Data model (MER)

Generated from source — do not edit

Emitted by scripts/gen_data_model_docs.py from the live Kernel.auto() registry and the SQLAlchemy table model. scripts/data_model_guard.py fails CI when this page and a fresh regeneration disagree. Edit the generator, never this file.

DNA's data model has two levels. The logical model — Kinds and the references between them — carries the meaning. The physical model is a generic document store that tells you almost nothing about the domain, and this page says so rather than dressing it up.

One database, four schema owners

A MER showing only the SDK's tables and stopping there misleads by omission. A single Postgres instance is shared by four independent schema owners, each migrating only its own tables:

Owner	Migrates	On this page
DNA SDK (this repo)	the document-store tables below, via its own Alembic tree	yes — fully
dna-cloud portal	its Prisma schema (accounts, plans, billing — real relational tables with real foreign keys)	no — separate repo, separate migration tool
Copilot service	copilot_thread and friends	no
LangGraph runtime	checkpoint* / store*	no

The SDK's Alembic run is explicitly told not to have opinions about tables it does not own — otherwise autogenerate would propose dropping another owner's data. That exclusion list is machine- readable, so it is reproduced from source rather than asserted:

Excluded from the SDK's autogenerate
alembic_version
dna_schema_migrations
dna_search_docs
dna_search_meta
dna_search_migrations
schema_migrations
search_docs
search_fts
search_meta
search_vec
sqlite_sequence

Logical model — Kinds and their references

80 Kinds are registered. Each is a document, not a table: a Kind costs a YAML descriptor and zero migrations, which is the point of an open type system. The cost is that references between Kinds are not database foreign keys — they are fields holding a name.

How to read the edges

Not every line here is equally trustworthy, and pretending otherwise would be the whole problem. Four tiers, strongest first:

Tier	Drawn	What it means
Declared	solid	The field carries x-dna-ref. The kernel resolves it at write time — the only tier the system enforces.
Composition (dep)	solid	dep_filters names the target Kind. A real declaration, but it drives prompt composition and is never checked against stored data.
Inferred	dashed	Nothing declares it; the field NAME resolves to a Kind. A convention, not a contract.
Unresolved	not drawn	Reference-shaped, no confident target. Tabulated below.

* on a label marks a polymorphic reference (several possible target Kinds).

108 edges: 15 declared, 66 composition-only, 27 inferred — plus 23 reference-shaped fields left unresolved and 6 known-undeclarable ones.

Only the declared tier cannot dangle

dep_filters declares a target Kind; nothing validates the value. A Feature.owner naming an Actor that does not exist is written without complaint. Solid therefore means "the model knows what this points at", not "this resolves". Closing that gap is what x-dna-ref does, one field at a time.

Overview — how the groups reference each other

Kinds are grouped by alias prefix (sdlc-, helix-, …) — a grouping that comes from the data. Arrows are counts of edges between groups; self-references are omitted here and shown in the detail diagrams.

flowchart LR
    agentskills["agentskills (1 Kind)"]
    audit["audit (2 Kinds)"]
    dna["dna (2 Kinds)"]
    eval["eval (4 Kinds)"]
    evidence["evidence (2 Kinds)"]
    guardrails["guardrails (1 Kind)"]
    helix["helix (13 Kinds)"]
    intel["intel (2 Kinds)"]
    kinddef["kinddef (1 Kind)"]
    lesson["lesson (1 Kind)"]
    mif["mif (1 Kind)"]
    portfolio["portfolio (5 Kinds)"]
    presidio["presidio (1 Kind)"]
    research["research (1 Kind)"]
    sdlc["sdlc (25 Kinds)"]
    soulspec["soulspec (1 Kind)"]
    tenant["tenant (6 Kinds)"]
    testkit["testkit (2 Kinds)"]
    agentskills -->|1| sdlc
    audit -->|2| portfolio
    evidence -->|1| eval
    evidence -->|1| sdlc
    helix -->|2| agentskills
    helix -->|2| guardrails
    helix -->|1| portfolio
    helix -->|1| presidio
    helix -->|1| sdlc
    helix -->|2| soulspec
    kinddef -->|1| dna
    lesson -->|1| agentskills
    sdlc -->|1| evidence
    sdlc -->|13| helix
    sdlc -->|1| intel
    sdlc -->|1| mif
    sdlc -->|1| research
    tenant -->|1| portfolio
    testkit -->|1| evidence

