Struct RequestMeta 

pub struct RequestMeta<'a> {
    pub request_id: &'a str,
    pub tenant: &'a str,
    pub principal: Option<&'a str>,
    pub model: &'a str,
    pub route: &'static str,
    pub lane: &'static str,
    pub stream: bool,
    pub max_tokens: Option<u64>,
    pub reserved_ctx: Option<u64>,
}

The request identity a receipt is opened with. All borrowed: this is a view of state the handler already owns, taken for the duration of one open call.

Fields

request_id: &'a str

tenant: &'a str

principal: Option<&'a str>

The authenticated key’s non-secret identification prefix, when the request carries per-key identity (multi-key ring). What per-PRINCIPAL policy means — spend caps, quotas — is the implementation’s business.

model: &'a str

route: &'static str

lane: &'static str

stream: bool

max_tokens: Option<u64>

The request’s EFFECTIVE completion-token bound at receipt time: the caller’s max_tokens, or the model-default/max_output it resolved to (apply_model_request_limits); None when no bound resolved (context-bounded generation). D2 gap G4 (darklanes research/d2-shadow-20260831/RESULTS.md): the ledger row carries this so live-shadow replay reads the real bound instead of recovering it by inverting the billing reservation. An implementation with a positional row format appends this as a NEW trailing column (column order in an existing ledger is an API, so the discipline is append-only).

reserved_ctx: Option<u64>

The context reservation the budget admission charged for this request: prompt_tokens + completion bound, in tokens (the same quantities handed to Metering::reserve). None when no reservation ran (limits not enforced, or the receipt records a pre-reservation rejection). Same G4 append-only column rule as max_tokens.