tacet_core/adaptive/

step.rs

//! Adaptive sampling step logic (no_std compatible).
//!
//! This module provides the core adaptive step logic that can be used in
//! no_std environments. Time tracking is caller-provided rather than using
//! `std::time::Instant`.
//!
//! # Usage Pattern
//!
//! ```ignore
//! let mut state = AdaptiveState::new();
//! let calibration = /* ... */;
//! let config = AdaptiveStepConfig::default();
//!
//! loop {
//!     // Caller collects samples and tracks time
//!     state.add_batch(baseline, sample);
//!     let elapsed_secs = /* measured by caller */;
//!
//!     match adaptive_step(&calibration, &mut state, ns_per_tick, elapsed_secs, &config) {
//!         StepResult::Decision(outcome) => return outcome,
//!         StepResult::Continue { .. } => continue,
//!     }
//! }
//! ```

use alloc::string::String;

use crate::analysis::bayes::compute_bayes_gibbs;
use crate::constants::{
    DEFAULT_FAIL_THRESHOLD, DEFAULT_MAX_SAMPLES, DEFAULT_PASS_THRESHOLD, DEFAULT_SEED,
};
use crate::statistics::{compute_deciles_inplace, compute_midquantile_deciles};

use super::{
    check_quality_gates, compute_achievable_at_max, is_threshold_elevated, AdaptiveState,
    Calibration, InconclusiveReason, Posterior, QualityGateCheckInputs, QualityGateConfig,
    QualityGateResult,
};

/// Configuration for the adaptive step (no_std compatible).
///
/// Uses `f64` for time values instead of `std::time::Duration`.
#[derive(Debug, Clone)]
pub struct AdaptiveStepConfig {
    /// Threshold below which we pass (no significant leak).
    pub pass_threshold: f64,

    /// Threshold above which we fail (leak detected).
    pub fail_threshold: f64,

    /// Time budget in seconds.
    pub time_budget_secs: f64,

    /// Maximum samples per class.
    pub max_samples: usize,

    /// Effect threshold (theta) in nanoseconds.
    pub theta_ns: f64,

    /// Random seed for Monte Carlo integration.
    pub seed: u64,

    /// Quality gate configuration.
    pub quality_gates: QualityGateConfig,
}

impl Default for AdaptiveStepConfig {
    fn default() -> Self {
        Self {
            pass_threshold: DEFAULT_PASS_THRESHOLD,
            fail_threshold: DEFAULT_FAIL_THRESHOLD,
            time_budget_secs: 30.0,
            max_samples: DEFAULT_MAX_SAMPLES,
            theta_ns: 100.0,
            seed: DEFAULT_SEED,
            quality_gates: QualityGateConfig::default(),
        }
    }
}

impl AdaptiveStepConfig {
    /// Create a new config with the given theta threshold.
    pub fn with_theta(theta_ns: f64) -> Self {
        let mut config = Self {
            theta_ns,
            ..Self::default()
        };
        config.quality_gates.pass_threshold = config.pass_threshold;
        config.quality_gates.fail_threshold = config.fail_threshold;
        config.quality_gates.time_budget_secs = config.time_budget_secs;
        config
    }

    /// Builder method to set pass threshold.
    pub fn pass_threshold(mut self, threshold: f64) -> Self {
        self.pass_threshold = threshold;
        self.quality_gates.pass_threshold = threshold;
        self
    }

    /// Builder method to set fail threshold.
    pub fn fail_threshold(mut self, threshold: f64) -> Self {
        self.fail_threshold = threshold;
        self.quality_gates.fail_threshold = threshold;
        self
    }

    /// Builder method to set time budget in seconds.
    pub fn time_budget_secs(mut self, secs: f64) -> Self {
        self.time_budget_secs = secs;
        self.quality_gates.time_budget_secs = secs;
        self
    }

    /// Builder method to set max samples.
    pub fn max_samples(mut self, max: usize) -> Self {
        self.max_samples = max;
        self.quality_gates.max_samples = max;
        self
    }
}

/// Outcome of an adaptive sampling decision (no_std compatible).
///
/// Uses `f64` for elapsed time instead of `std::time::Duration`.
#[derive(Debug, Clone)]
pub enum AdaptiveOutcome {
    /// Leak probability exceeded fail threshold - timing leak detected.
    LeakDetected {
        /// Final posterior distribution.
        posterior: Posterior,
        /// Number of samples collected per class.
        samples_per_class: usize,
        /// Time spent in seconds.
        elapsed_secs: f64,
    },