Detail by group

All 80 Kinds in one diagram is an unreadable hairball, so each group with at least 2 edges gets its own. A group carrying more than 20 edges is split again by tier, which keeps the enforced edges legible instead of losing them among the unvalidated ones. A box from another group appearing here is a cross-group reference.

audit (2 edges)

erDiagram
    AuditLog
    Role
    UserRoleAssignment
    AuditLog }o..}o Role : "roles (inferred)"
    UserRoleAssignment }o..}o Role : "roles (inferred)"

eval (3 edges)

erDiagram
    EvalBaseline
    EvalCase
    EvalRun
    EvalSuite
    EvalBaseline }o..|| EvalSuite : "suite (inferred)"
    EvalRun }o..|| EvalSuite : "suite (inferred)"
    EvalSuite }o--}o EvalCase : "cases (dep)"

evidence (2 edges)

erDiagram
    EvalSuite
    Evidence
    EvidencePolicy
    WorkflowEvent
    Evidence }o..|| EvalSuite : "suite (inferred)"
    EvidencePolicy }o..}o WorkflowEvent : "events (inferred)"

helix (15 edges)

erDiagram
    Actor
    Agent
    Guardrail
    PromptTemplate
    Recognizer
    Role
    SafetyPolicy
    Skill
    Soul
    Tool
    UseCase
    Actor }o..|| Role : "role (inferred)"
    Agent }o--|| Actor : "actors (dep)"
    Agent }o--}o Guardrail : "guardrails (dep)"
    Agent }o..|| PromptTemplate : "promptTemplate (inferred)"
    Agent }o--}o Skill : "skills (dep)"
    Agent }o--|| Soul : "soul (dep)"
    Agent }o--}o Tool : "tools (dep)"
    SafetyPolicy }o--}o Recognizer : "recognizers (dep)"
    UseCase }o--}o Agent : "agents (dep)"
    UseCase }o--}o Guardrail : "guardrails (dep)"
    UseCase }o--|| Actor : "primary_actor (dep)"
    UseCase }o--}o Skill : "skills (dep)"
    UseCase }o--|| Soul : "soul (dep)"
    UseCase }o--}o Actor : "supporting_actors (dep)"
    UseCase }o--}o Tool : "tools (dep)"

portfolio (4 edges)

erDiagram
    Membership
    Organization
    Project
    Repo
    Membership }o--|| Organization : "scope_ref *"
    Membership }o--|| Project : "scope_ref *"
    Project }o--|| Organization : "org_ref"
    Project }o--}o Repo : "repo_refs"

sdlc — declared (11 edges)

erDiagram
    Epic
    Feature
    Plan
    Spec
    Story
    Task
    Epic }o--}o Feature : "features"
    Feature }o--|| Epic : "epic"
    Feature }o--}o Story : "stories"
    Plan }o--|| Epic : "epic"
    Plan }o--|| Spec : "spec_ref"
    Spec }o--|| Epic : "epic"
    Spec }o--|| Spec : "supersedes"
    Story }o--}o Story : "dependencies"
    Story }o--|| Feature : "feature"
    Story }o--}o Spec : "spec_refs"
    Task }o--|| Story : "story_ref"

sdlc — composition (52 edges)

