Struct Calibration 

pub struct Calibration {

    pub sigma_rate: Matrix9,
    pub block_length: usize,
    pub sigma_t: f64,
    pub l_r: Matrix9,
    pub prior_cov_marginal: Matrix9,
    pub theta_ns: f64,
    pub calibration_samples: usize,
    pub discrete_mode: bool,
    pub mde_ns: f64,
    pub calibration_snapshot: CalibrationSnapshot,
    pub timer_resolution_ns: f64,
    pub samples_per_second: f64,
    pub c_floor: f64,
    pub projection_mismatch_thresh: f64,
    pub theta_tick: f64,
    pub theta_eff: f64,
    pub theta_floor_initial: f64,
    pub rng_seed: u64,
    pub batch_k: u32,
    pub preflight_result: PreflightResult,
}

Calibration results from the initial measurement phase (no_std compatible).

This struct contains the essential statistical data needed for the adaptive sampling loop. It is designed for use in no_std environments like SGX enclaves.

For full calibration with preflight checks, see tacet::Calibration.

Student’s t Prior (v5.4+)

The prior is a Student’s t distribution with ν=4 degrees of freedom: δ ~ t_ν(0, σ_t²R)

This is implemented via scale mixture: λ ~ Gamma(ν/2, ν/2), δ|λ ~ N(0, (σ_t²/λ)R). The marginal variance is (ν/(ν-2)) σ_t² R = 2σ_t² R for ν=4.

Fields

sigma_rate: Matrix9

Covariance “rate” - multiply by 1/n to get Sigma_n for n samples. Computed as Sigma_cal * n_cal where Sigma_cal is calibration covariance. This allows O(1) covariance scaling as samples accumulate.

block_length: usize

Block length from Politis-White algorithm. Used for block bootstrap to preserve autocorrelation structure.

sigma_t: f64

Calibrated Student’s t prior scale (v5.4). This is the σ in δ|λ ~ N(0, (σ²/λ)R).

l_r: Matrix9

Cholesky factor L_R of correlation matrix R. Used for Gibbs sampling: δ = (σ/√λ) L_R z.

prior_cov_marginal: Matrix9

Marginal prior covariance: 2σ²R (for ν=4). This is the unconditional prior variance of δ under the t-prior.

theta_ns: f64

The theta threshold being used (in nanoseconds).

calibration_samples: usize

Number of calibration samples collected per class.

discrete_mode: bool

Whether discrete mode is active (< 10% unique values). When true, use mid-quantile estimators and m-out-of-n bootstrap.

mde_ns: f64

Minimum detectable effect in nanoseconds.

calibration_snapshot: CalibrationSnapshot

Statistics snapshot from calibration phase for drift detection.

timer_resolution_ns: f64

Timer resolution in nanoseconds.

samples_per_second: f64

Measured throughput for time estimation (samples per second). The caller is responsible for measuring this during calibration.

c_floor: f64

Floor-rate constant. Computed once at calibration: 95th percentile of max_k|Z_k| where Z ~ N(0, Σ_rate). Used for analytical theta_floor computation: theta_floor_stat(n) = c_floor / sqrt(n).

projection_mismatch_thresh: f64

Projection mismatch threshold. 99th percentile of bootstrap Q_proj distribution.

theta_tick: f64

Timer resolution floor component. theta_tick = (1 tick in ns) / K where K is the batch size.

theta_eff: f64

Effective threshold for this run. theta_eff = max(theta_user, theta_floor) or just theta_floor in research mode.

theta_floor_initial: f64

Initial measurement floor at calibration time.

rng_seed: u64

Deterministic RNG seed used for this run.

batch_k: u32

Batch size K for adaptive batching. When K > 1, samples contain K iterations worth of timing. Effect estimates must be divided by K to report per-call differences.

preflight_result: PreflightResult

Results of core preflight checks run during calibration (no_std compatible). Platform-specific checks (like system configuration) are handled by the tacet crate’s wrapper.

Implementations

impl Calibration

pub fn new( sigma_rate: Matrix9, block_length: usize, sigma_t: f64, l_r: Matrix9, theta_ns: f64, calibration_samples: usize, discrete_mode: bool, mde_ns: f64, calibration_snapshot: CalibrationSnapshot, timer_resolution_ns: f64, samples_per_second: f64, c_floor: f64, projection_mismatch_thresh: f64, theta_tick: f64, theta_eff: f64, theta_floor_initial: f64, rng_seed: u64, batch_k: u32, ) -> Self

Create a new Calibration with v5.4+ Student’s t prior.

pub fn with_preflight( sigma_rate: Matrix9, block_length: usize, sigma_t: f64, l_r: Matrix9, theta_ns: f64, calibration_samples: usize, discrete_mode: bool, mde_ns: f64, calibration_snapshot: CalibrationSnapshot, timer_resolution_ns: f64, samples_per_second: f64, c_floor: f64, projection_mismatch_thresh: f64, theta_tick: f64, theta_eff: f64, theta_floor_initial: f64, rng_seed: u64, batch_k: u32, preflight_result: PreflightResult, ) -> Self

Create a new Calibration with all fields including preflight results.

pub fn n_eff(&self, n: usize) -> usize

Compute effective sample size accounting for dependence (spec §3.3.2 v5.6).

Under strong temporal dependence, n samples do not provide n independent observations. The effective sample size approximates the number of effectively independent blocks.

n_eff = max(1, floor(n / block_length))

where block_length is the estimated dependence length from Politis-White selector.

pub fn covariance_for_n(&self, n: usize) -> Matrix9

Scale sigma_rate to get covariance for n samples using effective sample size (v5.6).

Σ_n = Σ_rate / n_eff

where n_eff = max(1, floor(n / block_length)) to correctly account for reduced information under temporal dependence.

pub fn covariance_for_n_raw(&self, n: usize) -> Matrix9

Scale sigma_rate to get covariance using raw n samples (ignoring dependence).

Σ_n = Σ_rate / n

Use this only when you need the raw scaling without n_eff correction.

pub fn estimate_collection_time_secs(&self, n: usize) -> f64

Estimate time to collect n additional samples based on calibration throughput.

Returns estimated seconds. Caller should add any overhead.

pub fn to_summary(&self) -> CalibrationSummary

Convert to an FFI-friendly summary containing only scalar fields.

Trait Implementations

impl Clone for Calibration

fn clone(&self) -> Calibration

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Calibration

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

Formats the value using the given formatter.