> The shared canvas a knowledge workshop runs on, and the three acts that fill it. The board is a **knowledge dependency [[Graph|graph]]** read in two layers: **shape says what a thing is** - a **card** is a piece of knowledge, a **frame** is a container that holds it - and **colour says what state it is in**. It turns one idea into a working diagnosis: a piece of knowledge is pinned down by [[Ostension|ostension]], [[Intention|intension]], or [[Extension|extension]], and the organisation's job is the climb toward a written rule. Built for a live board (Miro/FigJam) and exportable to [JSON Canvas](https://jsoncanvas.org/). ^definition An [[AI agent|agent]] is the consumer that cannot be pointed at anything - it needs every piece it depends on written, reachable, and machine-readable. So the board is filled in three acts, each asking one question and handing its answer to the next: **what does the question mean**, **where does each piece live and in what state**, and **what do we do about the gaps**. ## The lens There are three ways to pin down anything an organisation knows: ![[Kinds of definition#^lens]] Most organisations are stuck partway up - the answer lives in someone's head, or scattered across past cases nobody has distilled into a rule. The whole workshop is the climb to a **written rule** an [[AI agent|agent]] can read and apply to a case it has never seen. The board makes that climb visible, one prompt at a time. ## Dependencies *What does the question mean?* The first act takes one real question and decomposes what answering it depends on into the knowledge beneath it - pure meaning, no people or storage yet, every card grey. ^dependencies The board has two card shapes. A **prompt** is a **card with a flag** (▶) - the root a cone starts from, a real request someone raises ("what is our refund policy for the EU?"). Everything beneath it is plain **[[Knowledge|knowledge]]** - a card that **depends on** other knowledge, an [[Edge|arrow]] running to the cards its meaning rests on. Traced down from the prompt, the arrows form a [[Directed acyclic graph|DAG]]; you stop a branch when its leaf is a single, nameable piece. A card's meaning *is* its [[Intention|intension]], decomposed into the cards it points to. A dependency already points ahead to the next act: an arrow into a **specific card** is a link you can follow, while an arrow into a whole **tool** means the answer is somewhere inside and must be *searched for* - worth only as much as that tool's findability. Once the cone is drawn you can **walk it**, in one of two modes - and the gap between them is the workshop's headline finding: - **human mode** - a person may bottom out at another person ("ask Sarah"); [[Ostension|ostension]] is allowed. - **agent mode** - an [[AI agent|agent]] cannot point or ask; every leaf must be written, reachable, and machine-readable. You leave with one prompt's [[Intention|intension]] fully decomposed - a shared map of what depends on what, settled before a word about who holds it. ## Containers *Where does each piece live, and in what state?* The second act drops every card into the place it is really kept, nests it under whoever is accountable, and colours the board - turning the map of meaning into a diagnosis. ^containers A **container** is a **frame**, placed on one **portability axis**: - a **person** - a head: [[Ostension|ostensive]] only (you can ask them, not read them), gone when they leave, never machine-readable. - a **tool** - a written store (a wiki, a repo, a drive), from locked and proprietary up to portable and machine-readable. Its place on that axis is its [[Survivability|survivability]] - and what lets an [[AI agent|agent]] read it at all. The spatial rule that replaces "written vs tacit": **a piece whose only container is a person is unwritten; a piece in a tool is written.** Containers **nest, and the outer owns the inner** - a tool sits inside the person or role accountable for it, and that nesting *is* ownership. A container nobody is above has no keeper, and unkept knowledge drifts. Now **colour every shape** - card, container, and [[Edge|edge]] - against one list. Default is **grey = not yet assessed**; pair each colour with its glyph, because a board gets read on a projector. - 🩶 **grey** - not yet assessed (the default) - 🟢 **green** ✓ - defined: written, one findable source; a kept tool; a named owner; a reachable edge - 🟡 **yellow** ? - ambiguous: written, but no single source; an unclear owner - 🟠 **orange** ~ - stale: rotting, reachable only with friction, or a diffuse / whole-team owner - 🔴 **red** ✗ - absent: nobody has it; or an edge that cannot be satisfied - 🟣 **purple** ◐ - tacit: in a head only ^colours A card's colour is its existence climbing 🔴 absent → 🟣 tacit → *written* (🟢, unless 🟡 ambiguous or 🟠 stale). A container's colour is its **holder clarity** - 🟢 a named individual, 🟡 unclear who, 🟠 a whole team - and an edge's colour is whether the dependency it carries is reachable. The act of colouring *is* the assessment: you leave with a fully coloured board, every [[Bus factor|bus factor]] visible and an agent-readiness verdict per prompt. ## Methods *What do we do about each gap, and who keeps it true?* The third act reads the move straight off the colour - every non-green shape names both the missing attribute and the refactor that restores it - then binds each fix to an owner. ^methods Nothing green needs touching; everything else is a shortfall sitting on a card, a container, or an edge: - 🟣 **tacit** card (its only container a person) → **capture** it into a tool ([[Capture pattern]]), then [[Distillation pattern|distil]] if bulky - closing the [[Bus factor|bus factor]], restoring [[Captureability]]. - 🔴 **absent** card, or an edge running into nothing → **author** it, or [[Decoupling pattern|decouple]] the dependency. - 🟡 **ambiguous** card → **[[Single source of truth|single-source]]** it and [[Linking pattern|link]] the scattered rest - restoring [[Discoverability]]. - 🟠 **stale** card, or a container with no keeper → **refresh** it and nest it under an accountable [[Stewardship pattern|steward]] - restoring [[Maintainability]]. - 🔴 / 🟠 **siloed edge** crossing into a container the asker cannot reach → **relocate** or [[Linking pattern|link]] it - restoring [[Interoperability]]. - a **fragmented** cone, hopping many containers before it assembles → **[[Centralization pattern|centralise]]**, **[[Federation pattern|federate]]**, or link. These are the [[Scale-free knowledge patterns|scale-free patterns]]; for the concrete method that restores each attribute - personal or collective - read it off the map: ![[Knowledge attribute to method map#^table]] Bind every move to the container accountable for keeping it green, pick the [[Minimum viable scope|scope]] worth greening first, and lift the decisions into a few implementation principles that survive any tool change. A move with no owner is an aspiration; an owner with no move is overhead. What the room signs at the end is its first [[Lightweight knowledge methodology|lightweight knowledge methodology]] - small, operational from Monday, expected to evolve - written up as a [[Decision record|decision record]].