Knowledge Substrate Core
The governed knowledge layer that sits beneath both AI agents and the people who make decisions, unifying licensed, organisational, external, and personal content into one trust-scored base.
Causal Substrate Core
The typed-knowledge graph that governs and proves: knowledge items in families, the causal and dependency edges between them, the trust tiers, and the scores. It traces why an answer holds and what would cascade if a source were retracted.
Retrieval graph Core
The graph used for grounded retrieval, holding Documents, Chunks, and Entities. Chunks carry the embeddings; retrieval is hybrid (vector plus full-text) with a short graph expansion.
Corvair Ingestion Pipeline Core
The batch, idempotent pipeline that brings a source in and prepares it: collect, parse, classify, chunk, type and extract, embed, write, then catalog and score. It populates both graphs and feeds scoring and the warrant.
The two graphs Core
The retrieval graph and the Causal Substrate, co-located so one query grounds its words on the first and proves its trust on the second.
Recipe Core
A declarative file that states a base's intent, scope, sources, trust, freshness, access, and curation policy. Knowledge as code: hydrate it to get a running, governed base.
Knowledge as code Core
The principle that a base is defined by a versioned recipe and reconstituted reproducibly, auditable and portable like software.
Hydrate Core
To build a running, governed base from a recipe against the sources available at that moment.
Snapshot Core
An immutable, sealed point-in-time state of a base. Snapshots can be diffed, and a new one promoted by an atomic blue/green swap.
Blue/green swap Core
Promoting a new snapshot atomically so the served base never shows a half-built state.
FrameGraph Core
Applying a selectable analytical framework (5W1H, PESTEL, SWOT, Ecosystem, and more) over a topic graph to produce a navigable, reusable, exportable view.
Topic graph Core
The entity-and-relationship layer a FrameGraph is built over; the structure the Topics capability walks.
Virtual Expert Capabilities
The domain assistant that is the face of the substrate for most users: it answers from one governed base, grounded and warranted, and is reached identically by people and agents.
Ask Capabilities
A plain-language answer grounded only in the base, carrying a premise chain, sources and tier, and a Validity Warrant.
Search Capabilities
One semantic query across everything a user is allowed to see, ranked and attributed.
Deep Research Capabilities
A multi-pass investigation that clarifies the objective, plans and reasons in steps, and returns a structured, cited report with a warrant.
Topics Capabilities
Following the topic graph to see what a base knows and where it is thin, filtered by geography, framework, time, or sector.
News Capabilities
Monitored news on chosen subjects, each story tied to its source and tier, with a digest and a path into an inquiry.
Vault Capabilities
A user's own files and saved results, governed under the same model and served only to them unless shared.
Cases Capabilities
A raised gap, dispute, or help request routed to a steward, so weaknesses are fixed rather than worked around.
Validity Warrant Governance
A signed, replayable record attached to every answer and governed decision, in five parts: question, knowledge, reasoning, authority, and signature. It verifies offline after the fact.
Premise chain Governance
The traced path from question to answer: which claims rest on which sources, with per-item validity and gravity.
Trust tiers (T1 to T4) Governance
The classification carried by every source and item: T1 authoritative, T2 reliable, T3 contextual, T4 unverified. The tier travels into every answer.
Validity Governance
A score for how current and dependable an item is, with a decay over time unless refreshed, so stale material is visibly stale.
Decay / half-life Governance
The rate at which an item's validity falls; the half-life sets when it decays to half its score and when it is due for re-validation.
Epistemic gravity Governance
A measure of how load-bearing an item is: how much would cascade if it failed or were retracted.
Contextual confidence Governance
A score for an answer in the context of the specific question, not in the abstract.
Veracity gate Governance
A screen at ingestion and answer-time that refuses to let low-trust or contradicted content silently ground an answer.
Provenance Governance
The recorded origin of an item or action: source, content hash, connector, and the agent or person responsible.
Retraction & cascade Governance
Invalidating a source and propagating the effect through the dependency graph to everything that rested on it.
Trickle update Governance
Applying incremental change to a base as sources move, without a full rebuild.
Federation Governance
Posing one query to several bases at once, each answering from its own governed content, then composing the responses with attribution. Governance is not merged.
