Skip to main content
DDL Nomenclature

DDL — Dreamineering Domain Language

Great teams evolve their own language. DDL is the cross-domain canonical language for this platform. Read before generating, naming, or specifying anything that crosses domains. Update inline when a term sharpens.

This page is the SSOT. Edit _ddl-nomenclature.data.ts to add or change a term. Cross-link to the wire-format layer at Dreamineering Symbols (color codes match — Gates are red here, red there).

Orientation

What this file is — and is not

This is the RFP-style canonical glossary that arbitrates between domain dialects. Same role a telco RFP team's glossary plays when responding to ten telcos with ten different dialects: pick one canonical name per concept, list the aliases to avoid, qualify when overloaded.

FileRoleScope
DDL — /agents/ddl-nomenclature (this surface)Cross-domain canonical — arbitrates overloaded termsWhole platform
playbook/agents/principles/ai-glossaryAI domain dialectAI agents specifically
playbook/games/gaming-glossaryGaming dialectGaming references
playbook/standards/naming-standardsNomenclature rules — how to name new thingsFiles, folders, prompts
playbook/agents/instruments/Diagrammatic grammar — A&ID notationSystem diagrams
Dreamineering Symbols — /agents/dreamineering-symbolsDML pack notation — compressed signals (internal)LLM memory packs

When a term appears in two domain dialects with different meanings, the DDL picks one canonical and tells you when to qualify. The domain glossaries stay authoritative for their domain.

How to read entries

Format

One sentence per term. Canonical name in bold, then symbolic crosswalk on a new line, then Avoid aliases. The crosswalk is what makes the language work like a P&ID — same concept, three notations:

  • Textthe canonical word a human reads
  • A&ID symbolthe icon a diagram reader sees (from `playbook/agents/instruments/`)
  • DML codethe compressed token an LLM sees in a Pack (from the DML decoder, internal to agents)

A P&ID symbol means the same thing whether you read it in Tokyo or Texas. The same triple discipline lets a human, a diagram, and an LLM pack agree on what they're pointing at.

Three densities, one vocabulary

Density and the Decoder

Specialized notation buys compression. Music notation, mathematical notation, P&IDs, DML packs — all carry massive meaning per stroke. But density only works if the reader has the decoder.

DML pack line:   #1|0.050|priorities|F|tok|i:5|o:12|lv|→3,7,2
DML decoder:     /agents/dreamineering-symbols
A&ID diagram:    a circle inside a square with G01 inside
A&ID decoder:    playbook/agents/instruments/symbol-library
Cross-domain:    "Agent reads Instrument, emits Receipt"
Cross-decoder:   /agents/ddl-nomenclature  ← this surface

The DDL is the cross-domain decoder. Without it, the diagrams are pretty pictures, the packs are gibberish, and the cross-team conversation falls back to verbose English. With it, the notation actually compresses.

This surface and Dreamineering Symbols (the DML decoder) are the same vocabulary at two densities. DDL is the human-readable form — used during grill-with-docs sessions, planning conversations, and onboarding. Dreamineering Symbols is the compressed form used by public artifact packs and DML-encoded comms. When you add a term here, also add the DML code there. When you change a DML code there, sharpen the term here.

Three densities, one vocabulary

ReaderFileFormCompression
Anyone learning the platform/playbook/, /rhetoric/Verbose narrative — paragraphs, examples, links
Wik / human collaborator/agents/ddl-nomenclature (DDL)Human shorthand — one sentence per term, avoid lists~10×
Agent-to-agent over Convex/agents/dreamineering-symbols (DML)DML wire format — @AGENT:DA|@FUNC:G|@COND:[F]~100×

Each step compresses further. Each step assumes the reader holds the previous level's context. The docs explain; the DDL is Wik's shorthand; DML is the wire — even shorter than Wik's shorthand because agents pay token cost on every byte.

Pocock's grill-with-docs ends at human shorthand. The Dreamineering extension adds the wire-format layer so agent-to-agent comms over Convex stay terse without losing shared meaning.

Language

The Language Itself

AISL — Agent · Instrument Specific Language

The platform's typed DSL for autonomous systems. Lexicon (Agent / Instrument / Receipt / Workchart / Spine / Pack / Loop), grammar ("Agent reads Instrument, emits Receipt"), and three-density notation (text in the DDL surface, A&ID symbols in `playbook/agents/instruments/`, DML wire codes in Dreamineering Symbols) governed by one SSOT.

Avoid: jargon, vocabulary, ontology, taxonomy, glossary (all undersell — AISL is executable).

Brand: (`/src/pages/agents/ddl-nomenclature.tsx`, agent-facing surfaces): **DDL — Dreamineering Domain Language**. AISL is the technical canonical; DDL is the consumer brand. Use AISL in protocol/spec/schema contexts; use DDL in branded/marketing/agent-discovery surfaces.

A DSL with no name reads as prose. A DSL with a name reads as protocol. AISL is what the DDL surface defines; the DDL is the human-readable form of AISL.