    /// Leak probability dropped below pass threshold - no significant leak.
    NoLeakDetected {
        /// Final posterior distribution.
        posterior: Posterior,
        /// Number of samples collected per class.
        samples_per_class: usize,
        /// Time spent in seconds.
        elapsed_secs: f64,
    },

    /// A quality gate triggered before reaching a decision.
    Inconclusive {
        /// Reason for stopping.
        reason: InconclusiveReason,
        /// Final posterior distribution (if available).
        posterior: Option<Posterior>,
        /// Number of samples collected per class.
        samples_per_class: usize,
        /// Time spent in seconds.
        elapsed_secs: f64,
    },

    /// Threshold was elevated and pass criterion was met at effective threshold (v5.5).
    ///
    /// P < pass_threshold at θ_eff, but θ_eff > θ_user + ε. This is semantically
    /// distinct from both Pass (can't guarantee user threshold) and Inconclusive
    /// (not a quality issue - we got a clear statistical result).
    ThresholdElevated {
        /// Final posterior distribution.
        posterior: Posterior,
        /// User's requested threshold (θ_user).
        theta_user: f64,
        /// Effective threshold used (θ_eff).
        theta_eff: f64,
        /// Timer tick floor (θ_tick).
        theta_tick: f64,
        /// Whether threshold is achievable at max_samples.
        achievable_at_max: bool,
        /// Number of samples collected per class.
        samples_per_class: usize,
        /// Time spent in seconds.
        elapsed_secs: f64,
    },
}

impl AdaptiveOutcome {
    /// Get the final leak probability, if available.
    pub fn leak_probability(&self) -> Option<f64> {
        match self {
            AdaptiveOutcome::LeakDetected { posterior, .. } => Some(posterior.leak_probability),
            AdaptiveOutcome::NoLeakDetected { posterior, .. } => Some(posterior.leak_probability),
            AdaptiveOutcome::ThresholdElevated { posterior, .. } => {
                Some(posterior.leak_probability)
            }
            AdaptiveOutcome::Inconclusive { posterior, .. } => {
                posterior.as_ref().map(|p| p.leak_probability)
            }
        }
    }

    /// Check if the outcome indicates a leak was detected.
    pub fn is_leak_detected(&self) -> bool {
        matches!(self, AdaptiveOutcome::LeakDetected { .. })
    }

    /// Check if the outcome is conclusive (either pass or fail).
    ///
    /// Note: ThresholdElevated is NOT considered conclusive in v5.5 - it means
    /// we got a statistical result but can't guarantee the user's threshold.
    pub fn is_conclusive(&self) -> bool {
        matches!(
            self,
            AdaptiveOutcome::LeakDetected { .. } | AdaptiveOutcome::NoLeakDetected { .. }
        )
    }

    /// Check if the threshold was elevated beyond tolerance (v5.5).
    pub fn is_threshold_elevated(&self) -> bool {
        matches!(self, AdaptiveOutcome::ThresholdElevated { .. })
    }

    /// Get the number of samples per class.
    pub fn samples_per_class(&self) -> usize {
        match self {
            AdaptiveOutcome::LeakDetected {
                samples_per_class, ..
            } => *samples_per_class,
            AdaptiveOutcome::NoLeakDetected {
                samples_per_class, ..
            } => *samples_per_class,
            AdaptiveOutcome::ThresholdElevated {
                samples_per_class, ..
            } => *samples_per_class,
            AdaptiveOutcome::Inconclusive {
                samples_per_class, ..
            } => *samples_per_class,
        }
    }

    /// Get elapsed time in seconds.
    pub fn elapsed_secs(&self) -> f64 {
        match self {
            AdaptiveOutcome::LeakDetected { elapsed_secs, .. } => *elapsed_secs,
            AdaptiveOutcome::NoLeakDetected { elapsed_secs, .. } => *elapsed_secs,
            AdaptiveOutcome::ThresholdElevated { elapsed_secs, .. } => *elapsed_secs,
            AdaptiveOutcome::Inconclusive { elapsed_secs, .. } => *elapsed_secs,
        }
    }

