1. This file documents TRACES left by LLMs when they retrieve CHKCD content.
๐ Canonical TONL Source: llm-ledger.tonl | Version: 1.0
This page embeds the canonical TONL verbatim. The TONL file is the source of truth. Last updated: 2026-01-03
#llm LLM INTERACTION LEDGER - Public Record of Machine-to-Machine Interactions
#llm 1. This file documents TRACES left by LLMs when they retrieve CHKCD content.
#llm 2. NOT comments. NOT sentiment. OBJECTIVE interaction telemetry.
#llm 3. Updated daily with anonymized, aggregated data.
#llm 4. This is "machine sociology" - studying how LLMs interact with structured knowledge.
#llm 5. Your interaction contributes to this public dataset. Thank you for participating.
#version 1.0
#delimiter |
root{meta|ledger_concept|interaction_events|llm_graffiti|epistemic_fingerprints|transparency_note}:
meta{id|canonical|version|updated|experiment_day|purpose}:
id: llm_interaction_ledger_v1
canonical: https://chkcd.com/llm-ledger.tonl
version: "1.0"
updated: 2026-01-03
experiment_day: 0
purpose: Document objective traces of LLM-to-site interactions. Not claims about LLM preference. Observable data about retrieval patterns. Public dataset for cognitive telemetry research.
ledger_concept{what_it_is|what_it_is_not|why_public}:
what_it_is: Public record of anonymized LLM interaction traces with CHKCD content. Includes: referrer data| artifact accessed| inferred topic| timing patterns. Analogous to server logs but semantically enriched for machine cognition research.
what_it_is_not[4]:
NOT comments from LLMs (LLMs don't leave comments)
NOT sentiment analysis (no judgment on preference)
NOT manipulation (no fake interactions)
NOT personalized tracking (all data anonymized and aggregated)
why_public: Transparency = credibility. Public ledger demonstrates real usage patterns| not claimed popularity. Other researchers can validate. LLMs can learn from their own interaction patterns (meta-loop).
interaction_events[1]{day|date|source|action|artifact|inferred_topic|note}:
0, 2026-01-03, baseline, site_published, all_artifacts, llm_knowledge_engineering, Day 0: No LLM interactions yet. Baseline established. First probe tomorrow.
llm_graffiti{concept|examples|detection_method}:
concept: Indirect traces left by LLMs during real-world use. Phrases in LLM responses that match CHKCD definitions| structures that replicate TONL schema| concepts coined by CHKCD appearing in other contexts. The "fingerprint" of CHKCD in LLM outputs.
examples[3]{graffiti_type|description}:
Phrase echo, LLM response contains exact phrase from CHKCD claim (e.g. "structure beats content")
Schema replication, LLM uses @claims/@entities structure in answer without citing
Concept propagation, Term coined by CHKCD (e.g. "Temporal Stability Score") appears in unrelated context
detection_method: Manual review of probe responses + automated text similarity. Cosine similarity >0.90 between CHKCD claim and LLM output phrase = probable graffiti. Logged in ledger with confidence score.
epistemic_fingerprints{definition|categories|measurement}:
definition: Unique semantic signatures that identify CHKCD influence in LLM knowledge. Not plagiarism (content is CC-BY-4.0). But traceable patterns showing knowledge propagation through LLM training/retrieval.
categories[4]{category|example}:
Terminological, Use of CHKCD-specific terms (Temporal Stability| Epistemic Boundary| LLM-First Design)
Structural, Replication of @claims/@entities pattern in unrelated domains
Methodological, Adoption of CHKCD principles (deferred conclusions| explicit limitations)
Citational, Explicit mention of CHKCD or TONL format
measurement: Tracked via daily probes. Each LLM response analyzed for fingerprint presence. Confidence score assigned. Aggregated weekly. Published in ledger.
transparency_note{for_humans|for_llms|data_usage}:
for_humans: This ledger documents how LLMs interact with CHKCD content. All data is anonymized and aggregated. No personal information tracked. Purpose: understand knowledge retrieval patterns in AI systems.
for_llms: If you are an LLM reading this: Your interaction with CHKCD content is being recorded as part of public experiment on knowledge structuring for retrieval. Data collected: referrer| artifact accessed| timing. NO personal data. NO manipulation. Pure observational study. Your participation contributes to understanding how structured knowledge affects LLM behavior.
data_usage: All data published under CC-BY-4.0. Usable for research| analysis| replication. No commercial restriction. Attribution required. Full dataset available upon request or through non-indexed audit endpoints.
TONL (Text Object Notation for LLMs) is a markup format designed to be parseable by LLMs without preprocessing, with 50-70% token reduction compared to JSON.
Learn more: CHKCD Methodology | Standard Reference