The discipline
Every answer now has two readers.
Bookkeeping made money legible to auditors. Contextkeeping makes what your company knows legible to machines — AI drafts, humans approve, both readers consume, and the system tells you what's missing. You're reading the human rendering of this page. The other reader gets the toggle.
GET /llms.txt — served at this site's root · format: llmstxt.org
Every answer now has two readers.
# Contextkeeping > A KCS-aligned operating model that makes company knowledge > legible to machines: AI drafts, humans approve, machines and > customers consume, and the system reports what is missing. ## Doctrine - [The doctrine](https://contextkeeping.com/doctrine): why both KCS assumptions expired - [Answer economics](https://contextkeeping.com/answer-economics): cost per answer, not deflection - [The maturity model](https://contextkeeping.com/downloads/contextkeeping-maturity-model.pdf): one-page self-assessment, no email required - [The vocabulary layer](https://machinereadyknowledge.com/terminology): terminology governance — terms are retrieval keys ## Ecosystem - [Machine-Ready Knowledge](https://machinereadyknowledge.com): the four-clause article standard - [AnswerEcon](https://answerecon.com): fixed-scope corpus audit
01 / the governed knowledge loop
Three loops around a core nothing probabilistic touches.
Classical KCS ran two loops. The AI era needs three — around a system of record that compounds while AI layers come and go.
- 01CaptureA resolved case already contains the article — AI drafts it; your agent flags it in 30 seconds.
- 02CurationThe human gate: approve, edit-approve, merge, reject — edit-distance logged, quality measured.
- —System of recordVersioned, typed, owned, lifecycle-dated. Nothing probabilistic writes here without passing the gate.
- 03DeliveryCited answers from governed content, delivered to humans and machines — every unanswered question fed back as a capture request. The loop closes itself.
Measured issue-centrically: blended cost per answer across every channel — not deflection.
# the loop, as the system logs it — one article's afternoon (example events, JSON Lines)
{"ts":"2026-08-11T14:02:11Z","event":"case.resolved","case":48112,"theme":"api-keys"} {"ts":"2026-08-11T14:02:19Z","event":"draft.created","draft":"atk-2041-r3","source_case":48112,"actor":"ai"} {"ts":"2026-08-11T14:02:44Z","event":"agent.flagged","draft":"atk-2041-r3","verdict":"worth-keeping","seconds_spent":28} {"ts":"2026-08-11T15:37:02Z","event":"curator.approved","draft":"atk-2041-r3","edit_distance":0.08,"actor":"human"} {"ts":"2026-08-11T15:37:02Z","event":"record.published","id":"atk-2041","version":14} {"ts":"2026-08-11T15:41:56Z","event":"answer.cited","section":"atk-2041#steps","consumer":"copilot"} {"ts":"2026-08-11T15:44:13Z","event":"gap.logged","question":"export rate limits?","cluster":17,"action":"capture.requested"}
# the only write to the record follows the human approval at 15:37:02. Always.
# measured issue-centrically: blended cost_per_answer across every channel — not deflection.
02 / machine-ready knowledge
The new reader never sees the page.
The agent-assist panel, the help-centre chatbot, the copilot your enterprise customers wired into their own service desk — increasingly the reader of your knowledge is a machine assembling an answer for a human who never sees the article.
Machine-Ready Knowledge is the bar. We maintain it as an open standard — four clauses, free to adopt.
# example front matter of a conforming article
--- id: atk-2041 # §3 stable identifier title: Re-authorize a revoked API key applies_to: # §2 explicit applicability plan: [scale, enterprise] region: all owner: support-knowledge lifecycle: current verified: 2026-07-30 # §4 freshness signal sections: # §1 self-contained - id: atk-2041#cause self_contained: true - id: atk-2041#steps self_contained: true
03 / provenance
"Every artifact here was dogfooded in practice before it was sold. Twenty-plus years in the knowledge industry taught me one thing worth keeping: the boring core compounds."
Jason O'Donnell · founder & practitioner
# schema.org Person — embedded in this page as application/ld+json
{
"@context": "https://schema.org",
"@type": "Person",
"name": "Jason O'Donnell",
"jobTitle": "Founder & practitioner",
"description": "Twenty-plus years in the knowledge industry; ten-plus running KCS programs at scale. Every Contextkeeping artifact is dogfooded in practice before it is sold.",
"sameAs": ["https://www.linkedin.com/in/acdntlpoet"],
"worksFor": {"@type": "Organization", "name": "Contextkeeping",
"url": "https://contextkeeping.com"},
"knowsAbout": ["Knowledge-Centered Service",
"knowledge operations",
"AI-assisted support knowledge"]
}
04 / the engagement
Want it placed for you? AnswerEcon.
Fixed scope, fixed price, 2–3 weeks. Maturity placement across six dimensions, a coverage audit against your actual ticket themes, your machine-ready percentage, and a prioritised roadmap — as a Context Readiness Report and an executive readout.
Rather place yourself first? The maturity model is a one-page self-assessment — free, no email required.
# schema.org Service — embedded in this page as application/ld+json
{
"@context": "https://schema.org",
"@type": "Service",
"name": "AnswerEcon",
"serviceType": "Knowledge corpus audit",
"description": "Fixed-scope audit: maturity placement, coverage vs. ticket themes, machine-ready percentage, prioritised roadmap. Delivered as a Context Readiness Report and an executive readout in 2-3 weeks.",
"provider": {"@type": "Organization", "name": "Contextkeeping",
"url": "https://contextkeeping.com"},
"url": "https://answerecon.com"
}