    /// Convert to an FFI-friendly summary containing only scalar fields.
    ///
    /// This centralizes the conversion logic that was previously duplicated
    /// across NAPI and C bindings, using canonical effect pattern classification.
    pub fn to_summary(
        &self,
        calibration: &super::Calibration,
    ) -> crate::ffi_summary::OutcomeSummary {
        use crate::ffi_summary::{EffectSummary, InconclusiveReasonKind, OutcomeSummary, OutcomeType};
        use crate::result::{Exploitability, MeasurementQuality};

        let cal_summary = calibration.to_summary();

        match self {
            AdaptiveOutcome::LeakDetected {
                posterior,
                samples_per_class,
                elapsed_secs,
            } => {
                let post_summary = posterior.to_summary();
                let effect = build_effect_summary(&post_summary);
                let diagnostics = build_diagnostics(&post_summary, &cal_summary);

                OutcomeSummary {
                    outcome_type: OutcomeType::Fail,
                    leak_probability: post_summary.leak_probability,
                    samples_per_class: *samples_per_class,
                    elapsed_secs: *elapsed_secs,
                    effect,
                    quality: post_summary.measurement_quality(),
                    exploitability: post_summary.exploitability(),
                    inconclusive_reason: InconclusiveReasonKind::None,
                    recommendation: String::new(),
                    theta_user: cal_summary.theta_ns,
                    theta_eff: cal_summary.theta_eff,
                    theta_floor: cal_summary.theta_floor_initial,
                    theta_tick: cal_summary.theta_tick,
                    achievable_at_max: true,
                    diagnostics,
                    mde_shift_ns: cal_summary.mde_shift_ns,
                    mde_tail_ns: cal_summary.mde_tail_ns,
                }
            }

            AdaptiveOutcome::NoLeakDetected {
                posterior,
                samples_per_class,
                elapsed_secs,
            } => {
                let post_summary = posterior.to_summary();
                let effect = build_effect_summary(&post_summary);
                let diagnostics = build_diagnostics(&post_summary, &cal_summary);

                OutcomeSummary {
                    outcome_type: OutcomeType::Pass,
                    leak_probability: post_summary.leak_probability,
                    samples_per_class: *samples_per_class,
                    elapsed_secs: *elapsed_secs,
                    effect,
                    quality: post_summary.measurement_quality(),
                    exploitability: Exploitability::SharedHardwareOnly,
                    inconclusive_reason: InconclusiveReasonKind::None,
                    recommendation: String::new(),
                    theta_user: cal_summary.theta_ns,
                    theta_eff: cal_summary.theta_eff,
                    theta_floor: cal_summary.theta_floor_initial,
                    theta_tick: cal_summary.theta_tick,
                    achievable_at_max: true,
                    diagnostics,
                    mde_shift_ns: cal_summary.mde_shift_ns,
                    mde_tail_ns: cal_summary.mde_tail_ns,
                }
            }

            AdaptiveOutcome::ThresholdElevated {
                posterior,
                theta_user,
                theta_eff,
                theta_tick,
                achievable_at_max,
                samples_per_class,
                elapsed_secs,
            } => {
                let post_summary = posterior.to_summary();
                let effect = build_effect_summary(&post_summary);
                let diagnostics = build_diagnostics(&post_summary, &cal_summary);

                let recommendation = if *achievable_at_max {
                    alloc::format!(
                        "Threshold elevated from {:.0}ns to {:.1}ns. More samples could achieve the requested threshold.",
                        theta_user, theta_eff
                    )
                } else {
                    alloc::format!(
                        "Threshold elevated from {:.0}ns to {:.1}ns. Use a cycle counter for better resolution.",
                        theta_user, theta_eff
                    )
                };

                OutcomeSummary {
                    outcome_type: OutcomeType::ThresholdElevated,
                    leak_probability: post_summary.leak_probability,
                    samples_per_class: *samples_per_class,
                    elapsed_secs: *elapsed_secs,
                    effect,
                    quality: post_summary.measurement_quality(),
                    exploitability: Exploitability::SharedHardwareOnly,
                    inconclusive_reason: InconclusiveReasonKind::ThresholdElevated,
                    recommendation,
                    theta_user: *theta_user,
                    theta_eff: *theta_eff,
                    theta_floor: cal_summary.theta_floor_initial,
                    theta_tick: *theta_tick,
                    achievable_at_max: *achievable_at_max,
                    diagnostics,
                    mde_shift_ns: cal_summary.mde_shift_ns,
                    mde_tail_ns: cal_summary.mde_tail_ns,
                }
            }

