content_chars

content_chars is how an agent knows the token cost of a node before fetching it. If your declared count is wrong, agents budget wrong. This check keeps it honest.

content_chars is the declared Unicode code point count for a clean llm_url response.

The validator checks content_chars as part of Level 2a Agent Index validation through both validateIndexAi() and the index-ai CLI.

What is implemented now

The validator now checks content_chars for fetched clean endpoints.

Implemented behavior:

• clean endpoint fetch through each node content.llm_url
• allowed clean endpoint content types: text/markdown and text/plain
• Unicode NFC normalization before counting
• code point counting, not UTF-16 .length
• content_chars_mode: exact
• content_chars_mode: max
• content_chars must be an integer greater than or equal to 1
• emoji counts as one code point
• decomposed accents are normalized before counting

Why .length is not enough

JavaScript string .length counts UTF-16 code units. That is not the same as the character count required for content_chars.

Examples:

abc        -> 3 code points
e + accent -> normalized, then 1 code point
rocket     -> 1 code point, but JavaScript .length is 2

The validator normalizes text to Unicode NFC before counting code points. This makes composed and decomposed accented text count consistently.

STEP-1 - Fetch the clean content

For Level 2a, each graph node declares:

content.llm_url

The validator resolves that URL, fetches it, and checks that the response is served as one of:

text/markdown
text/plain

STEP-2 - Normalize and count

The fetched body is normalized to Unicode NFC before counting.

e + combining acute accent -> one normalized code point

The count must be a whole number. Decimal values such as 1.5 are invalid, and 0 is invalid because zero is only a placeholder.

STEP-3 - Compare by mode

content_chars_mode controls how the declared value is compared.

Mode	Rule
exact	The measured count must equal content_chars.
max	The measured count must be less than or equal to content_chars.

Measurement flow

Scope

content_chars validation runs on Level 2a Agent Index clean endpoints through validateIndexAi(). This is experimental documentation of current validator behavior, not compliance certification or a traffic promise. See Scope.