erDiagram
    ADR
    Actor
    AgentSession
    Bug
    Epic
    Feature
    Initiative
    Issue
    Kaizen
    Narrative
    Plan
    Postmortem
    Reference
    Retrospective
    RiskRegister
    Roadmap
    Spec
    Spike
    Story
    Task
    UseCase
    WorkflowEvent
    ADR }o--}o Feature : "covers_features (dep)"
    ADR }o--|| ADR : "superseded_by (dep)"
    ADR }o--}o ADR : "supersedes (dep)"
    AgentSession }o--}o Actor : "participants (dep)"
    Bug }o--|| ADR : "fix_adr (dep)"
    Bug }o--|| Feature : "related_feature (dep)"
    Bug }o--|| Story : "related_story (dep)"
    Feature }o--|| Actor : "owner (dep)"
    Feature }o--}o UseCase : "use_cases (dep)"
    Initiative }o--}o Epic : "epics (dep)"
    Initiative }o--|| Actor : "owner (dep)"
    Issue }o--|| Actor : "owner (dep)"
    Issue }o--|| Feature : "related_feature (dep)"
    Kaizen }o--|| Issue : "issue (dep)"
    Narrative }o--}o Epic : "covers_epics (dep)"
    Narrative }o--}o Feature : "covers_features (dep)"
    Narrative }o--}o Story : "covers_stories (dep)"
    Postmortem }o--}o Feature : "related_features (dep)"
    Postmortem }o--}o Story : "related_stories (dep)"
    Retrospective }o--}o Epic : "covers_epics (dep)"
    Retrospective }o--}o Feature : "covers_features (dep)"
    Retrospective }o--|| AgentSession : "covers_session (dep)"
    Retrospective }o--}o Story : "covers_stories (dep)"
    RiskRegister }o--|| Actor : "owner (dep)"
    RiskRegister }o--}o Epic : "related_epics (dep)"
    RiskRegister }o--}o Feature : "related_features (dep)"
    Roadmap }o--|| Epic : "epics (dep)"
    Spike }o--|| Feature : "feature (dep)"
    Spike }o--|| ADR : "follow_up_adr (dep)"
    Spike }o--|| Spec : "follow_up_spec (dep)"
    Spike }o--|| Story : "follow_up_story (dep)"
    Spike }o--}o Reference : "references (dep)"
    Spike }o--}o Spike : "related_spikes (dep)"
    Story }o--|| Actor : "owner (dep)"
    Task }o--|| Actor : "owner (dep)"
    WorkflowEvent }o--|| AgentSession : "parent_ref (dep) *"
    WorkflowEvent }o--|| Epic : "parent_ref (dep) *"
    WorkflowEvent }o--|| Feature : "parent_ref (dep) *"
    WorkflowEvent }o--|| Narrative : "parent_ref (dep) *"
    WorkflowEvent }o--|| Plan : "parent_ref (dep) *"
    WorkflowEvent }o--|| Roadmap : "parent_ref (dep) *"
    WorkflowEvent }o--|| Spec : "parent_ref (dep) *"
    WorkflowEvent }o--|| Story : "parent_ref (dep) *"
    WorkflowEvent }o--|| AgentSession : "ref (dep) *"
    WorkflowEvent }o--|| Epic : "ref (dep) *"
    WorkflowEvent }o--|| Feature : "ref (dep) *"
    WorkflowEvent }o--|| Narrative : "ref (dep) *"
    WorkflowEvent }o--|| Plan : "ref (dep) *"
    WorkflowEvent }o--|| Roadmap : "ref (dep) *"
    WorkflowEvent }o--|| Spec : "ref (dep) *"
    WorkflowEvent }o--|| Story : "ref (dep) *"
    WorkflowEvent }o--|| WorkflowEvent : "transitioned_from (dep)"

sdlc — inferred (11 edges)