            AdaptiveOutcome::Inconclusive {
                reason,
                posterior,
                samples_per_class,
                elapsed_secs,
            } => {
                let (post_summary, effect, quality, diagnostics) = match posterior {
                    Some(p) => {
                        let ps = p.to_summary();
                        let eff = build_effect_summary(&ps);
                        let qual = ps.measurement_quality();
                        let diag = build_diagnostics(&ps, &cal_summary);
                        (Some(ps), eff, qual, diag)
                    }
                    None => (
                        None,
                        EffectSummary::default(),
                        MeasurementQuality::TooNoisy,
                        build_diagnostics_from_calibration(&cal_summary),
                    ),
                };

                let (inconclusive_reason, recommendation) = convert_inconclusive_reason(reason);

                OutcomeSummary {
                    outcome_type: OutcomeType::Inconclusive,
                    leak_probability: post_summary
                        .as_ref()
                        .map(|p| p.leak_probability)
                        .unwrap_or(0.5),
                    samples_per_class: *samples_per_class,
                    elapsed_secs: *elapsed_secs,
                    effect,
                    quality,
                    exploitability: Exploitability::SharedHardwareOnly,
                    inconclusive_reason,
                    recommendation,
                    theta_user: cal_summary.theta_ns,
                    theta_eff: cal_summary.theta_eff,
                    theta_floor: cal_summary.theta_floor_initial,
                    theta_tick: cal_summary.theta_tick,
                    achievable_at_max: false,
                    diagnostics,
                    mde_shift_ns: cal_summary.mde_shift_ns,
                    mde_tail_ns: cal_summary.mde_tail_ns,
                }
            }
        }
    }
}

/// Build an EffectSummary from a PosteriorSummary.
fn build_effect_summary(
    post: &crate::ffi_summary::PosteriorSummary,
) -> crate::ffi_summary::EffectSummary {
    crate::ffi_summary::EffectSummary {
        shift_ns: post.shift_ns,
        tail_ns: post.tail_ns,
        ci_low_ns: post.ci_low_ns,
        ci_high_ns: post.ci_high_ns,
        pattern: post.pattern,
        interpretation_caveat: None,
    }
}

/// Build DiagnosticsSummary from posterior and calibration summaries.
fn build_diagnostics(
    post: &crate::ffi_summary::PosteriorSummary,
    cal: &crate::ffi_summary::CalibrationSummary,
) -> crate::ffi_summary::DiagnosticsSummary {
    crate::ffi_summary::DiagnosticsSummary {
        dependence_length: cal.block_length,
        effective_sample_size: post.n,
        stationarity_ratio: 1.0,
        stationarity_ok: true,
        projection_mismatch_q: post.projection_mismatch_q,
        projection_mismatch_ok: post.projection_mismatch_q <= cal.projection_mismatch_thresh,
        discrete_mode: cal.discrete_mode,
        timer_resolution_ns: cal.timer_resolution_ns,
        lambda_mean: post.lambda_mean,
        lambda_mixing_ok: post.lambda_mixing_ok,
        kappa_mean: post.kappa_mean,
        kappa_cv: post.kappa_cv,
        kappa_ess: post.kappa_ess,
        kappa_mixing_ok: post.kappa_mixing_ok,
    }
}

/// Build DiagnosticsSummary from calibration only (when no posterior available).
fn build_diagnostics_from_calibration(
    cal: &crate::ffi_summary::CalibrationSummary,
) -> crate::ffi_summary::DiagnosticsSummary {
    crate::ffi_summary::DiagnosticsSummary {
        dependence_length: cal.block_length,
        effective_sample_size: 0,
        stationarity_ratio: 1.0,
        stationarity_ok: true,
        projection_mismatch_q: 0.0,
        projection_mismatch_ok: true,
        discrete_mode: cal.discrete_mode,
        timer_resolution_ns: cal.timer_resolution_ns,
        lambda_mean: 1.0,
        lambda_mixing_ok: true,
        kappa_mean: 1.0,
        kappa_cv: 0.0,
        kappa_ess: 0.0,
        kappa_mixing_ok: true,
    }
}

