rchd 2.1.0

//! Build history tracking.
//!
//! Maintains a ring buffer of recent builds for status reporting and analytics.

use chrono::{DateTime, Duration as ChronoDuration, Utc};
use rch_common::{
    BuildCancellationMetadata, BuildHeartbeatPhase, BuildHeartbeatRequest, BuildLocation,
    BuildRecord, BuildStats, CommandTimingBreakdown, SavedTimeStats,
};
use serde::{Deserialize, Serialize};
use std::collections::{HashMap, HashSet, VecDeque};
use std::fs::File;
use std::io::{BufRead, BufReader, Write};
use std::path::{Path, PathBuf};
use std::sync::RwLock;
use std::sync::atomic::{AtomicU64, Ordering};
use std::time::Instant;
use tokio::fs::OpenOptions as AsyncOpenOptions;
use tokio::io::AsyncWriteExt;
use tracing::{debug, warn};

#[derive(Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
struct DurableOwnership {
    version: u32,
    active: Vec<ActiveBuildState>,
    completed: Vec<TerminalOwnership>,
    #[serde(default)]
    cancelled_wrappers: HashSet<String>,
    #[serde(default)]
    next_queue_id: Option<u64>,
}

/// Cancellation and admission compete under the same ownership lock.
pub enum WrapperCancellation {
    BeforeStart,
    Active(u64),
    Completed(Box<BuildRecord>),
    NotQueued,
}

const MAX_CANCELLED_WRAPPERS: usize = 100_000;
/// Queue IDs occupy a separate numeric namespace from daemon build IDs.
const QUEUE_ID_NAMESPACE: u64 = 1 << 63;

#[derive(Clone, Serialize, Deserialize)]
struct TerminalOwnership {
    record: BuildRecord,
    local_wrapper_id: Option<String>,
}

/// Terminal result supplied by completion or cancellation; ownership stays separate.
pub struct BuildCompletion {
    pub exit_code: i32,
    pub duration_ms: Option<u64>,
    pub bytes_transferred: Option<u64>,
    pub timing: Option<CommandTimingBreakdown>,
    pub cancellation: Option<BuildCancellationMetadata>,
}

/// Linux boot identity plus process start ticks distinguish PID reuse and reboot.
pub fn process_identity(pid: u32) -> Option<String> {
    if pid <= 1 {
        return None;
    }
    let boot = std::fs::read_to_string("/proc/sys/kernel/random/boot_id").ok()?;
    let stat = std::fs::read_to_string(format!("/proc/{pid}/stat")).ok()?;
    let (_, fields) = stat.rsplit_once(") ")?;
    let start_ticks = fields.split_whitespace().nth(19)?;
    Some(format!("{}:{start_ticks}", boot.trim()))
}
/// Default maximum number of builds to retain.
const DEFAULT_CAPACITY: usize = 100;

/// Whether the recorded command completed successfully.
///
/// History carries a terminal exit code, not compilation-stage evidence.
/// In particular, Cargo's exit 101 can mean compilation or test failure, and
/// a signal can interrupt either stage. Never infer successful compilation
/// from the command kind. These counts follow the `BuildStats` exit-code
/// contract; stage-specific success requires separate evidence.
fn build_record_succeeded(record: &BuildRecord) -> bool {
    record.exit_code == 0
}

/// In-flight build state tracked for active build visibility.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ActiveBuildState {
    pub id: u64,
    pub project_id: String,
    pub worker_id: String,
    pub command: String,
    pub started_at: String,
    #[serde(skip, default = "Instant::now")]
    pub started_at_mono: Instant,
    pub hook_pid: u32,
    #[serde(default)]
    pub hook_process_identity: Option<String>,
    /// Client wrapper identity, allowing the daemon's active build to be
    /// joined with the client-side durable lease without relying on a reusable PID.
    pub local_wrapper_id: Option<String>,
    pub remote_pgid_file: Option<String>,
    pub slots: u32,
    pub location: BuildLocation,
    pub heartbeat_phase: BuildHeartbeatPhase,
    pub heartbeat_detail: Option<String>,
    pub heartbeat_counter: u64,
    pub heartbeat_percent: Option<f64>,
    pub heartbeat_count: u64,
    pub last_heartbeat_at: String,
    #[serde(skip, default = "Instant::now")]
    pub last_heartbeat_mono: Instant,
    pub last_progress_at: String,
    #[serde(skip, default = "Instant::now")]
    pub last_progress_mono: Instant,
    pub detector_hook_alive: bool,
    pub detector_heartbeat_stale: bool,
    pub detector_progress_stale: bool,
    pub detector_confidence: f64,
    pub detector_build_age_secs: u64,
    pub detector_slots_owned: u32,
    pub detector_last_evaluated_at: Option<String>,
    /// Restarted ownership must not authorize signalling a reused local PID.
    #[serde(skip)]
    pub recovered: bool,
}

