dict[str, str | None] = PydanticField(
        ...,
        description=(
            "Per-vertex description: ``{vertex_name: text}``. ``None`` clears it, "
            "which is what makes the op invertible."
        ),
        min_length=1,
    )

Attributes

descriptions = PydanticField(..., description='Per-vertex description: ``{vertex_name: text}``. ``None`` clears it, which is what makes the op invertible.', min_length=1) class-attribute instance-attribute

op = 'set_vertex_descriptions' class-attribute instance-attribute

SetVertexSemanticsOp

Bases: ConfigBaseModel

Ground vertex types in an external vocabulary.

Semantics were authorable only when a type was first written: no operation could attach an iri to a type that already existed, so a manifest that arrived ungrounded — an inferred one, or anything predating the block — could never be grounded through the op system, only rewritten by hand.

Grounding is purely additive and never consulted at execution time, so this op cannot change how anything ingests or stores. What it changes is whether a reader who did not author the schema can tell what a type denotes.

Source code in graflo/architecture/evolution/ops.py

class SetVertexSemanticsOp(ConfigBaseModel):
    """Ground vertex types in an external vocabulary.

    Semantics were authorable only when a type was first written: no operation
    could attach an ``iri`` to a type that already existed, so a manifest that
    arrived ungrounded — an inferred one, or anything predating the block — could
    never be grounded through the op system, only rewritten by hand.

    Grounding is purely additive and never consulted at execution time, so this
    op cannot change how anything ingests or stores. What it changes is whether a
    reader who did not author the schema can tell what a type denotes.
    """

    op: Literal["set_vertex_semantics"] = "set_vertex_semantics"
    semantics: dict[str, Semantics | None] = PydanticField(
        ...,
        description=(
            "Per-vertex grounding: ``{vertex_name: Semantics}``. ``None`` clears "
            "the block, which is what makes the op invertible."
        ),
        min_length=1,
    )

Attributes

op = 'set_vertex_semantics' class-attribute instance-attribute

semantics = PydanticField(..., description='Per-vertex grounding: ``{vertex_name: Semantics}``. ``None`` clears the block, which is what makes the op invertible.', min_length=1) class-attribute instance-attribute

SideRenames

Bases: ConfigBaseModel

Renames applied to one side's names that no equivalence groups.

Applied simultaneously with the side's groups and vocabulary, so a chain ({Asset: WorkOrder, WorkOrder: Ticket}) and a swap resolve without an intermediate name, and any name either side uses or vacates may be a target. Every source must exist on the side. A class or relation that an equivalence groups is named by that equivalence's into, not here.

Source code in graflo/architecture/evolution/ops.py

class SideRenames(ConfigBaseModel):
    """Renames applied to one side's names that no equivalence groups.

    Applied simultaneously with the side's groups and vocabulary, so a chain
    (``{Asset: WorkOrder, WorkOrder: Ticket}``) and a swap resolve without an
    intermediate name, and any name either side uses or vacates may be a
    target. Every source must exist on the side. A class or relation that an
    equivalence groups is named by that equivalence's ``into``, not here.
    """

    vertices: dict[str, str] = PydanticField(
        default_factory=dict, description="Class renames: ``{old: new}``."
    )
    relations: dict[str, str] = PydanticField(
        default_factory=dict, description="Relation renames: ``{old: new}``."
    )
    properties: dict[str, dict[str, str]] = PydanticField(
        default_factory=dict,
        description=(
            "Attribute renames keyed by the class's own name: ``{class: {old: new}}``."
        ),
    )
    resources: dict[str, str] = PydanticField(
        default_factory=dict,
        description="Resource renames, applied before the union: ``{old: new}``.",
    )

    @model_validator(mode="after")
    def _validate_renames(self) -> SideRenames:
        validate_rename_map_is_injective(
            {s: t for s, t in self.vertices.items() if s != t},
            kind="renames.vertices",
            merge_hint="a VertexEquivalence",
        )
        validate_rename_map_is_injective(
            {s: t for s, t in self.relations.items() if s != t},
            kind="renames.relations",
            merge_hint="a RelationEquivalence",
        )
        validate_rename_map_is_injective(
            {s: t for s, t in self.resources.items() if s != t},
            kind="renames.resources",
            merge_hint="distinct resource names",
        )
        for cls, attr_map in self.properties.items():
            validate_rename_map_is_injective(
                attr_map,
                kind=f"renames.properties (class {cls!r})",
                merge_hint="a transform that combines the fields upstream",
            )
        return self

    def is_empty(self) -> bool:
        """Whether no rename is declared."""
        return not (
            self.vertices or self.relations or self.properties or self.resources
        )