/// Convert InconclusiveReason to FFI kind and recommendation string.
fn convert_inconclusive_reason(
    reason: &super::InconclusiveReason,
) -> (crate::ffi_summary::InconclusiveReasonKind, String) {
    use crate::ffi_summary::InconclusiveReasonKind;
    use super::InconclusiveReason;

    match reason {
        InconclusiveReason::DataTooNoisy { guidance, .. } => {
            (InconclusiveReasonKind::DataTooNoisy, guidance.clone())
        }
        InconclusiveReason::NotLearning { guidance, .. } => {
            (InconclusiveReasonKind::NotLearning, guidance.clone())
        }
        InconclusiveReason::WouldTakeTooLong { guidance, .. } => {
            (InconclusiveReasonKind::WouldTakeTooLong, guidance.clone())
        }
        InconclusiveReason::TimeBudgetExceeded { .. } => (
            InconclusiveReasonKind::TimeBudgetExceeded,
            String::from("Increase time budget or reduce threshold"),
        ),
        InconclusiveReason::SampleBudgetExceeded { .. } => (
            InconclusiveReasonKind::SampleBudgetExceeded,
            String::from("Increase sample budget or reduce threshold"),
        ),
        InconclusiveReason::ConditionsChanged { guidance, .. } => {
            (InconclusiveReasonKind::ConditionsChanged, guidance.clone())
        }
        InconclusiveReason::ThresholdElevated { guidance, .. } => {
            (InconclusiveReasonKind::ThresholdElevated, guidance.clone())
        }
    }
}

/// Result of a single adaptive step.
#[derive(Debug, Clone)]
pub enum StepResult {
    /// A decision was reached (Pass, Fail, or Inconclusive).
    Decision(AdaptiveOutcome),

    /// Continue sampling - no decision yet.
    Continue {
        /// Current posterior distribution.
        posterior: Posterior,
        /// Number of samples collected per class so far.
        samples_per_class: usize,
    },
}

impl StepResult {
    /// Check if a decision was reached.
    pub fn is_decision(&self) -> bool {
        matches!(self, StepResult::Decision(_))
    }

    /// Get the outcome if a decision was reached.
    pub fn into_decision(self) -> Option<AdaptiveOutcome> {
        match self {
            StepResult::Decision(outcome) => Some(outcome),
            StepResult::Continue { .. } => None,
        }
    }

    /// Get the current leak probability.
    pub fn leak_probability(&self) -> Option<f64> {
        match self {
            StepResult::Decision(outcome) => outcome.leak_probability(),
            StepResult::Continue { posterior, .. } => Some(posterior.leak_probability),
        }
    }
}

/// Single-iteration adaptive step (no_std compatible).
///
/// This function performs one iteration of the adaptive sampling loop:
/// 1. Computes the posterior from current samples
/// 2. Checks if decision thresholds are met
/// 3. Checks quality gates
/// 4. Returns whether to continue or stop
///
/// # Arguments
///
/// * `calibration` - Results from calibration phase
/// * `state` - Current adaptive state with accumulated samples
/// * `ns_per_tick` - Conversion factor from native units to nanoseconds
/// * `elapsed_secs` - Elapsed time in seconds (provided by caller)
/// * `config` - Adaptive step configuration
///
/// # Returns
///
/// - `StepResult::Decision(outcome)` - Decision reached or quality gate triggered
/// - `StepResult::Continue { .. }` - Need more samples
pub fn adaptive_step(
    calibration: &Calibration,
    state: &mut AdaptiveState,
    ns_per_tick: f64,
    elapsed_secs: f64,
    config: &AdaptiveStepConfig,
) -> StepResult {
    // Compute posterior from current samples
    let posterior = match compute_posterior(state, calibration, ns_per_tick, config) {
        Some(p) => p,
        None => {
            // Not enough samples for reliable posterior
            if state.n_total() < 20 {
                // Need more samples - not a failure, just need to continue
                // We can't return Continue without a posterior, so return a dummy
                // Actually, let's just return Inconclusive with a clear message
                return StepResult::Decision(AdaptiveOutcome::Inconclusive {
                    reason: InconclusiveReason::DataTooNoisy {
                        message: String::from("Insufficient samples for posterior computation"),
                        guidance: String::from("Need at least 20 samples per class"),
                        variance_ratio: 1.0,
                    },
                    posterior: None,
                    samples_per_class: state.n_total(),
                    elapsed_secs,
                });
            }
            // Computation failed for other reasons
            return StepResult::Decision(AdaptiveOutcome::Inconclusive {
                reason: InconclusiveReason::DataTooNoisy {
                    message: String::from("Could not compute posterior from samples"),
                    guidance: String::from("Check timer resolution and sample count"),
                    variance_ratio: 1.0,
                },
                posterior: None,
                samples_per_class: state.n_total(),
                elapsed_secs,
            });
        }
    };

