CKF
Specification · v0.2 · experimental

The Protocol

CKF is a portable, textual, extensible knowledge representation protocol. It is human-auditable, AI-optimized and source-traceable.

Design Principles

Portable
Human-auditable
AI-optimized
Source-traceable
Modular
Versionable
Convertible
Agent-ready
Graph-compatible
Retrieval-ready

File Extension

The canonical extension is .ckf. Recommended encodings: .ckf.md (Markdown), .ckf.yaml (YAML), .ckf.json (JSON).

Markdown CKF
YAML CKF
JSON CKF

Core Objects

  • — Package Metadata
  • — Knowledge Nodes
  • — Relations
  • — Rules
  • — Procedures
  • — Triggers
  • — Chunks
  • — Atomic Units
  • — Agent Instructions
  • — Source Traceability

Anatomy of a .ckf package

Every .ckf package is a structured collection of cognitive blocks: a metadata header (identity, version, source, language and quality scores) plus 22 typed sections numbered below. Each section is independently useful and can be consumed by retrieval systems, agents, knowledge graphs or fine-tuning pipelines.

01

Core Intent

The primary purpose, intended user, intended agent use and transformation goal of the source.

02

Domain Map

Main domain, subdomains, adjacent domains and explicitly excluded domains.

03

Entity Graph

People, roles, tools, concepts, metrics, states and objects extracted from the source.

04

Concept Graph

Key ideas as connected semantic nodes with dependencies, contradictions and supporting relationships.

05

Principles

Foundational statements with conditions of application and rationale.

06

Heuristics

Tacit expertise turned into practical rules of thumb for agent reasoning.

07

Decision Rules

Conditional decisions with required context, output actions and failure modes.

08

Procedures

Step-by-step processes with inputs, outputs, success and failure criteria.

09

Patterns

Recurring situations, signals, mechanisms and response strategies.

10

Anti-patterns

Failure modes with warning signals and replacement behavior.

11

Causal Chains

Cause, mechanism, effect, secondary effects and intervention points.

12

Contextual Triggers

When-then activations: signals that should make agents recall specific knowledge.

13

IF-THEN Rules

Compact conditional logic with rationale and confidence.

14

Exceptions

Edge cases that override general rules with modified actions and explanations.

15

Mental Models

Reusable frames of reasoning, with usage and non-usage conditions.

16

Operational Playbooks

End-to-end agent runbooks: objective, steps, tone, tools and failure modes.

17

Agent Q&A

Reference question/answer pairs for evaluation, tutoring and onboarding.

18

Retrieval Chunks

Standalone, dense, context-rich units optimized for semantic search.

19

Atomic Units

One-idea statements suitable for embeddings, evaluation and memory systems.

20

Agent Instructions

Behavior, reasoning, response, forbidden actions and tool guidance.

21

Knowledge Limits

Missing context, weakly supported claims, biases and items needing human review.

22

Source Traceability

Links every generated item back to its original source location and excerpt.

Detailed per-section technical reference: /docs/schema.

Source Basis Labels

Every extracted item must be marked with one of the following labels.

explicit

Directly stated in the source.

inferred

Reasonably inferred from source content.

synthesized

Created by combining multiple source fragments.

author_opinion

Represents the source author's subjective claim.

uncertain

Weakly supported or ambiguous.

Comparison

Implicit prose vs labeled claims

Source basis labels exist so every claim carries its own evidence type. Compare a human paragraph (sources implied) with a .ckf fragment (sources tagged).

Good for humans

Good for humans

According to the internal manual, refunds above $500 generally require manager approval. The team also believes that responding within 2 hours improves retention, though this hasn't been formally measured. Some agents add that loyal customers tend to forgive delays — anecdotally.

.ckf · Labeled and traceable

.ckf · Labeled and traceable

explicit
inferred
synthesized
author_opinion
uncertain
confidence
claim: refunds > 500 require manager approval
  source_basis: explicit
  confidence: 0.96
  source: internal_manual#sec-4.2
claim: response under 2h improves retention
  source_basis: inferred
  confidence: 0.55
claim: loyal customers forgive delays
  source_basis: author_opinion
  confidence: 0.40
claim: response time correlates with NPS
  source_basis: synthesized
  confidence: 0.62
claim: agents share this folklore
  source_basis: uncertain
  confidence: 0.30

Confidence Scores

Use a 0.00–1.00 scale.

0.95–1.00Directly stated and central.
0.80–0.94Strongly supported.
0.60–0.79Reasonable inference.
0.40–0.59Weak hypothesis.
0.00–0.39Uncertain or speculative.

CKF v1.0 for this page has not been compiled yet. Downloads become available once an admin runs the compiler.