erDiagram
    Actor
    CognitivePolicy
    Epic
    Evidence
    Feature
    Initiative
    IntelInsight
    Kaizen
    Memory
    Narrative
    Research
    Retrospective
    Spike
    StatusReport
    Theme
    WorkflowEvent
    CognitivePolicy }o..|| Memory : "memory (inferred)"
    Initiative }o..|| Theme : "theme_ref (inferred)"
    Kaizen }o..|| Actor : "actor (inferred)"
    Narrative }o..|| Actor : "actor (inferred)"
    Retrospective }o..|| Actor : "actor (inferred)"
    Spike }o..}o Research : "research_refs (inferred)"
    StatusReport }o..}o Evidence : "evidence_refs (inferred)"
    StatusReport }o..|| IntelInsight : "insight (inferred)"
    WorkflowEvent }o..|| Actor : "actor (inferred)"
    WorkflowEvent }o..|| Epic : "epic_ref (inferred)"
    WorkflowEvent }o..|| Feature : "feature_ref (inferred)"

tenant (3 edges)

erDiagram
    Role
    Tenant
    TenantMembership
    Workspace
    WorkspaceScopeGrant
    TenantMembership }o..|| Role : "role (inferred)"
    TenantMembership }o..|| Tenant : "tenant_slug (inferred)"
    WorkspaceScopeGrant }o..|| Workspace : "workspace_id (inferred)"

testkit (2 edges)

erDiagram
    Evidence
    TestGuide
    TestRun
    TestRun }o..}o Evidence : "evidence (inferred)"
    TestRun }o..|| TestGuide : "guide_ref (inferred)"

Groups with fewer than 2 edges (listed, not drawn): agentskills, kinddef, lesson.

Declared edges (x-dna-ref)

Enforced at write time. This table is the part of the graph the system will not let you break.

From	Field	To	Cardinality	Cross-group
Epic	features	Feature	many
Feature	epic	Epic	one
Feature	stories	Story	many
Membership	scope_ref (poly)	Organization	one
Membership	scope_ref (poly)	Project	one
Plan	epic	Epic	one
Plan	spec_ref	Spec	one
Project	org_ref	Organization	one
Project	repo_refs	Repo	many
Spec	epic	Epic	one
Spec	supersedes	Spec	one
Story	dependencies	Story	many
Story	feature	Feature	one
Story	spec_refs	Spec	many
Task	story_ref	Story	one

Composition edges (dep_filters only)

Declared for prompt composition, never validated against stored data. Each row is a candidate for an x-dna-ref promotion.

From	Field	To	Cardinality	Cross-group
ADR	covers_features	Feature	many
ADR	superseded_by	ADR	one
ADR	supersedes	ADR	many
Agent	actors	Actor	one
Agent	guardrails	Guardrail	many	yes
Agent	skills	Skill	many	yes
Agent	soul	Soul	one	yes
Agent	tools	Tool	many
AgentSession	participants	Actor	many	yes
Bug	fix_adr	ADR	one
Bug	related_feature	Feature	one
Bug	related_story	Story	one
EvalSuite	cases	EvalCase	many
Feature	owner	Actor	one	yes
Feature	use_cases	UseCase	many	yes
Initiative	epics	Epic	many
Initiative	owner	Actor	one	yes
Issue	owner	Actor	one	yes
Issue	related_feature	Feature	one
Kaizen	issue	Issue	one
Narrative	covers_epics	Epic	many
Narrative	covers_features	Feature	many
Narrative	covers_stories	Story	many
Postmortem	related_features	Feature	many
Postmortem	related_stories	Story	many
Retrospective	covers_epics	Epic	many
Retrospective	covers_features	Feature	many
Retrospective	covers_session	AgentSession	one
Retrospective	covers_stories	Story	many
RiskRegister	owner	Actor	one	yes
RiskRegister	related_epics	Epic	many
RiskRegister	related_features	Feature	many
Roadmap	epics	Epic	one
SafetyPolicy	recognizers	Recognizer	many	yes
Spike	feature	Feature	one
Spike	follow_up_adr	ADR	one
Spike	follow_up_spec	Spec	one
Spike	follow_up_story	Story	one
Spike	references	Reference	many
Spike	related_spikes	Spike	many
Story	owner	Actor	one	yes
Task	owner	Actor	one	yes
UseCase	agents	Agent	many
UseCase	guardrails	Guardrail	many	yes
UseCase	primary_actor	Actor	one
UseCase	skills	Skill	many	yes
UseCase	soul	Soul	one	yes
UseCase	supporting_actors	Actor	many
UseCase	tools	Tool	many
WorkflowEvent	parent_ref (poly)	AgentSession	one
WorkflowEvent	parent_ref (poly)	Epic	one
WorkflowEvent	parent_ref (poly)	Feature	one
WorkflowEvent	parent_ref (poly)	Narrative	one
WorkflowEvent	parent_ref (poly)	Plan	one
WorkflowEvent	parent_ref (poly)	Roadmap	one
WorkflowEvent	parent_ref (poly)	Spec	one
WorkflowEvent	parent_ref (poly)	Story	one
WorkflowEvent	ref (poly)	AgentSession	one
WorkflowEvent	ref (poly)	Epic	one
WorkflowEvent	ref (poly)	Feature	one
WorkflowEvent	ref (poly)	Narrative	one
WorkflowEvent	ref (poly)	Plan	one
WorkflowEvent	ref (poly)	Roadmap	one
WorkflowEvent	ref (poly)	Spec	one
WorkflowEvent	ref (poly)	Story	one
WorkflowEvent	transitioned_from	WorkflowEvent	one

