KVN — how Kavanah frames work

KVN is the methodology Kavanah is built around. Every delegation carries three axes: Know-how (what capacity and capability the work needs), Vision (what the finished thing looks like), and Negation (the hard nos — what must not happen). Most failed delegation is a missing axis, usually Negation.

The three axes

Know-how — the skills, access, budget, and time the work actually requires. Naming it surfaces the gap between what you are asking for and what the person can do today.
Vision — the end state, described so someone else could recognise it. Not the task list; the finished result.
Negation— the constraints and explicit non-goals. "Don't touch the pricing page." "No new dependencies." This is the axis people skip, and it is the one that prevents rework.

Three scopes

KVN applies at three levels, and each inherits from the one above it.

1Workspace charter (/workspace-kvn) — the standing brief for the whole company. Written once and revised rarely. Admins edit it; everyone can read it.
2Project charter — per project, and it adds false friends: things that look in-scope for this project but explicitly are not.
3Task— each task carries its own K/V/N. The create and edit forms have a Generate button that drafts all three from the task's title and description plus the workspace charter.
4The North Star inherits down the same three scopes — see below. It is structured rather than prose, so it is not a fourth axis; it is what makes the V above it checkable.

The North Star — the number that makes Vision falsifiable

Vision is prose. Prose cannot tell you, on a Tuesday, whether the task in front of you moves the company toward it. The North Star can: one metric, a direction, a target, and a date to hit it by. You set it on the same page as the charter (/workspace-kvn), and on any project that has a leading indicator of its own.

It inherits the way the axes do. A project without its own metric is not a project without direction — it points at the company's. A task never carries a metric of its own; it carries one line saying how finishing it moves the metric it inherits. If it does not move it, leave that line empty. An empty one is a visible fact; an invented one is not.

The North Star is advisory, and that is deliberate. The Negation refuses; the North Star points. Nothing is ever declined for failing to move a number — a metric with a veto is how teams learn to game the metric. What changes is what the agent says: when it proposes, prioritises or sequences work, it names the effect on the number, and when there is no effect it says so instead of implying one.

Measured by hand or automatically. Most north stars are numbers only you have, so you record them and the trend builds up. If the metric is something Kavanah already knows — completion rate, open tasks, overdue tasks, throughput — it is re-measured for you once a day, and a verdict that gets worse routes an escalation once, on the transition, not every morning.
The honest scoreboard is coverage. Ask the agent about the north star and it tells you how many open tasks actually say how they move it. A metric no live work references is decoration — the same test the negation ledger applies to a Negation with zero hits.

The Negation is enforced, not advisory

The workspace charter is injected into the AI agent's system prompt on every conversation — but the Negation axis is more than context. It is a gate. Before the agent creates a task or a project, the work is checked against your charter. Work that crosses the Negation is refused before it is created, and the agent tells you the exact line of the charter that refused it.

This is the cheapest control you have over the agent. If it keeps proposing work you have to reject, the fix is one sentence in Negation — not a longer chat.

False friends

The gate also catches the subtle case: work that looks like it serves your Vision but actually serves a proxy for it. If your Vision says "users reach first value in one session" and the agent proposes an extra onboarding step because it lifts signup numbers, that is a false friend — it moves a metric, not the mission. It gets refused and cited like any other collision. You can also name false friends explicitly on a project charter so the agent refuses them by name.

Nothing is silent

The agent cannot quietly reword the request, drop the offending item, or reach for a different tool to get the same work in. It has to tell you. Then there are three honest moves: drop the work, reframe it, or change the charter — because sometimes the charter is what is out of date.
If you hear the refusal and still want it done, you can override. The override is recorded too.
A refusal never blocks you when the check itself fails. If the model is unavailable, the work is allowed through — this is a guard rail, not a lock.

What we said no to

Every refusal is logged, and /workspace-kvn shows the ledger: how much work the charter refused, how much was overridden anyway, and — the useful part — which lines of your charter are actually doing the work.

Read that list at your charter review. A Negation with hits is earning its keep. One with none is either irrelevant or invisible, and you should decide which. One that keeps getting overridden is a rule your team has quietly stopped believing in — either enforce it or delete it.

This is the thing Kavanah does that other tools do not. Every project tool measures what your team shipped. Kavanah also measures what it correctly declined.

Asking before you commit

You do not have to wait for a refusal to find out. Ask the agent "should we do X?" and it checks the idea against the charter without writing anything. Ask it whether a project has drifted off-mission and it audits every open task in that project against the charter and reports the ones that no longer fit.

Not sure how to start an axis? Each editor has an AI Generate button that drafts one axis from the other two plus your projects and clients. It drafts — it never saves — so you always review before it becomes the charter. Note that the gate only switches on once you have written a Negation: an empty charter refuses nothing.

Request Intake Pages NextKnowledge Base