Attributes

properties = PydanticField(default_factory=dict, description="Attribute renames keyed by the class's own name: ``{class: {old: new}}``.") class-attribute instance-attribute

relations = PydanticField(default_factory=dict, description='Relation renames: ``{old: new}``.') class-attribute instance-attribute

resources = PydanticField(default_factory=dict, description='Resource renames, applied before the union: ``{old: new}``.') class-attribute instance-attribute

vertices = PydanticField(default_factory=dict, description='Class renames: ``{old: new}``.') class-attribute instance-attribute

Methods:

is_empty()

Whether no rename is declared.

Source code in graflo/architecture/evolution/ops.py

def is_empty(self) -> bool:
    """Whether no rename is declared."""
    return not (
        self.vertices or self.relations or self.properties or self.resources
    )

VertexEquivalence

Bases: ConfigBaseModel

Collapse one or more left classes and one or more right classes into one.

GraFlo applies this map deterministically; it does not infer semantic matches. left / right accept a bare class name (a 1-1 equivalence) or a list (an n-ary cluster): {Company, Shop} ~ {Org, Branch} -> Company. A member may also be spelled by the canonical name a vocabulary gives it, which stands for every class the vocabulary sends there.

The group is everything the equivalence links: its members, and every class a vocabulary merges with one of them. Per-member maps (property equivalences, member-keyed identity sources) may name any class of the group by its own name.

Properties with the same spelling on every member after alignment fuse by exact name without an entry in properties — list only renames.

identity is the merged key, as ordered funnel branches in canonical names. A branch is a property the members carry (a name, or a composite [a, b]), a :class:DerivedBranch each source computes from its own columns, or a :class:LocalKeyBranch (the tagged fallback, last). One property branch keys the class on that natural key; anything else keys it on a funnel, where a record keys on its first complete branch. A funnel's digest is stored in digest_field (id by default).

derive declares attributes each source computes, without making each one a branch: a name or composite branch in identity keys on them, so [host_key, group_key] fires only when both are derived and the digest joins them. A DerivedBranch {name, sources} is the same as derive: {name: sources} with the branch name.

Source code in graflo/architecture/evolution/ops.py

class VertexEquivalence(ConfigBaseModel):
    """Collapse one or more left classes and one or more right classes into one.

    GraFlo applies this map deterministically; it does not infer semantic
    matches. ``left`` / ``right`` accept a bare class name (a 1-1 equivalence)
    or a list (an n-ary cluster): ``{Company, Shop} ~ {Org, Branch} ->
    Company``. A member may also be spelled by the canonical name a vocabulary
    gives it, which stands for every class the vocabulary sends there.

    The group is everything the equivalence links: its members, and every
    class a vocabulary merges with one of them. Per-member maps (property
    equivalences, member-keyed identity sources) may name any class of the
    group by its own name.

    Properties with the same spelling on every member after alignment fuse by
    exact name without an entry in ``properties`` — list only renames.

    ``identity`` is the merged key, as ordered funnel branches in canonical
    names. A branch is a property the members carry (a name, or a composite
    ``[a, b]``), a :class:`DerivedBranch` each source computes from its own
    columns, or a :class:`LocalKeyBranch` (the tagged fallback, last). One
    property branch keys the class on that natural key; anything else keys it
    on a funnel, where a record keys on its first complete branch. A funnel's
    digest is stored in ``digest_field`` (``id`` by default).

    ``derive`` declares attributes each source computes, without making each
    one a branch: a name or composite branch in ``identity`` keys on them, so
    ``[host_key, group_key]`` fires only when both are derived and the digest
    joins them. A ``DerivedBranch`` ``{name, sources}`` is the same as
    ``derive: {name: sources}`` with the branch ``name``.
    """

