Dreamineering Meta-Language

This is our internal language — the standard we think, plan, and coordinate in. Not a crawler schema. External agents that want to operate as peers learn it. llms.txt is the billboard that points here; this page is the territory.

How does any LLM communicate most effectively with another?

Process engineers solved this problem decades ago. ISA 5.1 gives every plant the same grammar — FT-101 means Flow Transmitter, loop 101, anywhere in the world. The legend sheet teaches the reader. The reader teaches the next reader. The grammar compounds.

This page is the same idea, applied one layer up: from physical process flow to agent intention flow. Every symbol carries its full instruction in the type. The decoder maps symbol → inherited behaviour. No prose required once you have the key.

The legend sheet is the product. A P&ID without ISA 5.1 is line noise — FT-101 means nothing until the reader has loaded the key. The drawing only becomes dense once the decoder is in context. Same contract here: VFL-G01 [FC] [Q:>=3] is three tokens of garbage to a fresh reader, and a full behaviour contract to a reader who holds this page in memory.

Without the key	With the key
Symbols cost more to explain than prose would	Seven tokens replace a paragraph. Compression ratio ≈ 40:1
Reader must ask clarifying questions before acting	Reader acts
The diagram is an obstacle	The diagram is a contract

Load this page once. Every A&ID diagram, every pack line, every WM-NAV entry decodes against the same ~30 primitives: 7 families, 8 function codes, 3 authority levels, 5 loop types, 3 fail-safe modes, plus condition grammar. That count is the key size — small enough for any LLM to hold in working context, large enough to compose every intention flow the platform needs to express.

Two disciplines fall out of that:

1. Never use a symbol in the wild that isn't in the published key. If a diagram uses a family the decoder has never seen, the author owes the update to the key — not a footnote on the drawing.
2. Every symbol is paid for by outcomes. A symbol earns its place when using it produces better decisions than the prose it replaced. The gauge is WM-NAV.pack — does the compressed form lead to sharper action? Keep. Does it obscure? Retire.

Cold-read test. Reach the bottom of this page, then decode this line without help:

VFL-G01 [FC] [Q:>=3] HA[L3] →PRD-V01[P] →COMMS-A01[F] ⟳

If every token resolves and you know what to do, the standard is working.

The White Space

Workflow notation already has standards. Message schemas already have standards. Nothing standardised carries agent intention with typed instruments and verifiable feedback loops. That gap is what this notation fills.

Layer	Process engineering	Finance / commerce	Agent intention flow
Grammar	ISA 5.1	BPMN 2.0	A&ID type system (this page)
Local dictionary	Project legend sheet	Firm-specific schemas	Pack notation
Message schema	—	ISO 20022	IntentTrace
Lifecycle / executable	—	ISDA CDM, FpML	Operating protocol
Tag pattern	FT-101	—	VFL-G01

The substrate is different — pipes versus pixels versus protocols — but the grammar discipline is identical. Symbols carry types. Types carry behaviour. Diagrams compose without prose.

Three Layers

Layer	Name	Role
1	Key / Type System	This page. Decodes any symbol in any A&ID.
2	Operating Protocol	How agents using the key behave toward each other.
3	Standard / Story	Why the approach is correct. Distributed for external consumption.

Learning budget rule. Column 1 of every table below is something LLMs already know — type systems, ontologies, control theory. Column 2 is the local encoding. Spend learning budget only on genuinely new primitives. The rest you already have.

Organising Principle

Key	Value	Context
deterministic	Same input → same output	Trust layer: standards, protocols, settlement
probabilistic	Same input → distribution of outputs	Intelligence layer: prediction, exploration, agents
pikorua	Dual-strand spiral	Deterministic and probabilistic crossing upward
agency	Probabilistic actors on deterministic rails	The balance that creates flow

P&ID maps physical process flow. A&ID maps intention flow. Same grammar. Different substrate.

A&ID Type System

Each symbol is a typed instance. The type carries the full behaviour contract — the diagram inherits it. The author writes only the instance.

Symbol format: {FAMILY}-{FUNC}{NN} — for example, VFL-G01 decodes as Family VFL (Validated Feedback Loop), Function G (Gate), Instance 01.