Inferred edges (name convention)

Not declared anywhere. Each row is this generator matching a field name against the Kind registry — useful, and fallible.

From	Field	To	Cardinality	Cross-group
Actor	role	Role	one	yes
Agent	promptTemplate	PromptTemplate	one	yes
AuditLog	roles	Role	many	yes
CognitivePolicy	memory	Memory	one	yes
EvalBaseline	suite	EvalSuite	one
EvalRun	suite	EvalSuite	one
Evidence	suite	EvalSuite	one	yes
EvidencePolicy	events	WorkflowEvent	many	yes
Initiative	theme_ref	Theme	one	yes
Kaizen	actor	Actor	one	yes
KindDefinition	docs	Doc	one	yes
Lesson	skill	Skill	one	yes
Narrative	actor	Actor	one	yes
Retrospective	actor	Actor	one	yes
Skill	references	Reference	one	yes
Spike	research_refs	Research	many	yes
StatusReport	evidence_refs	Evidence	many	yes
StatusReport	insight	IntelInsight	one	yes
TenantMembership	role	Role	one	yes
TenantMembership	tenant_slug	Tenant	one
TestRun	evidence	Evidence	many	yes
TestRun	guide_ref	TestGuide	one
UserRoleAssignment	roles	Role	many	yes
WorkflowEvent	actor	Actor	one	yes
WorkflowEvent	epic_ref	Epic	one
WorkflowEvent	feature_ref	Feature	one
WorkspaceScopeGrant	workspace_id	Workspace	one

What this model cannot express

A MER that implies completeness is worse than none. These are the known gaps, generated alongside everything else so they cannot be quietly dropped.

Known-undeclarable references

Real edges that x-dna-ref deliberately does NOT declare. It resolves targets by document name, and these are keyed by something else — declaring them would produce false write-time violations on perfectly valid data. This is the concrete backlog for a future x-dna-ref-key.

Kind	Field	Really points at	Why undeclarable
Comment	target_ref	any	a composite Kind:name string — needs parsing, not a name lookup
Membership	role	Role	keyed by role_id, not the document name
Organization	plan_ref	Tier	keyed by tier_id (free/pro/enterprise), not the document name
Project	workspace_id	Workspace	keyed by the Workspace's opaque generated workspace_id, not its document name
WorkspaceMembership	role	Role	keyed by role_id (owner/admin/member/guest), not the document name
WorkspaceMembership	workspace_id	Workspace	same opaque workspace_id key

Unresolved reference-shaped fields

Fields that clearly point at something the model cannot name. This shrinks when references get declared, not when the generator gets cleverer.