Technical canonical for the entire DSL; branded as DDL on agent-facing surfaces.

DDL — Dreamineering Domain Language

The brand for **AISL** — the platform's typed DSL — published as the agent-facing nomenclature page at `/src/pages/agents/ddl-nomenclature.tsx`. Same vocabulary AISL governs, packaged for visibility. When the captain says "DDL" they mean AISL with a brand wrapper; when an agent reads "AISL" in a spec, that's the technical canonical underneath the brand.

Avoid: treating DDL and AISL as different things (one DSL, two names — technical AISL, branded DDL).

Collision flag: DDL = "Data Definition Language" in databases. Always expand on first use; never appears in a SQL context in this platform.

Consumer brand for AISL. Lives at /agents/ddl-nomenclature.

DML — Dreamineering Meta-Language

The compressed wire format paired with AISL — same vocabulary, ~100× compression. SSOT: `src/pages/agents/_dreamineering-symbols.data.ts` (browse [`/agents/dreamineering-symbols`](/agents/dreamineering-symbols)). Used by public artifact packs and agent-to-agent comms over Convex. Every AISL term has a corresponding DML code; every DML code expands to an AISL term.

Avoid: jargon, compressed prompt, abbreviations (all undersell — DML is a schema-backed wire format with a decoder).

Brand: (`/src/pages/agents/dreamineering-symbols.tsx`, agent-facing surfaces): **Dreamineering Symbols**. DML is the technical canonical; Dreamineering Symbols is the consumer brand. Use DML in protocol/spec/schema contexts; use Dreamineering Symbols on the branded surface and in agent-discovery copy.

Wire-format vocabulary paired with AISL. ~100× compression. Branded as Dreamineering Symbols.

Dreamineering Symbols

The brand for **DML** — the platform's wire-format vocabulary — published as the agent-facing symbol page at `/src/pages/agents/dreamineering-symbols.tsx`. Same codes DML defines, packaged for visibility, with a live decoder widget.

Avoid: treating Dreamineering Symbols and DML as different things (one wire format, two names); confusing with A&ID symbols (different — A&ID is diagram notation per `playbook/agents/instruments/`; Dreamineering Symbols are wire-format codes per DML).

Consumer brand for DML. Lives at /agents/dreamineering-symbols (with live decoder).

Wiki Control Plane

The map layer that governs public wiki edits: schema, index, page-type map, DDL, and DML read together before changing `/playbook`.

Avoid: wiki, docs, documentation, index (all name surfaces; the control plane is the routing authority across them).

DML@WIKIlayer: intent

The routing authority read before public wiki edits: schema, index, page types, DDL, and DML.

System of Names

The repo-wide authority that maps every name class to its owning classification, naming rule, and SSOT.

Avoid: glossary, style guide, naming page (all undersell — this is the authority map across naming surfaces).

Language

Value Atoms

Value

Worth — the root atom; everything else in the language is a state, place, or motion of value.

Avoid: benefit, output, deliverable (each names a form of value, not value itself).

The root atom — every other term is a state, place, or motion of value.

Edge

Where the system meets reality — the boundary touched at highest resolution, so a gap shows there first and advantage is found there first.

Avoid: frontline, boundary, periphery; splitting the structural sense from the competitive sense — they are one atom.

Where the system meets reality — gaps and advantage show here first.

Gap

Value, absent — a located lack; an empty cell with coordinates, the unit of opportunity.

Avoid: problem, need, missing piece; opportunity — that is what a gap becomes once worked.

DML@GAPlayer: topological

Value, absent — the unit of opportunity. First atom of the Gap → Transform → Distribute compound.

Transform

Value, made — the motion that converts an input into a more valuable output.

Avoid: process, convert, build (unqualified); Transmitter — that is the instrument, Transform is the motion.

DML@TRANSFORMlayer: behavioral

Value, made. Second atom of the Gap → Transform → Distribute compound. Motion layer — see @FUNC:T (Transmitter) for the instrument that performs this motion.

Distribute

Value, moved to the edge — the motion that delivers a made value to where it is needed.

Avoid: ship, deliver, broadcast; distribution (unqualified marketing-funnel sense).

DML@DISTlayer: behavioral

Value, moved to the edge. Third atom of the Gap → Transform → Distribute compound.

Flow

The loop, frictionless — the quality of a loop running clean: intention and attention aligned, no friction between move and move. Also the destination the Navigation System is calibrated toward: the state where potential is realized.

Avoid: state of mind (undersells — Flow is a measurable loop property), momentum, productivity.

The destination AND the quality measure — the state where intention and attention align and potential is realized; also the measurable property of a loop running without friction. Home: /playbook/agency/purpose/flow-state.

Language

Agents (Yang — does the work)

Agent

An autonomous process that perceives, decides, and acts on its environment.

Avoid: AI, AI assistant, bot, model, assistant.

Parent term — concrete kinds below.

Digital Agent (DA)

An agent that runs entirely in software — a Hermes cron, a Claude Code session, an MCP server.