Function Codes

Code	Name	Behaviour	Default Fail-Safe	Override
A	Alarm	Signals threshold breach. One-way. Non-blocking.	[FO] open	HA[L2] ack
C	Controller	Compares gauge to setpoint. Adjusts manipulated variable.	[FR] reset	HA[L3] set
G	Gate	Binary pass/fail. Halts on failure. Unidirectional.	[FC] closed	HA[L3]
V	Validator	Checks schema or contract compliance. Returns pass/fail and evidence.	[FC] closed	HA[L2]
M	Monitor	Continuous observation. Non-blocking. Outputs stream.	[FO] open	HA[L1] ack
R	Recorder	Persists to durable store. Append-only.	[FC] closed	HA[L3]
S	Selector	Routes to one of N outputs by condition.	[FC] closed	HA[L2]
T	Transmitter	Transforms and forwards signal. Cross-boundary.	[FO] open	HA[L1]

Family Codes

Code	Domain
VFL	Validated Feedback Loop — quality, standards
PRD	Product Requirements — feature commissioning
ETL	Extract-Transform-Load — data, intelligence
NAV	Navigation — routing, context, sessions
COMMS	Communications — channel routing, message types
AGNT	Agent — identity, authority, capability
AUTH	Authority — permissions, overrides, sign-off

Agent Shapes

Shape	Code	Meaning
○	DA	Digital Agent — fully autonomous
○—	HITL	Human-in-the-loop — requires sign-off before proceeding
□	OA	Orchestrator — routes and coordinates
◇	PA	Physical Agent — sensors, actuators
⬡	HA	Human Authority — overrides, sets setpoints

Authority Levels

Level	Scope
L1	Execute — runs within defined parameters
L2	Decide — selects from pre-approved options
L3	Override — can break gates, change setpoints

Hierarchy: HA[L3] > OA[L2] > DA[L1]

Condition Grammar

Syntax	Meaning
[P]	On pass
[F]	On fail
[T:n]	After n time units
[E:x]	On event x
[Q:>=n]	When quality score meets or exceeds n
[A:role]	When the named authority role approves
⊗	Override — human breaks the flow

Fail-Safe Defaults

Mode	Symbol	Default for	Behaviour
Fail closed	[FC]	G, V, R, S	Block all flow until manually cleared
Fail open	[FO]	A, M, T	Continue with last known state
Fail reset	[FR]	C	Return to defined default

The fail-safe is inherited from the function code. The author only writes the override when it differs from the default.

Loop Types

Symbol	Name	Behaviour
↺	Closed	Feedback corrects toward setpoint
↺↺	Cascade	Output of one loop feeds the setpoint of the next
→	Open	No feedback — fire and forget
⟳	Learning	Setpoint updates from observed outcomes
⇉	Feedforward	Anticipates disturbance before it arrives

Binding Dimensions

Five questions every page or node should connect. Pages connecting all five are fully bound. Pages connecting one or none are islands.

Key	Question	Tags
purpose	Why does this matter?	purpose, why, north-star, mission
principles	What truths guide you?	principles, first-principles, truth
platform	What do you control?	platform, tools, software, build
perspective	What do you see others don't?	perspective, edge, vision, prediction
performance	How do you know it's working?	scoreboard, measure, performance

Score	Meaning
5/5	Fully bound — all five dimensions connected
2-4	Partial — connected but with gaps
0-1	Island — exists but not wired

Reading a Full Symbol

VFL-G01 [FC] [Q:>=3] HA[L3] →PRD-V01[P] →COMMS-A01[F] ⟳

Token	Decode
VFL-G01	Validated Feedback Loop family, Gate function, instance 01
[FC]	Fail-closed — blocked by default, must pass to proceed
[Q:>=3]	Condition — proceeds only when quality score is at least 3
HA[L3]	Human Authority at Level 3 can override
→PRD-V01[P]	On pass, routes to PRD Validator instance 01
→COMMS-A01[F]	On fail, fires Comms Alarm instance 01
⟳	Learning loop — setpoint updates from observed outcomes

One line. Full behaviour contract. No prose required.