    left: str | list[str] = PydanticField(
        ..., description="One or more left-manifest vertex type names."
    )
    right: str | list[str] = PydanticField(
        ..., description="One or more right-manifest vertex type names."
    )
    into: str | None = PydanticField(
        default=None,
        description=(
            "Merged vertex type name: any name, including one either side "
            "uses or vacates. Never translated by a vocabulary; when it differs "
            "from the vocabulary's name for the group it wins. Omitted, the "
            "name comes from the vocabulary, or from the one spelling every "
            "member shares."
        ),
    )
    properties: list[PropertyEquivalence] = PydanticField(
        default_factory=list,
        description="Property alignment map applied before the vertex merge.",
    )
    identity: list[IdentityBranchDecl] | None = PydanticField(
        default=None,
        description=(
            "The merged key, as ordered funnel branches in canonical names: a "
            "property the members carry, a composite ``[a, b]``, a derived "
            "branch ``{name, sources}``, or ``{local_key: ...}`` (last). One "
            "property branch is a natural key; anything else is a funnel. When "
            "unset, identity is carried through only if every member agrees; "
            "disagreement raises `MergeIdentityError`."
        ),
    )
    derive: dict[str, dict[str, DerivationSpec | dict[str, DerivationSpec]]] | None = (
        PydanticField(
            default=None,
            description=(
                "Attributes each source derives from its own columns, keyed by "
                "attribute name; each entry has the shape of `DerivedBranch."
                "sources`. Identity branches key on them by name: a composite "
                "``[a, b]`` of derived attributes fires only when every part is "
                "derived. Requires `identity`."
            ),
        )
    )
    digest_field: str = PydanticField(
        default="id",
        min_length=1,
        description=(
            "Property the merged funnel's digest is stored in -- the class's "
            "identity field. Only for a funnel `identity`; it must not be a "
            "branch field. Set it when a member carries a real `id`."
        ),
    )
    derive_at: dict[str, list[int]] = PydanticField(
        default_factory=dict,
        description=(
            "Per-resource pipeline level the derived and local-key branches "
            "derive at, as ``descend`` step indices. Omitted resources resolve "
            "to the single level producing the class (for member-keyed "
            "sources: the level producing the member on its side); supply a "
            "path only when a resource produces it at more than one level."
        ),
    )
    retire: Literal["demote", "keep"] = PydanticField(
        default="demote",
        description=(
            "What becomes of each member's pre-merge key once `identity` "
            "re-keys the merged class. `demote` keeps each as a lookup-only "
            "secondary identity on `into`, and points edge steps of resources "
            "that only reference a member at it; `keep` leaves the fields as "
            "plain properties. Unused while the merged class keeps its "
            "members' shared key."
        ),
    )
    allow: list[Literal["self_relations", "observation_fusion"]] = PydanticField(
        default_factory=list,
        description=(
            "Consequences of this merge it accepts. `self_relations`: an edge "
            "between two members becomes an edge from the class to itself. "
            "`observation_fusion`: two members produced in one accumulator "
            "slot (same pipeline level and `role`, or both bare) fuse into one "
            "node."
        ),
    )

    @field_validator("identity", mode="before")
    @classmethod
    def _branch_shapes(cls, value: Any) -> Any:
        """Name the four branch shapes instead of pydantic's union error."""
        if not isinstance(value, list):
            return value
        for entry in value:
            if isinstance(entry, str | DerivedBranch | LocalKeyBranch):
                continue
            if isinstance(entry, list) and all(isinstance(f, str) for f in entry):
                continue
            if isinstance(entry, dict) and ("sources" in entry or "local_key" in entry):
                continue
            raise ValueError(
                f"VertexEquivalence: identity branch {entry!r} has no known "
                f"shape; {_describe_branch_shapes()}"
            )
        return value

    def raw_branches(self) -> list[tuple[str, ...]]:
        """The property branches of ``identity``, each as its field tuple."""
        return [
            tuple(branch_fields(branch))
            for branch in self.identity or []
            if isinstance(branch, str | list)
        ]

    def derived_branches(self) -> list[DerivedBranch]:
        """The derived branches of ``identity``, in priority order."""
        return [b for b in self.identity or [] if isinstance(b, DerivedBranch)]

    def derive_attributes(self) -> list[DerivedBranch]:
        """The ``derive`` entries, each as the derivation of one attribute."""
        return [
            DerivedBranch(name=name, sources=sources)
            for name, sources in (self.derive or {}).items()
        ]

    def derivations(self) -> list[DerivedBranch]:
        """Every derived attribute: the ``derive`` entries, then derived branches."""
        return [*self.derive_attributes(), *self.derived_branches()]

    def local_key_branch(self) -> LocalKeyBranch | None:
        """The ``local_key`` branch of ``identity``, if declared."""
        return next(
            (b for b in self.identity or [] if isinstance(b, LocalKeyBranch)), None
        )