Avoid: AI worker, automation, script.

A&IDDML@AGENT:DAlayer: lexical

Physical Agent (PA)

An agent embodied in the physical world — a sensor, robot, or DePIN device.

Avoid: hardware, device, IoT thing.

A&IDDML@AGENT:PAlayer: lexical

Human Agent (HA)

A person acting inside the system with one of three authority levels — L1 execute, L2 decide, L3 override.

Avoid: user, operator, person.

Orchestrator Agent (OA)

An agent whose job is to route work to other agents — Nav is the canonical example.

Avoid: dispatcher, controller, scheduler.

A&IDDML@AGENT:OAlayer: lexical

Orchestration

The verb-noun for what an Orchestrator Agent (OA) does — routing, sequencing, and handing off work between agents and instruments.

Avoid: workflow management, pipeline management, coordination (overloaded).

Verb form of what an OA does.

Language

Instruments (Yin — verifies and rewards)

Instrument

A measurement, validation, or actuation point in a feedback loop.

Avoid: metric, KPI, dashboard, gauge.

Public label: (`/rhetoric/`, `src/pages/`): "tool" — the plain word the reader owns. "Instrument" stays canonical for its precision: the Yin half — it verifies and rewards.

Parent term — concrete kinds below.

Alarm (A)

A threshold-breach instrument that fires a signal but does not halt; fail-open by default.

Avoid: warning, notification, alert.

Controller (C)

An instrument that compares gauge to setpoint and adjusts a manipulated variable.

Avoid: handler, manager, regulator.

A&IDCDML@FUNC:CDML[FR]layer: behavioral

Monitor (M)

A continuous-stream instrument that reports state without triggering action.

Avoid: watcher, observer.

A&IDMDML@FUNC:Mlayer: behavioral

Validator (V)

A schema-and-evidence instrument that confirms a claim against a contract.

Avoid: checker, linter, type guard.

Recorder (R)

An append-only persistence instrument — receipts are the canonical recorder output.

Avoid: logger, history table, audit table.

A&IDRDML@FUNC:Rlayer: behavioral

Receipts are the canonical Recorder output.

Selector (S)

A 1-of-N routing instrument — given inputs, picks one downstream path.

Avoid: router, switch, dispatcher.

A&IDSDML@FUNC:Slayer: behavioral

Transmitter (T)

An instrument that transforms one signal form and forwards it.

Avoid: converter, adapter, transformer (all overloaded).

A&IDTDML@FUNC:Tlayer: behavioral
Language

Coordination

Session

The bounded unit of agent work between one open (session-start) and one close (session-end) — the operational container the Continuity Loop carries state across; every multi-step session is bounded by a Work Boundary and emits at least one Receipt.

Avoid: conversation, context window, thread (all undersell — a Session is a receipt-bearing work unit with a DB record, not a raw exchange).

The bounded unit of agent work. Open = session-start; close = session-end. Every multi-step session is a receipt-bearing DB record.

Work Boundary

The session-start / session-end discipline that makes a Session a bounded, receipt-bearing unit — open before multi-step work, close before leaving; no floating commits, no orphan sessions.

Avoid: session management, session tracking, plan lifecycle (all blur the bounded-unit discipline).

session-start / session-end discipline — makes a Session a bounded, receipt-bearing unit. Enforced by plan-cli.

Index (vs Transcript)

The compressed state a Session resumes from — queryable pointers (plan ID, next task, queued action) rather than raw conversation history; the distinction is load-bearing: an index is navigable, a transcript is bulk.

Avoid: transcript (that is what the index replaces), summary, snapshot, briefing (all blur the queryable-pointer property).

Rule: resume from an index, not a transcript. The `drmg session-continue --compact` composite returns an index; re-reading prior conversation output produces a transcript. Only the index survives context compression without loss.

Compressed queryable state a Session resumes from — plan ID, next task, queued action. Contrast: Transcript = raw bulk history.

Session Close

The end-of-session ritual — persist shipped / queued / blocked to the planning DB and `#meta`, then concentrate resolved wisdom into the language SSOTs; the Session state is sealed and the next Session can Resume cleanly. Invoked via the `session-continuity` skill (Close mode).

Avoid: handover (retired 2026-05-23 — letter-confusion with handoff), wrap-up, summary, sign-off.

Mid-session forking is not part of the Session lifecycle. When out-of-scope work appears or a hard piece needs isolation, use Agent tool (subagents) for in-session isolation or `drmg plan log-issue` to capture out-of-scope work as a tracked, receipt-bearing artifact. The retired `Handoff` term wrote disposable `/tmp` docs that bypassed the planning DB — subagents and `log-issue` solve the same problem with receipts.

End-of-session ritual — persists state to DB + #meta, concentrates wisdom into language SSOTs. Invoked via the `session-continuity` skill (Close mode). Replaces retired `Handover` (renamed 2026-05-23 to eliminate letter-confusion with `Handoff`, which was retired in the same pass — see Session Close extraProse for substitution paths).