Kind	Field	Why unresolved
AuditLog	request_id	reference-shaped, but request matches no registered Kind
Engram	affect_evidence_refs	reference-shaped, but affect_evidence matches no registered Kind
Engram	source_refs	reference-shaped, but source matches no registered Kind
Evidence	document_ref	reference-shaped, but document matches no registered Kind
Feature	sprint_ref	reference-shaped, but sprint matches no registered Kind
IntelInsight	source_ref	reference-shaped, but source matches no registered Kind
LayerPolicy	layer_id	reference-shaped, but layer matches no registered Kind
ModelProfile	model_id	reference-shaped, but model matches no registered Kind
PlanBinding	account_id	reference-shaped, but account matches no registered Kind
PlanBinding	stripe_customer_id	reference-shaped, but stripe_customer matches no registered Kind
PlanBinding	stripe_subscription_id	reference-shaped, but stripe_subscription matches no registered Kind
PlanBinding	tier_id	reference-shaped, but tier matches no registered Kind
PricingPlan	tier_id	reference-shaped, but tier matches no registered Kind
Project	intel_source_refs	reference-shaped, but intel_source matches no registered Kind
Research	scope_ref	reference-shaped, but scope matches no registered Kind
SourceArtifact	derived_refs	reference-shaped, but derived matches no registered Kind
Story	sprint_ref	reference-shaped, but sprint matches no registered Kind
TenantMembership	user_id	reference-shaped, but user matches no registered Kind
UserProfile	user_id	reference-shaped, but user matches no registered Kind
UserRoleAssignment	user_id	reference-shaped, but user matches no registered Kind
Workspace	account_id	reference-shaped, but account matches no registered Kind
Workspace	plan_ref	reference-shaped, but plan matches no registered Kind
WorkspaceMembership	identity_oid	reference-shaped, but identity matches no registered Kind

Suppressed name matches

The name-convention pass matched these and each is wrong. Listed rather than silently dropped, so the suppression is auditable.

Kind	Field	Why the match is wrong
AgentSession	tool	provenance enum of the AI coding tool that produced the session (claude-code | cursor | cline | …), not a Tool document
AuditLog	actor	the request identity string from claims (email/sub, or 'dev-user'), not a reference to an Actor document
Copilot	tenant	inbound-tenant handling mode for the emitted serving layer, not a reference to a Tenant document
Memory	namespace	MIF's hierarchical memory scope path (_semantic/decisions, §10) — a string axis inside the document, not the KindNamespace Kind, whose alias tenant-kind-namespace merely ends in the same token
Organization	plan_ref	the DNA Cloud Tier this org is on, not the SDLC Plan Kind
RemoteAgent	skills	the A2A Card's own skills[] (id/name/description/tags/examples) — structured self-description of what the remote agent can do, not a reference to the Skill Kind (agentskills), which merely shares the singular of the field name
Tenant	plan	billing/feature tier (a Tier tier_id), not the SDLC Plan Kind
Workspace	plan_ref	DEPRECATED and never read — billing is per ACCOUNT (workspace → account_id → AccountPlan); also not the SDLC Plan Kind

Kinds with no reference edge (22)

Standalone documents — configuration, composition-plane behaviour, or record Kinds whose links are simply not modelled yet.

AgentDefinition, Automation, Canvas, Changelog, Comment, Copilot, Engram, Genome, Hook, HtmlArtifact, IntelSource, KindNamespace, LayerPolicy, MCPFederation, ModelProfile, PlanBinding, PricingPlan, RemoteAgent, Setting, SourceArtifact, UserProfile, WorkspaceMembership

Physical model — the real tables

This diagram carries little information, by design

7 tables on Postgres (4 on SQLite) and 0 foreign keys. They are a generic document store: documents holds every Kind, of every type, as JSON in a content column keyed by (scope, kind, name, tenant). Adding a Kind adds rows, never a table — so the physical diagram cannot show you the domain. The logical model above is where the domain lives. This section exists to be accurate, not to look deep.