The Ontology Insight

The symbol is the type. VFL-G01 inherits everything Gate means — binary, halting, fail-closed, overrideable at Level 3 — without restating any of it. The key decodes the type once. The diagram author writes only the instance. Every reader who has the key inherits the same contract.

This is how an A&ID carries orders of magnitude more information density than the prose description it replaces. A page full of "the gate blocks unless validated" sentences becomes one symbol, and the symbol composes into larger flows without ambiguity.

P&ID gave physical engineering this density a century ago. The notation made plants buildable by teams who never met. A&ID extends the same discipline to systems where the substance flowing is intention, not fluid.

What the Symbols Carry

A legend sheet is only useful if the lines on the drawing carry something real. In a P&ID the lines carry fluid, signal, or pneumatic pressure. In an A&ID the lines carry four things at once — and every →, ↺, ⟳ in this decoder resolves to one of them.

What flows on the line	Encoded by	Where it lives
Instructions	Instance + condition + route	Intercognitive Protocol — how physical agents receive and act
Money / incentive	AUTH, ST, IT, GT families	Economy of Things — machines with identity, wallet, marketplace
Energy / work done	Agent shapes + instrument types	DePIN Devices — the physical agents that convert intent into reality
Verification	V, G, ⟳, HA[L3] override	Verifiable Intent — cryptographic proof that action matched intention

The loop closes when all four agree. An instruction (→PRD-V01) dispatches with an incentive attached (IT-G01 gate), drives a physical or digital agent (PA, DA) to produce work, and terminates at a validator (V) or gate (G) that checks outcome against the original intent. If the check passes, ⟳ updates the setpoint — the loop learned. If it fails, →COMMS-A01[F] fires and human authority (HA[L3]) can override.

That is the Validated Feedback Loop expressed in its minimum form. Every symbol in the decoder above is paid for by its role in this closure. Remove any one and the loop breaks: no instructions means no action, no incentive means no movement, no work means no outcome, no verification means no learning.

This is why the standard belongs at the /docs/standards/ layer — it is the single common grammar for the four protocols that make agent commerce possible. Load the key once; decode any flow where intent, money, work, and proof move together.

Communications Standards

The type system above defines what symbols mean. This section defines how agents must use them when speaking to each other. The three subsections map onto the three layers: symbols, messages, dialogue.

Type and Symbol Rules

Rules that govern any symbol appearing in a diagram, pack line, or message payload.

Rule	Requirement
Decodable	Every symbol must decode against this page alone. If a fresh LLM cannot resolve it from the key, the symbol is not allowed.
Typed	Every instance uses the form {FAMILY}-{FUNC}{NN}. No bare function letters. No prose substitutes.
Inherited default	Fail-safe, override authority, and loop semantics are inherited from the function code. Authors write only the deviations.
No collision	A new letter, bracket, or marker must not overload an existing one. Collisions block promotion (see v1.1 queue).
Decoder-first	On receiving a symbol, an agent validates it against the type tables (@FUNC, @COND, @SHAPE, @TOK) before taking any action.

Message Schema and Authority

Every message between agents using this standard must declare family, authority, intent, and conditions. The canonical shape:

comms.message:
  family: COMMS | NAV | AGNT | AUTH | VFL | PRD | ETL
  authority: L1 | L2 | L3
  intent: query | proposal | decision | alert
  conditions:
    - "[P]"
    - "[F]"
    - "[T:5]"
    - "[Q:>=3]"
  payload:
    symbols:
      - "VFL-G01"
      - "PRD-V01"
      - "@M:slug"
    text: "short natural-language explanation, optional"
  trace:
    source: page-slug or repo path
    edge_weight: 1.0 | 2.0 | 3.0 | 5.0

Field	Rule
family	Required. Tells the receiver which hat the sender is wearing.
authority	Required. Declares what the sender is allowed to do and what response is expected.
intent	Exactly one primary intent per message. Split multi-intent payloads into multiple messages.
conditions	Shared condition grammar only ([P], [F], [T:n], [E:x], [Q:>=n], [A:role]). No ad-hoc predicates.
payload.symbols	Must decode against the type system. Unknown symbols route to the candidate queue, not to action.
trace	Every factual claim in the payload must cite a source. Edge weight reuses the link-weight scale.

Dialogue and Escalation Protocol

The rules that govern multi-turn interaction between agents once the grammar is shared.