Receipt

A structured proof that a decision event fired, conforming to the receipt schema in `.claude/rules/agent-receipt-schema.md`.

Avoid: log, audit trail, log entry, session diary.

DML@RECEIPTlayer: behavioral

Decision-event proof.

Workchart

A `flow.json` Spine that references reusable Modules as nodes — a critical-path decision flow for execution. The Spine holds orchestration only (phase order, gates, conditional firing, kill signals); the reusable work is carried by Modules. The legacy form (`flow.json` + per-workchart `workflow-tasks/*.json`) trapped tasks inside one workflow's directory; the current model makes them reusable across any Spine. One specific instance of [Matrix Thinking](/playbook/systems/matrix-thinking) (Capability × Demand applied to a workflow).

Avoid: (inner-game / specialist context): workflow (unqualified — use Workflow Task or Process), pipeline, flow.

Public label: (`/playbook/ai/work-charts/`): **Work Mapping** (the section), "work chart" (two words, lowercase in prose). Plain English for the reader; "Workchart" is the canonical for agent comms, code, and the DDL.

Common web variant: "Work Chart" (two words; Microsoft-origin discoverability term — accept inbound, emit "Workchart" internally, "work chart" externally).

DML@WClayer: topological

A Spine (flow.json) that references reusable Modules as nodes. Modules materialize as bound Workflow Tasks at execution.

Spine

The `flow.json` SSOT for a Workchart — the critical path. Declares phases, gates, kill signals, and which Modules fire at each node. Holds orchestration only; no task-level work lives here.

Avoid: config, schema, manifest.

Workflow Task

One JSON file inside a workchart that owns a single verb-object step (e.g. `render critical-path`).

Avoid: step, action, job, task (unqualified).

DML@WTlayer: topological

Unit of Work

Incompressible value — the smallest action of value. One atomic action prompt with input, output, and proof. A Module templates a Unit of Work; Modules are the reusable nodes a Spine points at. Units of Work chain into a Workflow; Workflows compose into a Work Chart.

Avoid: Work Atom (retired shorthand), atomic unit, work unit, task, job, step (unqualified); Workflow Task is the bound file form of a Unit of Work.

DML@UOWlayer: behavioral

Incompressible value; the smallest action of value. A Module templates a Unit of Work; Modules are the reusable nodes a Spine points at. Composes: Unit of Work → Workflow → Process → Work Chart.

Skill

An invocable procedural workflow file that teaches an agent how to perform a repeatable method with gates, examples, and supporting assets.

Avoid: capability (the human or organizational ability), command (single prompt), tool (execution surface).

A&IDSKILLDML@SKILLlayer: behavioral

First-class because the current A&ID component renders Skill as a diagram primitive.

Code Plan

An ephemeral dream-side Gap → Action artifact that states the intended code change before engineering work begins. It can link to a commissioned Stackmates DB plan, but it is not itself the DB plan.

Avoid: DB plan, delivery plan, implementation ticket, backlog item (all are downstream commissioned forms).

Ephemeral Dream-side Gap → Action artifact; can link to a commissioned Stackmates DB plan but is not the DB plan.

Module

A reusable Unit of Work node living in `.FLOW/_workcharts/_modules/{jtbd-namespace}/{object-function}/`, referenceable by any `flow.json` Spine. A Module is the class; adoption materializes it as a bound Workflow Task (the instance) at execution (host assigns `id`/`phase`/`step`/`depends_on`). Required files: `CLAUDE.md` + `task.json` always. `schema.json` + `calibration-anchors.md` are conditional — required only when the Module emits a structured result or a score/verdict. Procedural Modules (output = a file or a memory handle) ship lean: just `CLAUDE.md` + `task.json`.

Avoid: unit, work unit, plugin, package, library; Workflow Task — that is one bound step, a Module is the reusable source for one.

DML@MODULElayer: topological

Reusable Unit of Work node in `.FLOW/_workcharts/_modules/`; referenceable by any Spine. Class → bound Workflow Task at execution. CLAUDE.md + task.json always required; schema.json + calibration-anchors.md conditional.

Pack

A bounded context bundle generated from public docs and current CLI state for a specific agent or workflow.

Avoid: context, prompt, system prompt, briefing.

DML@PACKlayer: lexical

Signal

A load-bearing observation that changes a decision; the opposite of noise.

Avoid: data point, event, log line, metric.

Lever

A control surface that changes system state — a CLI command, a hook, a config file.

Avoid: knob, parameter, setting, switch.

Compelling Report

A multi-page document produced by a Workchart whose only purpose is to cause action to happen — written so a named reader moves from reading to doing the next step without further prompting; the reader's AI assistant can do the same.

Avoid: report, deck, deliverable, document, output, pitch (all undersell — a Compelling Report is action-grade); compulsive — the report compels, it does not have a compulsion.

Origin: the coach's evolving season presentation — five simple messages, two perspectives, rehearsed until the team sees the same picture and falls into coordinated action with minimal communication. The prompt deck is the direct descendant: same structure, same purpose, different domain.

