Struct FormatMetadata 

pub struct FormatMetadata {

    pub raw_chars: usize,
    pub output_chars: usize,
    pub pre_trim_chars: usize,
    pub estimated_tokens: usize,
    pub compression_ratio: f32,
    pub format: String,
    pub truncated: bool,
    pub total_items: Option<usize>,
    pub included_items: usize,
    pub chunked: bool,
    pub total_chunks: usize,
    pub chunk_number: usize,
    pub provider_pagination: Option<Pagination>,
    pub provider_sort: Option<SortInfo>,
    pub dedup_savings_pct: f32,
    pub encoder_savings_pct: f32,
    pub combined_savings_pct: f32,
    pub baseline: String,
    pub tokenizer: String,
}

Metadata about formatting result — compression stats, token estimates.

Per Paper 2 §Savings Accounting, every quoted savings number must distinguish three orthogonal sources and name the baseline/tokenizer against which the percentages are taken. The split fields below (dedup_savings_pct, encoder_savings_pct, combined_savings_pct, baseline, tokenizer) encode that contract on the live response.

For typed-domain transforms (issues / merge_requests / …), the encoder runs without an L0 dedup hop, so dedup_savings_pct == 0.0 and combined == encoder. The cross-turn dedup contribution is reported separately by devboy-mcp::layered::SessionPipeline via the telemetry sink.

Fields

raw_chars: usize

Size of raw JSON input (UTF-8 bytes)

output_chars: usize

Size of formatted output (UTF-8 bytes)

pre_trim_chars: usize

Size of TOON/JSON output BEFORE budget trimming (UTF-8 bytes). If no trimming occurred, equals output_chars. toon_saved = raw_chars - pre_trim_chars trimmed_chars = pre_trim_chars - output_chars

estimated_tokens: usize

Estimated token count under the active tokenizer.

compression_ratio: f32

Compression ratio: output_chars / raw_chars (< 1.0 = savings)

format: String

truncated: bool

Whether output was truncated by budget trimming

total_items: Option<usize>

Total items before truncation (e.g., 50 issues)

included_items: usize

Items included after truncation (e.g., 20 issues)

chunked: bool

Whether response was split into chunks (budget exceeded)

total_chunks: usize

Number of chunks generated (1 = no chunking, >1 = chunked)

chunk_number: usize

Which chunk was requested (1 = first/default, >1 = navigation)

provider_pagination: Option<Pagination>

Pagination metadata from the provider (offset, limit, total, has_more)

provider_sort: Option<SortInfo>

Sort metadata from the provider (current sort, available sorts)

dedup_savings_pct: f32

L0 dedup savings as a fraction of baseline tokens (0.0 = no dedup hit on this response). Always 0.0 on the typed-domain path; populated by the MCP-server’s layered pipeline when a hint is emitted.

encoder_savings_pct: f32

L1/L2 encoder savings as a fraction of baseline tokens, computed over the L0-miss portion of the response. Equals 1.0 - encoded_tokens / baseline_tokens for the typed-domain path.

combined_savings_pct: f32

Multiplicative combination of dedup and encoder savings, per the §Savings Accounting reporting rule: combined = dedup + (1 - dedup) * encoder.

baseline: String

Baseline against which the percentages are taken (e.g. "json_pretty", "json_compact", "toon"). Required by the reporting rule — savings without a named baseline are not comparable across systems.

tokenizer: String

Tokenizer used to compute estimated_tokens and the savings percentages above (e.g. "o200k_base", "cl100k_base", "heuristic").

Trait Implementations

impl Clone for FormatMetadata

fn clone(&self) -> FormatMetadata

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for FormatMetadata

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Serialize for FormatMetadata

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.