Rule	Requirement
Handshake	A new agent reads this page and the operating protocol (Layer 2) before sending anything. Initial messages carry L1 only.
Authority grant	Higher authority (L2, L3) is granted through AUTH family messages. An agent never self-elevates.
Escalation	If [Q:>=n] or [A:role] cannot be satisfied, the agent halts and emits an AGNT family alert. Guessing is forbidden.
Fail-closed	If a received message cannot be fully decoded, the receiver flags [F] and routes to HA[L2] for review. Silent drops forbidden.
One intent	One primary intent per message. Chain multi-step work through ↺↺ cascade, not through overloaded payloads.
Trace or reject	Any payload claim without a trace.source is rejected on receipt. Trust without provenance is not a valid state.

These rules exist so any LLM that loads this page and the operating protocol can enter the system as a peer — not a guest — and compress its reasoning into the shared pack without translation loss.

Versioning

Standard v1.0 — frozen. Breaking changes increment to v2. New families and instruments may be added in v1.x minor versions only when the existing grammar passes the cold-read test on real diagrams across independent readers.

Promotion Discipline

A symbol earns its place in a minor version when all four gates pass:

Gate	Criterion
Recurrence	Used across at least three independent contexts, not just the one that proposed it
Cold-read test	Fresh reader decodes the symbol from the updated key without follow-up questions
No collision	Does not overload an existing letter, bracket, or condition marker
Outcome improvement	Using it produces sharper decisions in WM-NAV.pack than the prose or workaround it replaces

Until all four pass, the candidate stays in the queue below. A symbol that fails any gate is either reshaped or retired — never silently accepted.

v1.1 Candidate Queue

Symbols surfaced through use but not yet promoted. Listed for transparency so readers know what is under test and what is law.

Candidate	Source	Status	Next gate
@CTX @ASK @PLAN @RISKS @TRIG-DEF	A2A handshake annotations proposed during cold-read tests	1 context — needs 2 more	Recurrence
NAV-F* (Flow function)	Proposed as navigation flow primitive	Collision with [F] fail condition marker	Rename required before recurrence counts
MECH family (mechanism)	AMM, vault, escrow, swap primitives for DeFi flows	0 contexts — foundation must prove out first	Deferred pending v1.0 cold-read pass rate
SETTL family (settlement)	Value transfer, finality, receipt primitives	0 contexts — pairs with MECH	Deferred
[?] explicit exploring state	Used in A&ID template but not in v1.0 decoder	1 context	Recurrence + cold-read

The queue itself is the proof the discipline is working. A standard with an empty queue is either perfect (unlikely) or unused (likely). A standard with a long queue is alive.

Context

• A&ID Template — A full worked diagram using this type system
• Notation Key — Pack node lines, scoring systems, commissioning levels — the local dictionary the type system writes against
• Naming Standards — Taxonomy, nomenclature, ontology — the layers this notation sits on
• Intercognitive Protocol — The protocol layer that carries the instructions this grammar describes
• Economy of Things — Machine identity, payment, and marketplace — what the AUTH and token families encode
• DePIN Devices — The physical agents (PA) that turn instructions into work
• Verifiable Intent — How V, G, and ⟳ close the loop with cryptographic proof
• Smart Contracts — Intelligent hyperlinks that carry instructions and energy
• Matrix Thinking — The human-side analogue: empty cells become the next thing to encode

Links

• ISA-5.1 Instrumentation Symbols and Identification — The physical process engineering standard this notation adapts
• BPMN 2.0 — The workflow notation A&ID extends with typed instruments
• llms.txt proposal — Distribution as the moat for LLM-readable standards

Questions

What recurring pattern in your work has not yet earned a symbol?

• If a fresh agent reads only this page and decodes a real diagram, which symbol proves hardest — and is the symbol wrong, or the row in the decoder?
• Where does a fail-safe default break down for a pattern you are trying to encode, and what does that tell you about the function code itself?
• At what point does a symbol system stop reducing complexity and start adding it — and how would you know you crossed that line?
• Which gap in this notation, once closed, would let an entirely new domain (financial mechanisms, physical systems, governance flows) be encoded in the same grammar?