Knowledge Pack 1 (KP: 1) is a plain-text format for packaging epistemic state — claims with explicit confidence, evidence, provenance, relationships, and contradictions. A Knowledge Pack is a directory of human-readable Markdown files that records what a person or an AI system believes, on what evidence, with what confidence, and in tension with what else, as understood at a particular moment in time. Each claim carries an identifier, a confidence value, references to its evidence, a since-date, an assessment depth, and an evidence nature. Each relationship between claims (supports, contradicts, supersedes, refines, see-also) is typed. The result: epistemic state is encoded in the document, not reconstructed from prose by the reader — whether that reader is a human or a parser. Why a new format? Existing formats stop short. RDF and JSON-LD interchange machines beautifully but aren't meant to be read by humans. Knowledge graphs and argumentation frameworks capture structure but not the same bundle of provenance, time, and confidence conventions. Recent "context docs" conventions like llms. txt and AGENTS. md index content for AI consumption but stop at indexing — they don't say how knowledge itself should be represented. KP: 1 picks up where they leave off. This editor's draft, v0. 8. 0-preview, dated 2026-05, succeeds the v0. 7-preview public release (2026-04). It contains: CORE. md — the implementable normative core (the implementer surface) SPEC. md — the full normative specification Twenty topic-authoritative companion documents covering voice surfaces, composition, lifecycle, multilingual support, organization, storage, archival, extensions, playback, reconciliation (stub), authoring rubrics, and rationale AGENTS. md at the repository root — an AI-first task-routing file that tells a cold AI agent which spec files to load for which task (parse, author, reconcile, translate, compose, playback) A concept-by-concept mapping to RDF/JSON-LD, PROV-O, and Nanopublications, with each translation graded clean, lossy, or impossible A PEG grammar and a JSON Schema A nineteen-test conformance suite (fifteen fixtures — six valid, nine invalid — plus four reference example packs, all auto-validated) Four worked examples: a market-analysis pack, a self-assessment of KP: 1 written in KP: 1, and two complementary art-acquisition decision-support packs (buyer-side and consigner-side) that exercise the full feature surface end-to-end What's new since v0. 7-preview: Three normative lanes — CORE. md, SPEC. md, and topic-authoritative companions all carry normative weight; companions are not mere extensions of CORE/SPEC, they are the normative authority for matters within their domain AI-first packaging — AGENTS. md task-routing file plus spec/AUTHORING. md producer-side decision rubrics with worked examples and anti-patterns Bifurcation — spec/RATIONALE. md (new informative companion) houses Design Principles, Style Systems rationale, and the Relationship to Existing Standards table previously interleaved into normative prose Two new reference examples — art-acquisition-decision. kpack (buyer-side, walks the AUTHORING. md rubrics end-to-end) and auction-house-consignment-review. kpack (consigner-side decline path with cross-pack ↔ references) SC-12 (new semantic constraint) — predictions must have confidence ≤ 0. 95; predictions about future states have irreducible uncertainty and the 0. 99+ band is reserved for trivially-falsifiable claims Discoverability — new optional specᵤri and specᵥersion manifest fields let a sealed. kpack archive self-describe to a cold receiver Tooling — a --pack PATH flag for validating arbitrary single packs, a Makefile convenience wrapper, and cross-pack reference parsing in the conformance runner No core schema break: pre-v0. 8. 0 packs continue to validate. The specification is edited by Timothy Kompanchenko under an editor's-draft governance model. The preview phase is feedback-only: normative pull requests against the spec text, grammar, schema, and conformance fixtures are not accepted. Comments, implementation experiments, and adversarial critique are welcome through the public issue tracker at github. com/tymofiy/kp. A nonprofit foundation is in formation and will take over governance from the single-editor phase. Specification text is licensed under CC-BY-4. 0. Code, schemas, and examples are licensed under Apache-2. 0. KP: 1™ is a pending US trademark (USPTO Serial No. 99747548, Class 9).
Building similarity graph...
Analyzing shared references across papers
Loading...
TIMOTHY KOMPANCHENKO
Building similarity graph...
Analyzing shared references across papers
Loading...
TIMOTHY KOMPANCHENKO (Sat,) studied this question.
www.synapsesocial.com/papers/6a01720a3a9f334c2827217f — DOI: https://doi.org/10.5281/zenodo.20100689