The operating model
A knowledge operation an executive can actually own.
The doctrine argues why a governed knowledge loop is necessary. This page is the working layer above the instruments: the roles, the gates, the cadence, and the one number the program answers to — written for the person who owns the budget, not the person who writes the articles.
Everything below is in daily practice use, and every claim on this page has a matching machine-readable record in the dark rail. That is the discipline practicing itself.
Four roles. One of them is yours.
The loop runs on effort budgets, not job descriptions. The design targets are deliberate: if capture costs an agent more than thirty seconds, capture dies — everything else in the model is arranged so it never does.
| Role | Owns | Effort budget |
|---|---|---|
| Agent | Flagging a resolved case worth keeping — AI drafts, the agent judges | ≤ 30 sec / capture |
| Curator | The gate: approve, edit-approve, merge, or reject with a reason code | ≤ 3 min / clean approve |
| Knowledge owner | The record: lifecycle states, review cadences, the terminology standard | a standing role |
| Executive sponsor | The number — blended cost per answer — and the authority the gate borrows | one review / month |
Two to three trained curators are enough to pilot. The executive's role is not ceremonial: the gate holds because someone with budget authority says it holds — that is what ratifying the governance policies means.
Trust is a gate, not a hope.
Three gates cover the record's whole life: new-candidate (does this enter the record?), update-DIFF (does this change stand?), and retirement (does this leave honestly?). Every pass ends in one disposition — approve, edit-approve, merge, or a coded rejection — and every disposition lands in an append-only decision log.
The rule that makes AI drafting safe is one sentence: nothing probabilistic writes to the system of record without passing a human gate. AI models come and go at the edges; the record compounds in the middle.
The gate also measures itself. Edit-distance — how much a curator changed a draft before approving it — is the drafting-quality benchmark, trended over time. Reason codes turn rejections into tunable telemetry instead of friction.
The cadence that keeps it alive.
Knowledge operations fail on rhythm before they fail on tooling. The model runs on four clocks: capture at resolution (the article is flagged while the case is still warm), publish within one business day of capture, review on volatility tiers — evergreen, medium, and high-volatility content each get their own re-verification cadence — and placement quarterly, re-scoring the program against the maturity model so the roadmap answers to evidence.
Standing it up is a 90-day arc, not a big bang: foundations first, then a pilot with templates live, prompts wired to real ticket exports, and the gate running with two or three trained curators — then scale on what the pilot proved.
One number on the table: blended cost per answer.
The program answers to a single figure an executive can defend in a budget meeting — total answer spend divided by total answers delivered, across every channel that answers questions. The full argument is the Answer Economics essay; the working instrument is the measurement workbook.
Two supporting ratios keep it honest: the shift-left ratio (what share of answers is delivered cheaper than assisted support) and the known-vs-new mix (how much of the assisted queue is questions the record already answers — known issues leaking back in are a findability failure, not a demand signal).
Brief your CFO on the inversion first
As shift-left succeeds, assisted-queue averages — time to resolve, complexity, cost per assisted answer — get worse by design, because the easy volume left the queue. Put that expectation in the deck before someone reads the trend as failure.
Where to start.
Placement precedes prescription. The maturity model places your program in thirty evidence-gated questions — your level is your weakest dimension, and the shortest defensible path to the next level is the roadmap. The Toolkit packages every instrument this page references — templates, prompts, gate checklists, ratifiable policies, and the workbook — as working files.
The instruments — the Toolkit Read the doctrine
The doctrine, this page, and the maturity model are free — no email required. That is deliberate: the reading room holds all of it in one place.