    @property
    def has_derivation(self) -> bool:
        """Whether the key needs pipeline steps: ``derive``, a derived or local-key branch."""
        return bool(self.derive) or any(
            isinstance(b, DerivedBranch | LocalKeyBranch) for b in self.identity or []
        )

    @property
    def left_members(self) -> list[str]:
        return _member_list(self.left)

    @property
    def right_members(self) -> list[str]:
        return _member_list(self.right)

    def members(self, side: Literal["left", "right"]) -> list[str]:
        return self.left_members if side == "left" else self.right_members

    def property_maps(
        self, side: Literal["left", "right"]
    ) -> dict[str, dict[str, str]]:
        """``{member: {old_field: into_field}}`` for *side*, bare strings expanded."""
        member_names = self.members(side)
        out: dict[str, dict[str, str]] = {}
        for pe in self.properties:
            spec = pe.left if side == "left" else pe.right
            if spec is None:
                continue
            per_member = (
                dict.fromkeys(member_names, spec) if isinstance(spec, str) else spec
            )
            for member, old in per_member.items():
                if old == pe.into:
                    continue
                bucket = out.setdefault(member, {})
                existing = bucket.get(old)
                if existing is not None and existing != pe.into:
                    raise ValueError(
                        f"VertexEquivalence: {side}:{member}.{old!r} would rename "
                        f"to both {existing!r} and {pe.into!r}"
                    )
                bucket[old] = pe.into
        return out

    @model_validator(mode="after")
    def _validate_members(self) -> VertexEquivalence:
        for side, members in (
            ("left", self.left_members),
            ("right", self.right_members),
        ):
            if not members:
                raise ValueError(
                    f"VertexEquivalence: {side} must name at least one class"
                )
            if len(members) != len(set(members)):
                raise ValueError(
                    f"VertexEquivalence: {side} lists a class more than once: {members}"
                )
        if len(self.allow) != len(set(self.allow)):
            raise ValueError(
                f"VertexEquivalence: allow lists a value twice: {self.allow}"
            )
        # Per-member property maps may name any class of the group, which only
        # the manifests and the vocabulary decide: checked at resolution.
        return self

    @model_validator(mode="after")
    def _validate_identity(self) -> VertexEquivalence:
        if self.derive is not None:
            if not self.derive:
                raise ValueError(
                    "VertexEquivalence: derive lists no attribute; omit it"
                )
            if self.identity is None:
                raise ValueError(
                    f"VertexEquivalence: derive computes {sorted(self.derive)}, "
                    "but no identity keys on them; declare identity"
                )
            # Each entry is validated as a derived branch's sources are.
            self.derive_attributes()
        if self.identity is None:
            if self.derive_at:
                raise ValueError(
                    "VertexEquivalence: derive_at is set but no identity branch "
                    "derives anything"
                )
            if self.digest_field != "id":
                raise ValueError(
                    f"VertexEquivalence: digest_field {self.digest_field!r} is "
                    "set but no identity is declared; it names where a funnel "
                    "identity's digest is stored"
                )
            return self
        check_identity_branches(
            self.identity, label="VertexEquivalence", derived=list(self.derive or {})
        )
        if self.derive_at and not self.has_derivation:
            raise ValueError(
                "VertexEquivalence: derive_at is set but no identity branch "
                "derives anything"
            )
        if self.is_natural_key:
            if self.digest_field != "id":
                raise ValueError(
                    f"VertexEquivalence: digest_field {self.digest_field!r} is "
                    f"set, but identity {self.identity!r} is a natural key, "
                    "stored in its own fields; digest_field applies only to a "
                    "funnel"
                )
            return self
        branch_names = {f for b in self.identity for f in branch_fields(b)}
        if self.digest_field in branch_names:
            raise ValueError(
                f"VertexEquivalence: digest_field {self.digest_field!r} is also "
                "an identity branch field. A branch names what the digest is "
                "computed from, not where it is stored, and the digest would "
                "replace that branch's input; store the digest in another field"
            )
        return self

    @property
    def is_natural_key(self) -> bool:
        """Whether ``identity`` is one property branch, kept as a natural key."""
        return self.identity is not None and (
            len(self.identity) == 1 and not self.has_derivation
        )