Composite base Governance
Several bases grouped into one addressable base, queried under a chosen strategy and conflict policy, while each member stays governed in its own right.
Application user Roles
Asks, searches, researches, and receives cited, warranted answers; contributes within bounds.
Knowledge steward Roles
Owns what a base knows: sources, tiers, the curation queue, scope, and the recipe. The role is judgement, not labour.
Administrator Roles
Runs the platform (workers, jobs, budgets, access) and never sees knowledge content.
Auditor Roles
Verifies after the fact: reads warrants and framework coverage, samples, and exports evidence. Read, verify, and export only.
Domain partner Roles
Authors and packages virtual-expert packs, then licenses and maintains them on subscription.
Domain pack Packaging
A prebuilt, co-branded virtual expert with named capabilities, licensed reference databases, and how-to guides, riding on the same substrate every customer already runs.
Catalog & marketplace Packaging
Where partners publish versioned, signed packs and subscribers discover and subscribe. Distribution is catalog only.
Entitlement Packaging
The subscription that grants access to licensed content; it scopes who may receive a licensed pack or OKF bundle.
Tenant / tenancy Platform
A first-class, always-present scope; every base and principal belongs to a tenant, and single-tenant is one tenant.
Control plane / data plane Platform
The operator-scoped control plane provisions tenants and never reads content; the data plane serves and governs knowledge.
Data residency Platform
Pinning all of a tenant's data and AI calls to a chosen region; a change is an explicit migrate-and-repoint.
Hybrid split Platform
The loosely-coupled split between the governed core and research-and-monitor: only a recipe goes out, only staged candidate files come back.
Circuit breaker Platform
A cap on agent activity or a cascading change that trips rather than letting a runaway operation commit.
Autonomy level Platform
The bound on what an agent may do unattended; lower levels stage changes into the approval queue instead of committing.
Merkle transparency log Platform
An append-only log of warrants with inclusion proofs, run per installation, that makes the audit record independently verifiable.
CMEK Platform
Customer-managed encryption keys: encryption at rest under keys the customer controls.
Right to erasure Platform
The governed workflow that locates and erases a subject's personal data across every store and survives backups, while the warrant chain holds via salted commitments.
OKF (Open Knowledge Format) OKF
An open, Apache-2.0 specification (Google Cloud, v0.1) that represents knowledge as a directory of markdown files. In CKS it is the system of record and the canonical ingestion representation.
Bundle OKF
An OKF directory tree of markdown files; the unit of distribution. In CKS, one bundle per source.
Concept / Concept ID OKF
One markdown file; its identity is its file path minus .md. In CKS, one concept per extracted claim or entity, plus the source itself.
x-cks OKF
The namespaced frontmatter block that carries CKS governance (trust tier, validity, provenance, warrant, classification, entitlement, typed relationships) while keeping the bundle OKF-conformant.
Source key OKF
Canonical URI plus content hash plus version; the bundle identity and the idempotency key for re-ingestion.
Collection / Source / Concept OKF
The three-level boundary model: a collection groups source bundles; a source is the smallest citable artefact that gets a bundle; a concept is one claim or entity within it.
System of record vs projection OKF
The OKF bundle store is the durable record; the graph and search indexes are rebuildable projections (caches) regenerated by replaying the loader.
Round-trip import OKF
Importing an external OKF bundle directly through the same loader, skipping parse and extract, then scoring and warranting it.
Conformance OKF
A bundle is conformant when every non-reserved file has parseable frontmatter with a non-empty type. CKS bundles stay conformant even with x-cks present.
MCP cover API
The seven lean MCP servers that present a curated, governed subset of the API as tools, generated from the OpenAPI contract so they cannot drift.
Safety class API
The label on every operation (read-only, preview, mutating, destructive) that tells a client when to confirm or preview first.
Persona API
The role resolved from a caller's token (business user, steward, admin, compliance, partner author, agent, subscriber) that gates each operation.
Preview and apply API
The two-step pattern for decision-bearing changes: a preview returns an impact diff and a token; a separate apply commits, binds to the token, writes provenance, and emits a warrant.
Idempotency key API
A header that makes a mutation safe to retry: a replay returns the original result; a replay with a different body is rejected.
AnswerEnvelope API
The query response: the assembled answer together with its citation list, from which the trace and warrant can be fetched.