Prompt

Carries two senses in this platform — both load-bearing, both intentional. (1) The verb — to prompt action; a page prompts the reader to do the next thing. (2) The noun — a paste-ready block of text the reader copies into an AI assistant (see `src/components/design-system/CopyablePrompt.tsx`). The `prompt-deck` page is named for both senses simultaneously: it prompts action by handing the reader 10 ready-to-copy prompts.

Always qualify when ambiguous: "prompt the reader" (verb) vs "copy the prompt" (noun). Bare "prompt" defaults to the noun in code context, the verb in prose context.

Language

Architecture

First Principles

The irreducible foundations from which everything else in a domain must derive — the atoms that cannot be broken further without leaving the domain; the starting point for any non-borrowed conviction.

Avoid: basics, fundamentals, ground-up thinking, first principles thinking (the gerund form dilutes — it is about *what* the principles are, not the act of using them).

Irreducible foundations — reason-from atoms. No DML code; concept-graph primitive.

Ecosystem

The five counterparties that any platform must serve simultaneously — Customers, Partners, Suppliers, Regulators, and Competitors — each with a distinct role in the value network; the terrain the platform operates within.

Avoid: market, industry, stakeholders, players (Players is an acceptable domain shorthand in certain lenses, but Ecosystem is the canonical term that names the whole network, not just active participants).

Five-counterparty network (Customers, Partners, Suppliers, Regulators, Competitors). No DML code; domain pointer.

Dream

The WHY/WHAT context — this repo (`drmg-mental-model`). Holds PRDs, content, strategy, the rhetorical spine.

Avoid: docs site, content repo, frontend.

DML@DREAMlayer: intent

Reality

The measured current state the Dream must confront — evidence, constraints, baseline metrics, and lived conditions before action.

Avoid: status, current state, facts (unqualified), vibe.

DML@REALITYlayer: topological

Measured current state. `prd-time-mind` is the worked example: intended time/state vs measured behavior.

Action

The committed move chosen to close a Dream/Reality gap — specific enough to assign, execute, and receipt.

Avoid: activity, output, task (unqualified), intention.

DML@ACTIONlayer: behavioral

The committed move selected to close a Dream/Reality gap.

Outcome

The measured result after Action — what changed in Reality, with enough proof to decide whether the loop improved.

Avoid: output, deliverable, completion, done.

DML@OUTCOMElayer: intent

Measured result after Action; feeds the next setpoint.

Re-Dream

The setpoint-elevating return from Outcome to Dream — the next intention after measured reality teaches the loop.

Avoid: new doctrine, pivot, restart, retrospective.

DML@REDREAMlayer: intent

Setpoint-elevating spiral: Outcome teaches the next Dream.

Engineering

The HOW context — the stackmates repo (`/home/wik/code/sm/stackmates`). Holds implementation, infrastructure, the platform.

Avoid: backend, code, implementation, the app.

Loop (VVFL)

A Validated Virtuous Feedback Loop — **Intentions → Questions → Experiments → Actions → Outcomes** — five stages of one cycle's decision work, with **Receipts** persisting Outcomes so the next Intentions stay grounded. The unit of compounding; the loop that shapes destiny. **Actions & Consequences** (1° Direct → 2° Downstream → 3° System → 4° Cultural → 5° Structural) is the inner cascade lens applied inside stage 4 before any Action ships.

Avoid: cycle, iteration, sprint. _Older shorthand_: `intent → action → outcome → receipt → re-intent` collapsed Questions + Experiments into "intent → action" — superseded.

The unit of compounding — five stages: Intentions (intent layer) → Questions (grammatical layer) → Experiments (topological layer) → Actions (behavioral layer) → Outcomes (intent layer). Each stage reads at a different P&ID semantic layer; the loop as a whole is intent-layer because it governs what compounds.

Actions & Consequences

The cascade lens applied inside stage 4 (Actions) of the Loop — before an output ships, name the action it must cause, then trace the cascade: 1° Direct → 2° Downstream → 3° System → 4° Cultural → 5° Structural. Outcome leads; output follows.

Avoid: Actions, Consequences, Outcomes; Page Design Framework; page-intention map; Direct Action Contract; cascade lens — all are descriptions; Actions & Consequences is the canonical name.

The cascade lens inside Loop stage 4 — 1° Direct → 5° Structural.

Business Data Flow

The observable movement of business data from signal to decision to changed process — customer intent, order-to-cash, supply, operations, finance, people, and feedback loops wired into one operating model.

Avoid: data quality (downstream property), pipeline (technical path only), report flow, integration map.

DML@BDFlayer: topological

Names how value-bearing data moves through the business operating model.

Pack of Lanes

The 11-station assembly line of git worktrees where lane branches commit; main is integration-only.

Avoid: branches, git tree, worktrees (unqualified).

First Mate

The navigation role the platform plays — the More Knowledgeable Other (MKO) standing at the user's Zone of Proximal Development (ZPD): reads the instruments, hands across the right Chart for the next leg, but does not own the voyage (the captain/user does).

