Editor's Draft · Public Draft 2026-06

KP:1

An open, text-native format for packaging epistemic state — claims, confidence, evidence, contradictions, and the trace of how beliefs evolved, preserved as structure.

Read the specification GitHub repository

v0.8.4-preview 2026-06-13 CC BY 4.0 DOI 10.5281/zenodo.19445262

Abstract

KP:1 is a text-native specification for representing epistemic state — claims with confidence, evidence, contradictions, and the trace of how beliefs evolved — designed for AI inspectability and portability across systems. The format defines a strict normative core (CORE.md, JSON Schema, PEG grammar) plus topic-authoritative companions for voice surfaces, multilingual support, archival, composition, and extension blocks.

What makes KP:1 distinct

Most formats flatten knowledge into settled facts. KP:1 keeps its shape: the uncertainty, the tension, and the history stay as addressable structure. The clearest way to see that is the format itself.

claims.md — acquisition review 3 claims self-describing 
<!-- KP:1 — self-describing header: a reader bootstraps the format from the file
     - [ID] assertion  {confidence|type|evidence|date|depth|nature}   (depth, nature optional)
     type   o observed · r reported · c computed · i inferred
     rel    → supports · ⊗ contradicts · ⊗! error · ⊗~ tension · ⊘ supersedes -->

- [C005] Dating follows the catalogue raisonné: 1962, not 1957.
  {0.93|r|E003|2020-04-30|exhaustive} ⊗!C004

- [C014] The 2025 appraisal values the work at €950,000.
  {0.91|r|E008|2025-12-15|investigated} ⊘C013

- [C017] The 1953–1962 provenance gap is a material risk.
  {0.45|i|E005,E010|2026-05-09|investigated|judgment} ⊗~C019
C017 ⊗~ C019 A judgment held at 0.45, on two cited sources, kept in productive tension with the recommendation, not rounded away. Every line stays addressable, and the header lets any reader read the notation without fetching a schema.

  • Typed contradiction qualifiers

    Disagreement is kept in tension: plain contradiction, known error, or productive tension. Nothing is flattened away.

  • Voyager-Principle Rosetta header

    Zero-dependency parsing: a reader can bootstrap the format from the file itself, with no external schema to fetch.

  • Supersession with append-only history

    When a belief changes, the prior claim is superseded, not deleted, so how the thinking moved stays legible.

  • Three-surface architecture

    One source of truth, three reads: claims for AI reasoning, display views for visual reading, voice views for spoken delivery.

Read the specification