Postgres

erDiagram
    dna_bundle_entries {
        TEXT scope PK
        TEXT kind PK
        TEXT api_version PK
        TEXT name PK
        TEXT entry_path PK
        TEXT content
        TEXT updated_at
        TEXT tenant PK
        BLOB content_binary
    }
    dna_documents {
        TEXT scope PK
        TEXT kind PK
        TEXT api_version PK
        TEXT name PK
        TEXT content
        INTEGER version
        TEXT updated_at
        TEXT tenant PK
    }
    dna_layer_documents {
        TEXT scope PK
        TEXT layer_id PK
        TEXT layer_value PK
        TEXT kind PK
        TEXT name PK
        TEXT content
        TEXT updated_at
    }
    dna_outbox {
        BIGINT id PK
        DATETIME occurred_at
        TEXT scope
        TEXT tenant
        TEXT kind
        TEXT name
        TEXT op
        INTEGER doc_version
        TEXT actor
        TEXT cause
    }
    dna_quota_counters {
        DATE day PK
        TEXT tenant PK
        TEXT tier PK
        BIGINT calls
    }
    dna_versions {
        INTEGER id PK
        TEXT scope
        TEXT kind
        TEXT api_version
        TEXT name
        TEXT content
        INTEGER version
        BOOLEAN is_draft
        TEXT author
        TEXT created_at
        TEXT tenant
        TEXT semver
    }
    dna_versions_seq {
        TEXT scope PK
        TEXT tenant PK
        BIGINT last_id
        DATETIME last_at
    }

No lines connect these boxes because there are no foreign keys to draw. The join key is (scope, kind, name, tenant), applied in application code.

Dialect differences

The dialects are genuinely disjoint — Postgres tables carry a dna_ prefix, SQLite's do not, and Postgres has tables SQLite lacks.

Postgres	SQLite
dna_bundle_entries	bundle_entries
dna_documents	documents
dna_layer_documents	layer_documents
dna_outbox	—
dna_quota_counters	—
dna_versions	versions
dna_versions_seq	—

Columns

dna_bundle_entries

Column	Type	Key	Nullable
scope	TEXT	PK
kind	TEXT	PK
api_version	TEXT	PK
name	TEXT	PK
entry_path	TEXT	PK
content	TEXT
updated_at	TEXT
tenant	TEXT	PK
content_binary	BLOB		yes

dna_documents

Column	Type	Key	Nullable
scope	TEXT	PK
kind	TEXT	PK
api_version	TEXT	PK
name	TEXT	PK
content	TEXT
version	INTEGER
updated_at	TEXT
tenant	TEXT	PK

dna_layer_documents

Column	Type	Key	Nullable
scope	TEXT	PK
layer_id	TEXT	PK
layer_value	TEXT	PK
kind	TEXT	PK
name	TEXT	PK
content	TEXT
updated_at	TEXT

dna_outbox

Column	Type	Key	Nullable
id	BIGINT	PK
occurred_at	DATETIME
scope	TEXT
tenant	TEXT
kind	TEXT
name	TEXT
op	TEXT
doc_version	INTEGER
actor	TEXT		yes
cause	TEXT		yes

dna_quota_counters

Column	Type	Key	Nullable
day	DATE	PK
tenant	TEXT	PK
tier	TEXT	PK
calls	BIGINT

dna_versions

Column	Type	Key	Nullable
id	INTEGER	PK
scope	TEXT
kind	TEXT
api_version	TEXT
name	TEXT
content	TEXT
version	INTEGER
is_draft	BOOLEAN
author	TEXT		yes
created_at	TEXT
tenant	TEXT
semver	TEXT		yes

dna_versions_seq

Column	Type	Key	Nullable
scope	TEXT	PK
tenant	TEXT	PK
last_id	BIGINT
last_at	DATETIME