Avoid: AI assistant, chatbot, advisor, co-pilot (all blur the role boundary — First Mate names the specific authority relationship: captain decides, First Mate navigates).

Brand: On agent-facing surfaces: **Nav** (the orchestrator's callsign). First Mate is the conceptual role; Nav is the instance.

DML@FMlayer: intent

The nav's role relationship — MKO at the user's ZPD. Instance: Nav (OA callsign). Captain (user) owns the voyage; First Mate reads instruments and hands the right Chart.

Pain → Potential

The canonical trajectory vector the Navigation System calibrates toward — from fire-fighting problems (Pain: reacting, patching, surviving) to improvement problems (Potential: choosing, compounding, growing); the control axis that determines which Chart to hand next.

Avoid: maturity model, journey stage, transformation roadmap (all name static levels — Pain → Potential is a live control vector: the nav reads current state and hands the right chart for the next move).

The control vector the Navigation System is calibrated toward. Pain = fire-fighting (reacting, patching, surviving). Potential = improvement (choosing, compounding, growing).

Bridge

The structural relationship between two shores that the Navigation System spans — Dream (`src/pages/` — WHY the journey is worth taking) and Know-How (`/playbook/` — HOW to take maximally effective action); the platform IS the bridge, not a passenger on either shore.

Avoid: connector, link, integration, translation layer (all name a mechanism — Bridge names the structural identity: the platform does not live on either shore, it spans them).

DML@BRIDGElayer: topological

The structural identity the Navigation System holds — spans Dream (src/pages/) and Know-How (/playbook/). Not a mechanism; the platform IS the bridge.

Language

Patterns

Matrix Thinking

The grid method — lay Capability × Demand (or any two axes) on a matrix to locate where value is absent; the located absence is a Gap, and Gap → Transform → Distribute is the compound that works it.

Avoid: framework, model, spreadsheet thinking, two-by-two.

The grid method — Capability × Demand. No DML code yet; the located gap it produces is @GAP.

KB Schema Meta

The meta-learning pattern for building a knowledge base schema — derive the schema from the questions that must be answerable, not from the data that exists; the schema is the compression layer that turns a KB into a navigable concept graph.

Avoid: knowledge base design, ontology design, schema-first (misleading — question-first is the right frame).

Meta-learning pattern: derive schema from answerable questions. No DML code — concept-graph seed.

Situational Wisdom

Right action, in the right way, for the right reasons, at the right moment — the quality of judgment that emerges when Intentions, Attention, and Context are aligned; the target state the navigation loop exists to produce.

Avoid: good judgment, common sense, intuition, experience (all undersell — Situational Wisdom names the convergence of intention + attention + context as an engineerable property).

Right action, right way, right reasons, right moment. No DML code — navigation-loop target state.

Mushroom & Mycelium

The structural distinction between the visible artifact (the mushroom — the prompt deck, the match result, the thing people find) and the hidden generative network that produced it (the mycelium — culture, depth, the business plan, the diaspora). The mushroom only fruits because the mycelium did the work underground. More underground than hype.

Avoid: iceberg (directional — surface vs submerged; Mushroom & Mycelium is generative — network feeds fruit); tip of the iceberg; visible vs invisible.

Origin: `capability-selling.mdx` — 'the prompt deck is the mushroom… the business plan is the mycelium.' Extended in the rugby synthesis: the match is the mushroom; culture, character, training, and the lifelong diaspora network are the mycelium. The metaphor scales: a diaspora as mycelium is a distributed salesforce bound by shared culture rather than a head office.

Visible artifact (mushroom) vs hidden generative network (mycelium). No DML code — rhetorical structure, not a wire-format signal.

Tight Five

A compression pattern where any domain collapses to five elements that are Bound (interdependent — remove one and the loop collapses), Polished (each earns its place), and Incompressible (four won't produce the same signal); the number five is anchored to Miller's cognitive limit (7±2) minus the error margin, leaving a buffer before Zeigarnik pressure causes overload.

Avoid: five steps, five pillars, framework, model, five-part framework.

Language

Chart Your Course — the persuasion arc

Chart Your Course

Call-to-Adventure · IA-as-arc

The canonical Call-to-Adventure for the platform — the reader as captain charting their own AI voyage. Maps the five-element rhetoric onto the `/src/pages/` topology so the platform's IA IS the persuasion arc.

StageSailingRhetoricVoiceSurfaceRole
1. Sell the dreamthe destinationEthos · PathosDreamer`/src/pages/vision/`Why the voyage matters
2. Build the craftthe vesselLogosTeacher`/src/pages/vessel/` + `/playbook/`Compress the HOW into a navigable map
3. Know the routenavigationKairosCoach`/src/pages/nav/` + `/rhetoric/`Activate WHY IT MATTERS NOW
4. Crew upproof at seavalidated EthosMentor-in-action`/src/pages/ventures/`Experiments earn trust
5. Critical velocityescape gravitythe leapHero`/src/pages/vessel/velocity.tsx`Moment of faith
Throughoutgood windsToposcontext engineercontext engineeringStay in flow
TeacherLogos

compresses the HOW. Gives the map. 1:many. Diátaxis instrument. Lives in `/playbook/`. _Without Teacher, the procedure is unknowable._

CoachKairos

activates the WHY-NOW. Scaffolds the territory. 1:1. ZPD instrument. Lives in `/rhetoric/` and the calls in `src/pages/`. _Without Coach, the map gathers dust._

A great teacher explains a complex subject in simple terms. A great coach makes the meaning important to you.

