Tutorial #3: CGDL Signals, Part 1

What CGDL is for

CGDL is a disciplined-text notation with two major functions.

One function is building directed graphs incrementally.

The other function is passing signals. These signals may be used between software components, including from LLM to system, and they may also be used by people, especially in messaging forums and social networks.

This tutorial is about the signal side of CGDL.

Why signals matter

In ordinary writing, most distinctions must be expressed by ordinary language. That works well when both writer and reader are fluent and patient, and when no system needs to process the text in a disciplined way.

But many workflows are not like that.

A prompt may contain explanation, instruction, a formal block, a reusable anchor, and text intended for a downstream agent. A system may need to notice these differences immediately, without relying on deep linguistic interpretation of every line.

That is where CGDL signals become useful.

A signal is a short visible mark with a defined function. It appears in disciplined text when the writer or the system needs to mark something that should not be left to guesswork.

Compliance does not require signals everywhere

A text does not need to contain any signal in order to remain CGDL-compliant.

Plain text with zero signals is still compliant, as long as it does not use CGDL signals in the wrong way.

The role of CGDL is not to replace ordinary writing. Its role is to provide a small set of marks that can be added when a distinction needs to be made explicitly.

What counts as a CGDL signal

A CGDL signal is exactly two characters long.

A CGDL signal never contains letters or digits.

The characters apostrophe and dash are also excluded from signals.

This is a design rule. Signals should remain separate from ordinary words, names, numbers, and identifiers. That makes them easier for both people and systems to notice and interpret.

Most CGDL signals appear at the beginning of a line.

There are exceptions. Matching double parentheses may appear anywhere in the same line. Also, special paired markers may frame content in their own way.

Signals introduced so far

!!

Without parameter, this means emphasis of the preceding few lines. With parameter, it means emphasis of that parameter.

??

Without parameter, this means doubt about the preceding few lines. With parameter, it means doubt clarified by that parameter.

//

This signals that the rest of the line is in a formal language such as JSON.

##

This creates a new NODE. In a graph-building context this may be literal. In a more general context it unifies the following lines into a section.

@!

This signals that its parameter is intended to be a prompt to an LLM.

@@

At the beginning of a line, this begins addressing an AGENT described in the parameter, possibly even imaginably or “as if,” until the next @@-initial line.

{{ and }}

These can bound a block with a special instruction such as including a short description under a given condition by the current LLM.

[[ and ]]

These mean OFF THE RECORD.

Signals can be used without constant explanation

In disciplined text, it is acceptable, and often useful, to use a signal without re-explaining it every time.

A signal becomes part of the readable texture of the text. Once the reader or the system knows the function, the mark can do its job quickly and quietly.

This is one reason CGDL tries to keep signals short and memorable.

Signals are chosen, not assigned arbitrarily

A good signal is not only a free pair of characters.

The pair should fit the function.

It should be easy to recognize and hard to confuse with ordinary writing. It should be readable. It should carry some mnemonic value when possible.

For example, many readers already know that # marks something that may later be found again. CGDL extends that idea with ##, which introduces a NODE.

Likewise, many readers already know @-style attention from messaging culture around @ signs. In CGDL, @@ marks addressing an AGENT.

Such continuity helps disciplined text remain readable.

Prompt marking in CGDL

For AI application developers, one especially important signal is @!.

When @! appears at the beginning of a line, it marks that line as a potential prompt for an LLM.

This helps separate prompt material from surrounding explanation, notes, or workflow discussion.

In orchestration workflows, such separation can be very useful. A single region of text may contain several kinds of material. Without signals, these roles can become mixed in ways that are easy for a human to tolerate but harder for a system to process reliably.

Anchors and addressability

CGDL is not only about local line meaning. It is also about making disciplined text more addressable.

The signal ## introduces an intelligently referable anchor by creating a NODE.

This allows later retrieval, navigation, linking, and continuation to refer back to a marked point in the text.

In long prompts, long outputs, or multi-step workflows, such addressability becomes very useful.

Disciplined text can contain non-linguistic signals

Some distinctions are better carried by compact signals than by repeated natural-language explanation.

This is especially true in workflows involving software systems that are not as fluent as LLMs. There would be less need for such signals if all systems were highly fluent, but many systems are not.

In such settings, short digraphs can carry distinctions that would otherwise require much longer explanation.

That is one reason to extend the signal set carefully.

Summary

CGDL is a disciplined-text notation used both for graph-building and for passing signals.

This tutorial has focused on signals.

A CGDL signal is a two-character mark made from non-alphanumeric characters other than apostrophe and dash.

Signals help a system and a reader notice distinctions that should not be left implicit.

A text may remain fully compliant even when it contains no signals at all.

The currently discussed signal set already shows several important roles: emphasis, doubt, formal-language marking, prompt marking, node creation, agent addressing, bounded blocks, and off-the-record framing.