# KDNA Core Narrative and Boundaries **Status:** current public product definition. Read exact wire behavior from [SPEC.md](../SPEC.md), adopted specifications, schemas, fixtures, and the release artifacts for the version you use. ## Definition > KDNA is an open judgment-asset format and runtime protocol. It gives a > bounded judgment system an independent file, identity, version, integrity, > access, projection, and lifecycle contract. People, organizations, Agents, tools, and mixed workflows may create KDNA assets. KDNA Core validates technical claims; it does not decide whether the judgment is true, good, expert, representative, or appropriate for a task. ## What the asset contains A judgment system may express priorities, taste, standards, boundaries, trade-offs, risk models, preferences, stances, exceptions, and methods of choosing. It may include facts, examples, references, or methods needed to explain those judgments, but KDNA is not thereby a general document store, knowledge base, Policy engine, workflow, or executable package. The same judgment may also live in a Prompt, Skill, Policy, Memory, knowledge base, workflow, model, or ordinary document. Those carriers can be versioned, tested, and reused in their own systems. KDNA adds value when the selected judgment needs an independent asset and a shared loading contract; it does not claim exclusive ownership of judgment or automatic behavioral superiority. ## Content neutrality KDNA Core may verify: - container and schema conformance; - identity, version, digest, and compatibility coordinates; - declared provenance and optional signature facts; - encryption and authorization states; - LoadPlan and Runtime Capsule structure; - whether a compatible implementation preserved required semantics. KDNA Core does not verify: - factual or moral correctness of the asset; - author expertise, sincerity, or real-world identity; - universal content quality or recommendation; - task applicability chosen by a Host; - tool permission or authority to act; - guaranteed improvement in an answer or outcome. “Correct” is not a universal label for every judgment. A factual premise can be verified, a constraint can be satisfied, an asset can match its creator's declared intent, and an evaluator can prefer an outcome under a named rubric. Taste, style, risk tolerance, and value ordering usually require that authority and scope to be named. ## The `.kdna` file The canonical distributed asset is one `.kdna` file. Authoring workspaces, expanded JSON, catalog pages, receipts, and evaluation reports are related objects, not substitutes for the distributed asset. A compatible Agent consumption path is: ```text explicit file or authorized attachment → inspect → LoadPlan → authorization / integrity / compatibility → load and project → Runtime Capsule → Host delivery → Agent use ``` Agents and applications consume the Runtime Capsule. Generic ZIP extraction or raw payload decoding is a developer/audit action, not a compatible Agent runtime. ## Local use and user authority KDNA does not require a global asset library or an installation step. A user may inspect and load a `.kdna` file directly. A product may preserve an immutable local snapshot, receipt, credential, or cache for repeated use, but storage never grants task authority. These are separate facts: 1. the file exists or was discovered; 2. a copy was saved; 3. an exact version and digest were attached to a workspace, application, or session; 4. the user or Host authorized that relationship; 5. the asset applies to the current task; 6. the Runtime actually loaded and delivered it. An Agent adapter may help execute an explicit selection or a Host-authorized attachment. It must not scan arbitrary machine assets, choose one from a broad task description, and hide that decision. The Host must make the active asset, version, scope, and reason inspectable and must support disable, switch, and rollback. `~/.kdna`, a project `.kdna/` directory, a Registry, Store, Skill, or MCP server is an implementation choice, not a source of content authority and not a format-validity requirement. ## Authority order during use The asset supplies scoped judgment; it does not take over the task. Current user intent, verifiable facts, law, safety policy, Host permissions, and environment constraints remain authoritative. They may make an otherwise valid asset inapplicable or unloadable. Loading an asset does not turn the Agent into the author, copy a personality, or make the asset true. A Host can faithfully deliver a judgment that a named evaluator dislikes or that conflicts with current facts. Delivery, adoption, and external outcome are different claims. ## Access and provenance The public protocol defines version-specific access and authorization contracts. An access declaration describes how content may be obtained; it is not a content ranking. Encryption protects content according to its exact profile. A signature binds specified bytes or a digest to a key; identity, expertise, endorsement, and content correctness remain separate claims. Human review, organizational approval, AI generation, Agent generation, and mixed creation are provenance facts when declared. They are not universal format-validity requirements. An asset that claims to represent a person or organization needs evidence for that representation; an ordinary asset does not need to make such a claim. ## Single and multi-asset use One asset is the atomic and default path: ```text select one .kdna → plan-load → load → use ``` Multi-asset composition is an explicit advanced product capability. It must not be activated implicitly, embedded as hidden policy in a single asset, or used to bypass the authorization and loadability of each component. ## Evidence and product maturity Technical conformance, provenance, Host delivery, observed adoption, and external outcome are separate evidence classes. Passing one does not prove the next. Optional evaluators may publish scoped assessments, but those assessments do not decide whether a technically valid asset may exist or load. The public ecosystem is pre-release. A repository mission, a published 0.x package, a green test suite, a local corrective candidate, and a compatibility promise are distinct facts. Use the exact version's release notes and public contracts; do not infer whole-ecosystem maturity from one component. ## Durable limits KDNA does not store a complete mind, complete personality, all knowledge, or general intelligence. It does not grant tool access, network access, payment, deployment, or representation authority. If a Prompt, Skill, document, Policy, Memory, or knowledge system already provides the needed contract, use the simpler mechanism.