Struct ChatOutcome 

pub struct ChatOutcome {

    pub text: String,
    pub reasoning: String,
    pub tool_calls: Vec<ToolCall>,
    pub finish_reason: Option<String>,
    pub ttft_ms: Option<f64>,
    pub tpot_ms: Option<f64>,
    pub e2e_ms: f64,
    pub arrival_gaps: GapSample,
    pub completion_tokens: usize,
    pub prompt_tokens: usize,
    pub cached_prompt_tokens: usize,
    pub server_ttft_ms: Option<f64>,
    pub server_tps: Option<f64>,
    pub accepted_prediction_tokens: Option<usize>,
    pub server_decode_time_ms: Option<f64>,
    pub server_total_time_ms: Option<f64>,
}

2026-09-26: Everything one request produced, plus its timings.

Fields

text: String

reasoning: String

2026-09-26: The streamed reasoning_content, kept apart from text. Each non-empty reasoning delta still counts toward completion_tokens and can start the TTFT clock.

tool_calls: Vec<ToolCall>

finish_reason: Option<String>

ttft_ms: Option<f64>

2026-09-26: Client clock: request write → first chunk that carried a reasoning, content or tool-call delta.

tpot_ms: Option<f64>

2026-09-26: Client-clock inter-token latency, called TPOT here: first token-carrying chunk → end of stream ([DONE] or EOF), over completion_tokens − 1 (itl_ms). The window ends at the final chunk, not the last token. None below two tokens.

e2e_ms: f64

2026-09-26: Client clock: request write → end of stream, the same end instant as tpot_ms.

arrival_gaps: GapSample

2026-09-26: Gaps between socket reads that carried a token (gaps).

completion_tokens: usize

2026-09-26: One per token-carrying reasoning or content delta; a usage.completion_tokens replaces the running count, and later deltas add to it.

prompt_tokens: usize

cached_prompt_tokens: usize

server_ttft_ms: Option<f64>

2026-09-26: usage.time_to_first_token_ms, on the server’s clock.

server_tps: Option<f64>

2026-09-26: usage."response_token/s", which this engine’s server computes as (completion_tokens − 1) / decode seconds (ir::Usage::decode_rate_tok_s in the server crate).

accepted_prediction_tokens: Option<usize>

2026-09-26: Accepted speculative draft tokens (usage.completion_tokens_details.accepted_prediction_tokens). None when the field is absent, so it differs from Some(0).

server_decode_time_ms: Option<f64>

2026-09-26: usage.decode_time_ms, the numerator of Self::server_tpot_ms. None when absent.

server_total_time_ms: Option<f64>

2026-09-26: usage.total_time_ms. None when absent.

Implementations

impl ChatOutcome

pub fn server_tpot_ms(&self) -> Option<f64>

2026-09-26: Server-clock inter-token latency: decode_time_ms / (completion_tokens − 1), the same itl_ms rule as Self::tpot_ms. None without the window or below two tokens.

Trait Implementations

impl Clone for ChatOutcome

fn clone(&self) -> ChatOutcome

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ChatOutcome

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

Formats the value using the given formatter.

impl Default for ChatOutcome

fn default() -> ChatOutcome

Returns the “default value” for a type.

impl From<&ChatOutcome> for Transcript

fn from(o: &ChatOutcome) -> Self

Converts to this type from the input type.