Attributes

allow = PydanticField(default_factory=list, description='Consequences of this merge it accepts. `self_relations`: an edge between two members becomes an edge from the class to itself. `observation_fusion`: two members produced in one accumulator slot (same pipeline level and `role`, or both bare) fuse into one node.') class-attribute instance-attribute

derive = PydanticField(default=None, description='Attributes each source derives from its own columns, keyed by attribute name; each entry has the shape of `DerivedBranch.sources`. Identity branches key on them by name: a composite ``[a, b]`` of derived attributes fires only when every part is derived. Requires `identity`.') class-attribute instance-attribute

derive_at = PydanticField(default_factory=dict, description='Per-resource pipeline level the derived and local-key branches derive at, as ``descend`` step indices. Omitted resources resolve to the single level producing the class (for member-keyed sources: the level producing the member on its side); supply a path only when a resource produces it at more than one level.') class-attribute instance-attribute

digest_field = PydanticField(default='id', min_length=1, description="Property the merged funnel's digest is stored in -- the class's identity field. Only for a funnel `identity`; it must not be a branch field. Set it when a member carries a real `id`.") class-attribute instance-attribute

has_derivation property

Whether the key needs pipeline steps: derive, a derived or local-key branch.

identity = PydanticField(default=None, description='The merged key, as ordered funnel branches in canonical names: a property the members carry, a composite ``[a, b]``, a derived branch ``{name, sources}``, or ``{local_key: ...}`` (last). One property branch is a natural key; anything else is a funnel. When unset, identity is carried through only if every member agrees; disagreement raises `MergeIdentityError`.') class-attribute instance-attribute

into = PydanticField(default=None, description="Merged vertex type name: any name, including one either side uses or vacates. Never translated by a vocabulary; when it differs from the vocabulary's name for the group it wins. Omitted, the name comes from the vocabulary, or from the one spelling every member shares.") class-attribute instance-attribute

is_natural_key property

Whether identity is one property branch, kept as a natural key.

left = PydanticField(..., description='One or more left-manifest vertex type names.') class-attribute instance-attribute

left_members property

properties = PydanticField(default_factory=list, description='Property alignment map applied before the vertex merge.') class-attribute instance-attribute

retire = PydanticField(default='demote', description="What becomes of each member's pre-merge key once `identity` re-keys the merged class. `demote` keeps each as a lookup-only secondary identity on `into`, and points edge steps of resources that only reference a member at it; `keep` leaves the fields as plain properties. Unused while the merged class keeps its members' shared key.") class-attribute instance-attribute

right = PydanticField(..., description='One or more right-manifest vertex type names.') class-attribute instance-attribute

right_members property

Methods:

derivations()

Every derived attribute: the derive entries, then derived branches.

Source code in graflo/architecture/evolution/ops.py

def derivations(self) -> list[DerivedBranch]:
    """Every derived attribute: the ``derive`` entries, then derived branches."""
    return [*self.derive_attributes(), *self.derived_branches()]

derive_attributes()

The derive entries, each as the derivation of one attribute.

Source code in graflo/architecture/evolution/ops.py

def derive_attributes(self) -> list[DerivedBranch]:
    """The ``derive`` entries, each as the derivation of one attribute."""
    return [
        DerivedBranch(name=name, sources=sources)
        for name, sources in (self.derive or {}).items()
    ]

derived_branches()

The derived branches of identity, in priority order.

Source code in graflo/architecture/evolution/ops.py

def derived_branches(self) -> list[DerivedBranch]:
    """The derived branches of ``identity``, in priority order."""
    return [b for b in self.identity or [] if isinstance(b, DerivedBranch)]

local_key_branch()

The local_key branch of identity, if declared.

Source code in graflo/architecture/evolution/ops.py

def local_key_branch(self) -> LocalKeyBranch | None:
    """The ``local_key`` branch of ``identity``, if declared."""
    return next(
        (b for b in self.identity or [] if isinstance(b, LocalKeyBranch)), None
    )

members(side)

Source code in graflo/architecture/evolution/ops.py

def members(self, side: Literal["left", "right"]) -> list[str]:
    return self.left_members if side == "left" else self.right_members

property_maps(side)

{member: {old_field: into_field}} for side, bare strings expanded.