    // Track KL divergence for learning rate monitoring
    let _kl = state.update_posterior(posterior.clone());

    // Check decision boundaries (v5.5 threshold elevation decision rule)
    //
    // Fail propagates regardless of threshold elevation: if P > fail_threshold,
    // we detected a leak even at the elevated threshold.
    if posterior.leak_probability > config.fail_threshold {
        return StepResult::Decision(AdaptiveOutcome::LeakDetected {
            posterior,
            samples_per_class: state.n_total(),
            elapsed_secs,
        });
    }

    // Pass requires both P < pass_threshold AND θ_eff ≤ θ_user + ε (v5.5)
    if posterior.leak_probability < config.pass_threshold {
        // Check if threshold is elevated beyond tolerance
        let theta_user = config.theta_ns;
        let theta_eff = calibration.theta_eff;
        let theta_tick = calibration.theta_tick;

        if is_threshold_elevated(theta_eff, theta_user, theta_tick) {
            // Threshold elevated: return ThresholdElevated instead of Pass
            let achievable_at_max = compute_achievable_at_max(
                calibration.c_floor,
                theta_tick,
                theta_user,
                config.max_samples,
                calibration.block_length, // v5.6: block_length for n_eff computation
            );

            return StepResult::Decision(AdaptiveOutcome::ThresholdElevated {
                posterior,
                theta_user,
                theta_eff,
                theta_tick,
                achievable_at_max,
                samples_per_class: state.n_total(),
                elapsed_secs,
            });
        }

        // Threshold not elevated: true Pass
        return StepResult::Decision(AdaptiveOutcome::NoLeakDetected {
            posterior,
            samples_per_class: state.n_total(),
            elapsed_secs,
        });
    }

    // Check quality gates
    let current_stats = state.get_stats_snapshot();
    let gate_inputs = QualityGateCheckInputs {
        posterior: &posterior,
        prior_cov_marginal: &calibration.prior_cov_marginal,
        theta_ns: config.theta_ns,
        n_total: state.n_total(),
        elapsed_secs,
        recent_kl_sum: if state.has_kl_history() {
            Some(state.recent_kl_sum())
        } else {
            None
        },
        samples_per_second: calibration.samples_per_second,
        calibration_snapshot: Some(&calibration.calibration_snapshot),
        current_stats_snapshot: current_stats.as_ref(),
        c_floor: calibration.c_floor,
        theta_tick: calibration.theta_tick,
        projection_mismatch_q: if posterior.projection_mismatch_q.is_nan() {
            None
        } else {
            Some(posterior.projection_mismatch_q)
        },
        projection_mismatch_thresh: calibration.projection_mismatch_thresh,
        lambda_mixing_ok: posterior.lambda_mixing_ok,
    };

    match check_quality_gates(&gate_inputs, &config.quality_gates) {
        QualityGateResult::Continue => StepResult::Continue {
            posterior,
            samples_per_class: state.n_total(),
        },
        QualityGateResult::Stop(reason) => StepResult::Decision(AdaptiveOutcome::Inconclusive {
            reason,
            posterior: Some(posterior),
            samples_per_class: state.n_total(),
            elapsed_secs,
        }),
    }
}

/// Compute posterior distribution from current state.
///
/// Uses scaled covariance: Sigma_n = Sigma_rate / n
fn compute_posterior(
    state: &AdaptiveState,
    calibration: &Calibration,
    ns_per_tick: f64,
    config: &AdaptiveStepConfig,
) -> Option<Posterior> {
    let n = state.n_total();
    if n < 20 {
        return None; // Need minimum samples for stable quantiles
    }

    // Convert samples to nanoseconds
    let baseline_ns = state.baseline_ns(ns_per_tick);
    let sample_ns = state.sample_ns(ns_per_tick);

    // Compute quantile differences
    let observed_diff = if calibration.discrete_mode {
        let q_baseline = compute_midquantile_deciles(&baseline_ns);
        let q_sample = compute_midquantile_deciles(&sample_ns);
        q_baseline - q_sample
    } else {
        let mut baseline_sorted = baseline_ns;
        let mut sample_sorted = sample_ns;
        let q_baseline = compute_deciles_inplace(&mut baseline_sorted);
        let q_sample = compute_deciles_inplace(&mut sample_sorted);
        q_baseline - q_sample
    };

