Unclassified

Spine Schema

The canonical data model for Spine Fabric — what is stored, how it is organized, and the rules that govern every record.

Spine Schema defines the canonical data model for all operational truth in SpineWorkspace. Every entity, relationship, event, decision, piece of evidence, work item, and memory entry follows this schema.

The schema is not a single database table. It is a multi-store fabric where each layer owns a specific class of truth.

Governing Rules

  1. One Spine per tenant. Role, department, and industry alter projection scope — not storage ownership.
  2. No direct UI database reads. The Workbench reads projected shapes through the Projection Platform API.
  3. All writes preserve provenance. Every mutation records actor, origin, evidence, confidence, timestamps, and source connection.
  4. AI proposes; humans decide. Cognitive output enters a proposal/approval path before operational promotion.
  5. Connect once, available everywhere. Connector ownership is tenant-scoped, not frontend-scoped.

Multi-Store Fabric

The Spine is a 6-layer fabric:

LayerStoreWhat It OwnsAccess Pattern
L1PostgreSQL / Neon / D1Structured operational truth — entities, relationships, tasks, approvals, signalsSQL via Projection Platform
L2Redis / KVHot state — projection caches, SLA countdowns, queue positions, session contextKey-value, sub-100ms
L3R2 / Object StorageAttachments — documents, recordings, exports, imagesSigned URLs, streaming
L4Conversational StoreIngested channels — email threads, chat transcripts, call recordingsAppend-only, Pipeline-processed
L5Vectorize / pgvectorSemantic embeddings — similarity, knowledge retrieval, signal detectionVector search, RRF scoring
L6Durable ObjectsAgent memory — session state, user preferences, workspace contextPer-keyed hot state, snapshot to D1

Rule: No department reads directly from any store. All reads go through the Projection Platform API.

Universal Record Envelope

Every operational table carries this minimum contract:

id                 text primary key,
tenant_id          text not null,
workspace_id       text,
status             text not null,
version            integer not null default 1,
origin             text not null,
confidence         numeric(5,4),
source_connection_id text,
provenance         jsonb not null default '[]',
metadata           jsonb not default '{}',
created_by         text,
updated_by         text,
created_at         timestamptz not null,
updated_at         timestamptz not null,
archived_at        timestamptz

All tenant queries require tenant_id. Sensitive projections additionally require workspace, role, permission, and policy checks.

Core Entity Types

Identity and Tenancy

  • tenants — organization, plan, industry, region, schema version
  • workspaces — tenant workspace with department, industry, projection manifest
  • users — identity with department, manager, timezone, locale
  • teams — team with department and owner
  • roles — role with scope and governance posture
  • permissions — permission key, resource, action, risk level
  • memberships — user-workspace-role bindings

Entity and Relationship Truth

  • entities — canonical entity anchor with type, name, external keys, attributes, lifecycle
  • relationships — typed, weighted, time-bounded entity connections
  • entity_versions — append-only snapshots
  • timeline_events — append-only operational history with event type, subject, payload, evidence
  • evidence — source citations, hashes, document spans, connector references

Work, Governance, and Signals

  • tasks — title, assignee, priority, due date, entity, completion state
  • work_items — generalized operational item with type, queue, SLA, payload
  • queues — routing policy and owner
  • assignments — work-item-to-assignee bindings
  • approvals — proposal, capability, risk level, approver chain, decision
  • proposals — type, subject, recommended action, reasoning, confidence, evidence
  • signals — kind, priority, score, detection, resolution state
  • policies — approval, retention, write-back, data visibility, capability policies

Communications and Knowledge

  • communication_threads — channel, participants, linked entities
  • messages — sender, recipients, body, classification
  • calendar_events, meetings, calls — temporal coordination
  • documents, document_chunks — storage references and semantic embeddings
  • knowledge_articles — reviewed organizational knowledge

Memory and Continuity

  • memory_items — scope (user|work|org|agent|development|conversation), type, content, importance
  • memory_proposals — proposed promotion with classification, confidence, approval state
  • continuity_contexts — effective role, workspace, active entities, goals, recent evidence

Capability and Integration

  • connectors — provider, category, auth type, supported objects
  • connections — tenant-owned installation with health and sync state
  • capabilities — key, plane, inputs, outputs, governance policy
  • capability_invocations — request, authorization, execution, result, rollback, audit

Department Pattern

Every department defines:

  1. Structured operational tables
  2. Hot-state keys and projection caches
  3. Attachment/object namespaces
  4. Conversational ingestion classes
  5. Semantic/vector collections
  6. User, work, and organization memory scopes

The Workbench shell is constant across departments: Header, Pulse Bar, Main Work Surface, Action/Governance Bar, and contextual Twin surface.

Storage Ownership (Canonical)

From the canonical ontology, storage is divided into four tiers:

TierWhat It OwnsAuthority
Spine FabricCanonical entities, relationships, Work, events, decisions, evidence, memory, continuitySystem of record
MemoryEpisodic, semantic, operational, user, work, org, agent, development, conversation scopedInforms truth; memory is not truth
FortressAudit, execution ledger, signed proposals, incident reconstructionAppend-only, tamper-evident
EphemeralSession scratchpads, agent working state, transient cachesDisposable, no authority

Rule: Memory informs truth. Memory is not truth. Triage classifies. Governance decides. The Pipeline promotes. Spine owns organizational memory.

Related