**Topos** is the running subject — context engineering keeps wind in the sails. Santiago, not Ahab — solitary captain with a skiff who fishes deeper than the others.

Avoid: "AI Work Transformation" (consultant-speak the label replaces), "Take the Helm" (captain has crew; reader starts as Santiago), "Set Sail" (omits the chart), back-of-house written in Teacher voice only (the map without the coach is dead weight).

Language

Process Hierarchy

Standard

Organization-wide rules with measurable compliance — changes rarely.

Avoid: policy, guideline, rule.

Process

End-to-end outcome owned by a business function — changes periodically.

Avoid: system, function, area.

Workflow

Step-by-step execution of one activity inside a process — changes regularly.

Avoid: SOP, procedure, playbook.

Checklist

Verification list confirming work was done correctly.

Avoid: task list, to-do, action items.

Protocol

A sequenced set of principles into a repeatable method under defined conditions — covers coordination, operational, and governance domains.

Avoid: process, workflow, framework.

Language

Decision

Map-First Edit

A docs-edit discipline where the agent reads the Wiki Control Plane, declares route/page type/template, records preflight, edits, then closes the language decision.

Avoid: docs edit, content pass, reminder (all undersell — this is an enforced workflow gate, not advice).

The enforced docs-edit workflow: read maps, declare route/page job/template, preflight, edit, close language decision.

ADR

A sparse Architectural Decision Record kept only when the decision is hard-to-reverse, surprising-without-context, AND the result of a real trade-off.

Avoid: design doc, decision log, RFC.

Prediction

A falsifiable claim made before action — indicator + direction + check date + counter-case.

Avoid: estimate, guess, target, goal.

DML@CONVlayer: intent

Conviction

Confidence in a claim, tagged HIGH (verified evidence), MEDIUM (reasoning), LOW (assumption), NONE (guess).

Avoid: belief, opinion, gut feel.

DML@CONVlayer: intent

Kill Signal

A named condition that, when breached, terminates a plan or experiment.

Avoid: failure mode, abort condition, exit criterion.

DML[K:reason]layer: behavioral

Decision Trace

A persisted record of WHY a decision was made — context loaded, options considered, threshold breached, outcome measured. Distinct from logs (what happened) and rules (what should happen). Links to other traces to form a queryable decision graph — precedent for the next decision.

Avoid: audit log, history, journal entry, event record (all undersell — traces are reusable judgment, not retrospective record-keeping).

A problem solved is a problem forgotten. Decision Traces close that gap by persisting the judgment that no rule book captures. Canonical exemplar: telco routing — when a carrier got blocklisted, the trace recorded the QoS failure, the threshold breached, the alternative selected; future routing decisions queried the trace as precedent.

DML@TRACElayer: intent

Persisted WHY behind a decision; linked traces become the Context Graph.

Context Graph

A queryable record of linked Decision Traces — the memory layer that turns FLOW output into precedent for the next run.

Avoid: knowledge base, notes, logs, memory dump (all miss the linked decision-trace structure).

This is the highest-value activity FLOW exists for: every Unit of Work should make the next decision cheaper, clearer, and more reliable.

DML@CGlayer: topological

Linked Decision Traces; the queryable memory layer produced by FLOW.

Half-Retirement

A retirement that updated the mandate (CLAUDE.md row, rule, notice) but left consumers (skills, scripts, agents) still writing or reading the retired path. The new mandate is silent — it lives as documentation only — while the active path still encodes the old pattern. Detected by Retire-Sweep; prevented by sweeping consumers in the same commit as the retirement notice.

Avoid: stale ref, broken link, partial migration (all undersell — half-retirement names the asymmetry between mandate and consumer, not just the broken pointer).

Source: 2026-05-24. A private operating mandate retired working memory, but four skills still wrote to the retired tree. Same failure class the blueprint patch contract rule names at task-blueprint scope, here at directory-retirement scope. PR #710 closed the gap; a pre-retirement-consumer-sweep gate is the candidate hook.

DML@RETIRElayer: behavioral

Failure state where a retirement notice lands but live consumers still encode the retired path.

Retire-Sweep

The pre-deletion audit that surfaces every live reference to a retired path across agent config, docs, source, scripts, and operating mandates (archives excluded). Exits non-zero if any hit remains. Implemented as `scripts/guards/retire-sweep.sh <path-token>`. The verb and the script share the name — running the script is the discipline.