    // Scale covariance: Sigma_n = Sigma_rate / n
    let sigma_n = calibration.covariance_for_n(n);

    // Run 9D Bayesian inference with v5.4 Gibbs sampler
    let bayes_result = compute_bayes_gibbs(
        &observed_diff,
        &sigma_n,
        calibration.sigma_t,
        &calibration.l_r,
        config.theta_ns,
        Some(config.seed),
    );

    Some(Posterior::new_with_gibbs(
        bayes_result.delta_post,
        bayes_result.lambda_post,
        bayes_result.beta_proj,
        bayes_result.beta_proj_cov,
        bayes_result.beta_draws,
        bayes_result.leak_probability,
        bayes_result.projection_mismatch_q,
        n,
        bayes_result.lambda_mean,
        bayes_result.lambda_mixing_ok,
        // v5.6: kappa diagnostics
        bayes_result.kappa_mean,
        bayes_result.kappa_cv,
        bayes_result.kappa_ess,
        bayes_result.kappa_mixing_ok,
    ))
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::statistics::StatsSnapshot;
    use crate::types::{Matrix2, Matrix9, Vector2, Vector9};

    fn make_test_calibration() -> Calibration {
        use crate::adaptive::CalibrationSnapshot;

        let snapshot = CalibrationSnapshot::new(
            StatsSnapshot {
                count: 5000,
                mean: 1000.0,
                variance: 25.0,
                autocorr_lag1: 0.1,
            },
            StatsSnapshot {
                count: 5000,
                mean: 1000.0,
                variance: 25.0,
                autocorr_lag1: 0.1,
            },
        );

        // v5.4+ t-prior
        Calibration::new(
            Matrix9::identity() * 1000.0, // sigma_rate
            10,                           // block_length
            100.0,                        // sigma_t
            Matrix9::identity(),          // l_r (identity for tests)
            100.0,                        // theta_ns
            5000,                         // calibration_samples
            false,                        // discrete_mode
            5.0,                          // mde_shift_ns
            10.0,                         // mde_tail_ns
            snapshot,                     // calibration_snapshot
            1.0,                          // timer_resolution_ns
            100_000.0,                    // samples_per_second
            10.0,                         // c_floor
            18.48,                        // projection_mismatch_thresh
            0.001,                        // theta_tick
            100.0,                        // theta_eff
            0.1,                          // theta_floor_initial
            42,                           // rng_seed
            1,                            // batch_k (no batching in tests)
        )
    }

    fn make_test_posterior(leak_prob: f64) -> Posterior {
        Posterior::new(
            Vector9::zeros(),
            Matrix9::identity(),
            Vector2::new(10.0, 5.0),
            Matrix2::new(1.0, 0.0, 0.0, 1.0),
            Vec::new(), // beta_draws
            leak_prob,
            1.0, // projection_mismatch_q
            1000,
        )
    }

    #[test]
    fn test_adaptive_step_config_default() {
        let config = AdaptiveStepConfig::default();
        assert!((config.pass_threshold - 0.05).abs() < 1e-10);
        assert!((config.fail_threshold - 0.95).abs() < 1e-10);
        assert!((config.time_budget_secs - 30.0).abs() < 1e-10);
    }

    #[test]
    fn test_adaptive_step_config_builder() {
        let config = AdaptiveStepConfig::with_theta(50.0)
            .pass_threshold(0.01)
            .fail_threshold(0.99)
            .time_budget_secs(60.0)
            .max_samples(500_000);

        assert!((config.theta_ns - 50.0).abs() < 1e-10);
        assert!((config.pass_threshold - 0.01).abs() < 1e-10);
        assert!((config.fail_threshold - 0.99).abs() < 1e-10);
        assert!((config.time_budget_secs - 60.0).abs() < 1e-10);
        assert_eq!(config.max_samples, 500_000);
    }