Source code in graflo/architecture/evolution/ops.py

def property_maps(
    self, side: Literal["left", "right"]
) -> dict[str, dict[str, str]]:
    """``{member: {old_field: into_field}}`` for *side*, bare strings expanded."""
    member_names = self.members(side)
    out: dict[str, dict[str, str]] = {}
    for pe in self.properties:
        spec = pe.left if side == "left" else pe.right
        if spec is None:
            continue
        per_member = (
            dict.fromkeys(member_names, spec) if isinstance(spec, str) else spec
        )
        for member, old in per_member.items():
            if old == pe.into:
                continue
            bucket = out.setdefault(member, {})
            existing = bucket.get(old)
            if existing is not None and existing != pe.into:
                raise ValueError(
                    f"VertexEquivalence: {side}:{member}.{old!r} would rename "
                    f"to both {existing!r} and {pe.into!r}"
                )
            bucket[old] = pe.into
    return out

raw_branches()

The property branches of identity, each as its field tuple.

Source code in graflo/architecture/evolution/ops.py

def raw_branches(self) -> list[tuple[str, ...]]:
    """The property branches of ``identity``, each as its field tuple."""
    return [
        tuple(branch_fields(branch))
        for branch in self.identity or []
        if isinstance(branch, str | list)
    ]

Functions:

__dir__()

Source code in graflo/architecture/evolution/__init__.py

def __dir__() -> list[str]:
    return sorted(__all__)

__getattr__(name)

Source code in graflo/architecture/evolution/__init__.py

def __getattr__(name: str) -> Any:
    if name in _APPLY_EXPORTS:
        from . import apply as apply_mod

        return getattr(apply_mod, name)
    if name in _MERGE_EXPORTS:
        from . import merge as merge_mod

        return getattr(merge_mod, name)
    if name in _NAMING_GRAPH_EXPORTS:
        from . import naming_graph as naming_graph_mod

        return getattr(naming_graph_mod, name)
    if name in _MERGE_COMMIT_EXPORTS:
        from . import merge_commit as merge_commit_mod

        return getattr(merge_commit_mod, name)
    if name in _INGESTION_APPLY_EXPORTS:
        from . import ingestion as ingestion_mod

        return getattr(ingestion_mod, name)
    if name in _ALIGNMENT_EXPORTS:
        from . import alignment as alignment_mod

        return getattr(alignment_mod, name)
    if name in _CANONICAL_EXPORTS:
        from . import canonical as canonical_mod

        return getattr(canonical_mod, name)
    if name in _EQUIVALENCE_EXPORTS:
        from . import equivalence as equivalence_mod

        return getattr(equivalence_mod, name)
    if name in _IDENTITY_EXPORTS:
        from . import identity as identity_mod

        return getattr(identity_mod, name)
    if name in _STRUCTURE_EXPORTS:
        from . import structure as structure_mod

        return getattr(structure_mod, name)
    if name in _PHYSICAL_EXPORTS:
        from . import physical as physical_mod

        return getattr(physical_mod, name)
    if name in _SEMANTICS_EXPORTS:
        from . import semantics as semantics_mod

        return getattr(semantics_mod, name)
    if name in _CODEC_EXPORTS:
        from . import codec as codec_mod

        return getattr(codec_mod, name)
    if name in _HASHING_EXPORTS:
        from . import hashing as hashing_mod

        return getattr(hashing_mod, name)
    if name in _CANONICALIZE_EXPORTS:
        from . import canonicalize as canonicalize_mod

        return getattr(canonicalize_mod, name)
    if name in _AUTOGENERATE_EXPORTS:
        from . import autogenerate as autogenerate_mod

        return getattr(autogenerate_mod, name)
    if name in _INVERSE_EXPORTS:
        from . import inverse as inverse_mod

        return getattr(inverse_mod, name)
    if name in _INVERSE_PLAN_EXPORTS:
        from . import inverse_plan as inverse_plan_mod

        return getattr(inverse_plan_mod, name)
    if name in _COMMIT_EXPORTS:
        from . import commit as commit_mod

        return getattr(commit_mod, name)
    if name in _HISTORY_EXPORTS:
        from . import history as history_mod

        return getattr(history_mod, name)
    if name in _MERGE3_EXPORTS:
        from . import merge3 as merge3_mod

        return getattr(merge3_mod, name)
    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

ops_reaching_ingestion(ops)

Names of ops whose effect extends into ingestion_model, in order.

Source code in graflo/architecture/evolution/ops.py

def ops_reaching_ingestion(ops: Sequence[Any]) -> list[str]:
    """Names of *ops* whose effect extends into ``ingestion_model``, in order."""
    return [
        name
        for name in (getattr(op, "op", None) for op in ops)
        if isinstance(name, str) and name in INGESTION_REWRITING_OPS
    ]