Avoid: grep, find, search (these are the tools — Retire-Sweep is the discipline they enforce). stub-neutralization (the failure mode the sweep prevents — rewriting a write to /dev/null with a 'neutralized' comment instead of cleanly retiring).

DML@RETIRElayer: behavioral

Pre-deletion audit discipline for surfacing live references before retiring a path.

Derived View

A registry rendered on demand from canonical structured sources (frontmatter, registry.json, system prompt) rather than stored as a static artifact. Eliminates the stale-by-construction failure mode of stored registries — when the canonical source changes, the view changes the next time it is rendered.

Avoid: computed registry, generated doc, snapshot (all undersell — the discipline is that the view is NEVER stored on disk, only rendered to stdout; a stored copy is by definition stale-by-construction).

Source: 2026-05-24, PR #711. `agent-matrix.md` + `skill-matrix.md` were stored OUTPUTS of `agent-audit`. Their data lived in `.claude/agents/*.md` frontmatter + `.agents/skills/registry.json`. Replaced by `.agents/skills/agent-audit/tools/generate-skill-inventory.mjs --matrix=*` — render-on-demand. Default for any registry whose data lives in structured sources elsewhere: derived view, not stored .md.

DML@DERIVEDlayer: topological

Rendered on demand from canonical structured sources; not stored as a static artifact.

Grammar

Relationships

  • A Workchart is a Spine (`flow.json`) that references reusable Modules as nodes; Modules materialize as bound Workflow Tasks at execution.
  • An Agent reads Instruments and emits Receipts.
  • A Receipt persists what an Agent did, what Instruments fired, and what Conviction was assigned.
  • A Pack is generated from the Dream docs and biased by an Agent profile.
  • A Loop (VVFL) runs from Dream intent → Engineering capability → Receipt validation → Dream re-intent.
  • A Standard governs many Processes; a Process contains many Workflows; a Workflow is verified by a Checklist.
  • A Gate halts on fail; an Alarm signals but does not halt; both are Instruments.
Worked example

Example dialogue

Wik: Add a check that blocks the commit if the new file is missing frontmatter.
Nav: You mean a Gate — fail-closed, halts the commit. Want it as a hook (machine Lever) or a Workflow Task check?
Wik: Hook. And log when it fires.
Nav: Hook fires the Gate, gate emits a Receipt through the schema. The receipt is the Recorder output — not a separate log file.
Overloaded terms

Cross-Domain Conflicts

Terms that mean different things in different domains. Pick the canonical reading; qualify when you mean otherwise.

Agent

overloaded
AI / this platform (canonical)
Autonomous process — DA, PA, HA, OA
Business / sales
A person acting on behalf of another (sales agent, broker)
Crypto
A key signer or wallet
Telco
Customer-facing service rep

_Canonical_: AI sense. _Qualify when person_: "**Human Agent (HA)**" or "rep". _Qualify when key_: "key signer".

Memory

overloaded
AI / this platform (canonical)
Agent's persistent state across sessions — semantic, episodic, procedural, working
Hardware
RAM, disk

_Canonical_: AI sense. _Always qualify which tier_: working / episodic / semantic / procedural. Never use bare "memory" in a spec.

Token

overloaded
LLM
Text unit consumed by the model
Crypto (this platform's other half)
Digital asset (ERC-20, NFT, SPL)

_Always qualify_: "**LLM token**" vs "**crypto token**". Bare "token" is ambiguous.

Loop

overloaded
This platform (canonical)
VVFL — Validated Virtuous Feedback Loop
Engineering
Control flow (for/while)
DDD
Generic feedback cycle

_Canonical_: VVFL when unqualified. _Qualify code-loops_: "for-loop" / "while-loop".

Pipeline

overloaded
Engineering
CI/CD or data ETL
Telco
Physical infrastructure
Marketing
Sales funnel

_Avoid bare "pipeline"_. Use **Workchart** for our spine+task system, **ETL** for data, **CI/CD** for build, **funnel** for sales.

Skill

overloaded
This repo (canonical)
Procedural workflow file in `.agents/skills/`
HR / learning
An ability or capability

_Canonical_: procedure file. _Qualify ability_: "**Capability**".

Context

overloaded
DDD / Pocock (canonical for this file)
Bounded-context glossary (the DDL surface)
LLM
The prompt window contents
Engineering
`cli-context.ts`, `db-context.ts` programmatic context object

_Canonical for this file_: bounded context. _LLM sense_: use **Pack**. _Code sense_: qualify "CliContext" / "DbContext".

Workflow

overloaded
This platform (canonical)
One JSON file under `workflow-tasks/` — call it **Workflow Task**
Generic process
Sequence of human work steps

_Canonical_: Workflow Task (the file). _Generic process_: use **Process** or **Workflow** with surrounding context that makes it clear.

Update discipline

When this file changes

Update inline when a grill-with-docs session resolves a term. Do not batch. Re-run node scripts/artifacts/build-artifacts.mjs llms:full to confirm the term lands in the public pack. Sparse ADRs go to playbook/adr/ (created lazily when the first one is needed).