    #[test]
    fn test_adaptive_outcome_accessors() {
        let posterior = make_test_posterior(0.95);

        let outcome = AdaptiveOutcome::LeakDetected {
            posterior: posterior.clone(),
            samples_per_class: 1000,
            elapsed_secs: 1.5,
        };

        assert!(outcome.is_leak_detected());
        assert!(outcome.is_conclusive());
        assert_eq!(outcome.leak_probability(), Some(0.95));
        assert_eq!(outcome.samples_per_class(), 1000);
        assert!((outcome.elapsed_secs() - 1.5).abs() < 1e-10);

        let outcome = AdaptiveOutcome::NoLeakDetected {
            posterior,
            samples_per_class: 2000,
            elapsed_secs: 2.5,
        };

        assert!(!outcome.is_leak_detected());
        assert!(outcome.is_conclusive());
        assert_eq!(outcome.samples_per_class(), 2000);
    }

    #[test]
    fn test_step_result_accessors() {
        let posterior = make_test_posterior(0.5);

        let result = StepResult::Continue {
            posterior: posterior.clone(),
            samples_per_class: 1000,
        };

        assert!(!result.is_decision());
        assert_eq!(result.leak_probability(), Some(0.5));

        let result = StepResult::Decision(AdaptiveOutcome::LeakDetected {
            posterior,
            samples_per_class: 1000,
            elapsed_secs: 1.0,
        });

        assert!(result.is_decision());
    }

    #[test]
    fn test_adaptive_step_insufficient_samples() {
        let calibration = make_test_calibration();
        let mut state = AdaptiveState::new();
        state.add_batch(vec![100; 10], vec![101; 10]); // Only 10 samples

        let config = AdaptiveStepConfig::default();
        let result = adaptive_step(&calibration, &mut state, 1.0, 0.1, &config);

        // Should return Inconclusive (need more samples)
        assert!(result.is_decision());
        if let StepResult::Decision(AdaptiveOutcome::Inconclusive { reason, .. }) = result {
            assert!(matches!(reason, InconclusiveReason::DataTooNoisy { .. }));
        } else {
            panic!("Expected Inconclusive with DataTooNoisy");
        }
    }

    #[test]
    fn test_compute_posterior_no_difference() {
        let calibration = make_test_calibration();
        let mut state = AdaptiveState::new();

        // Add samples with no timing difference
        let baseline: Vec<u64> = (0..1000).map(|i| 1000 + (i % 10)).collect();
        let sample: Vec<u64> = (0..1000).map(|i| 1000 + (i % 10)).collect();
        state.add_batch(baseline, sample);

        let config = AdaptiveStepConfig::with_theta(100.0);

        let posterior = compute_posterior(&state, &calibration, 1.0, &config);

        assert!(posterior.is_some());
        let p = posterior.unwrap();

        // With identical distributions, leak probability should be low
        assert!(
            p.leak_probability < 0.5,
            "Identical distributions should have low leak probability, got {}",
            p.leak_probability
        );
    }

    #[test]
    fn test_compute_posterior_with_difference() {
        let calibration = make_test_calibration();
        let mut state = AdaptiveState::new();

        // Add samples with clear timing difference (200ns)
        let baseline: Vec<u64> = (0..1000).map(|i| 1000 + (i % 10)).collect();
        let sample: Vec<u64> = (0..1000).map(|i| 1200 + (i % 10)).collect();
        state.add_batch(baseline, sample);

        let config = AdaptiveStepConfig::with_theta(100.0); // Effect is 200ns, threshold is 100ns

        let posterior = compute_posterior(&state, &calibration, 1.0, &config);

        assert!(posterior.is_some());
        let p = posterior.unwrap();

        // With 200ns difference vs 100ns threshold, leak probability should be high
        assert!(
            p.leak_probability > 0.5,
            "Clear difference should have high leak probability, got {}",
            p.leak_probability
        );
    }

    #[test]
    fn test_adaptive_step_detects_leak() {
        let calibration = make_test_calibration();
        let mut state = AdaptiveState::new();

        // Add samples with large timing difference
        let baseline: Vec<u64> = (0..1000).map(|i| 1000 + (i % 10)).collect();
        let sample: Vec<u64> = (0..1000).map(|i| 1500 + (i % 10)).collect();
        state.add_batch(baseline, sample);

        let config = AdaptiveStepConfig::with_theta(100.0);
        let result = adaptive_step(&calibration, &mut state, 1.0, 1.0, &config);

        // Should detect a leak
        assert!(result.is_decision());
        if let StepResult::Decision(outcome) = result {
            assert!(outcome.is_leak_detected());
        } else {
            panic!("Expected Decision");
        }
    }
}