/// Snapshot of stuck-detector evidence for an active build.
#[derive(Debug, Clone, Copy)]
pub struct StuckDetectorSnapshot {
    pub hook_alive: bool,
    pub heartbeat_stale: bool,
    pub progress_stale: bool,
    pub confidence: f64,
    pub build_age_secs: u64,
    pub slots_owned: u32,
}

/// Queued build state for builds waiting for available workers.
///
/// When all workers are busy and `queue_when_busy` is enabled,
/// builds are queued here instead of falling back to local execution.
#[derive(Debug, Clone)]
pub struct QueuedBuildState {
    /// Queue position ID (monotonically increasing).
    pub id: u64,
    /// Project identifier (hash or path).
    pub project_id: String,
    /// Command to execute.
    pub command: String,
    /// When the build was queued (ISO 8601).
    pub queued_at: String,
    /// Monotonic timestamp for duration calculations.
    pub queued_at_mono: Instant,
    /// Hook process ID (for cancellation).
    pub hook_pid: u32,
    pub local_wrapper_id: Option<String>,
    /// Number of slots needed.
    pub slots_needed: u32,
    /// Estimated start time (ISO 8601), updated as queue advances.
    pub estimated_start: Option<String>,
}

/// Build history manager.
///
/// Thread-safe ring buffer of recent builds with optional persistence.
pub struct BuildHistory {
    /// Ring buffer of recent builds.
    records: RwLock<VecDeque<BuildRecord>>,
    /// Active builds (in-flight).
    active: RwLock<HashMap<u64, ActiveBuildState>>,
    /// Queued builds (waiting for workers).
    queued: RwLock<VecDeque<QueuedBuildState>>,
    /// Maximum capacity for history.
    capacity: usize,
    /// Maximum queue depth (0 = unlimited).
    max_queue_depth: usize,
    /// Next build ID.
    next_id: AtomicU64,
    /// Next queue ID.
    next_queue_id: AtomicU64,
    /// Persistence path (optional).
    persistence_path: Option<PathBuf>,
    /// Terminal receipts share the atomic ownership commit, not the JSONL log.
    terminal: RwLock<HashMap<u64, TerminalOwnership>>,
    /// Never evict an intent while a delayed same-identity admission can arrive.
    cancelled_wrappers: RwLock<HashSet<String>>,
    ownership_failed: std::sync::atomic::AtomicBool,
    #[cfg(test)]
    fail_after_ownership_rename: std::sync::atomic::AtomicBool,
}

/// Default maximum queue depth.
const DEFAULT_MAX_QUEUE_DEPTH: usize = 100;

impl BuildHistory {
    /// Create a new build history with the given capacity.
    pub fn new(capacity: usize) -> Self {
        // Use a timestamp-based epoch for build IDs to prevent collisions
        // with orphaned remote processes if the daemon crashes and restarts.
        let epoch_secs = std::time::SystemTime::now()
            .duration_since(std::time::UNIX_EPOCH)
            .unwrap_or_default()
            .as_secs();
        let initial_id = (epoch_secs << 24) | 1;

        Self {
            records: RwLock::new(VecDeque::with_capacity(capacity)),
            active: RwLock::new(HashMap::new()),
            queued: RwLock::new(VecDeque::new()),
            capacity,
            max_queue_depth: DEFAULT_MAX_QUEUE_DEPTH,
            next_id: AtomicU64::new(initial_id),
            next_queue_id: AtomicU64::new(QUEUE_ID_NAMESPACE | initial_id),
            persistence_path: None,
            terminal: RwLock::new(HashMap::new()),
            cancelled_wrappers: RwLock::new(HashSet::new()),
            ownership_failed: std::sync::atomic::AtomicBool::new(false),
            #[cfg(test)]
            fail_after_ownership_rename: std::sync::atomic::AtomicBool::new(false),
        }
    }

    /// Create a new build history with default capacity.
    pub fn with_default_capacity() -> Self {
        Self::new(DEFAULT_CAPACITY)
    }

    /// Set the maximum queue depth.
    pub fn with_max_queue_depth(mut self, depth: usize) -> Self {
        self.max_queue_depth = depth;
        self
    }

    /// Enable persistence to the given path.
    pub fn with_persistence(mut self, path: PathBuf) -> Self {
        self.persistence_path = Some(path);
        self
    }
    pub fn ownership_failed(&self) -> bool {
        self.ownership_failed.load(Ordering::SeqCst)
    }

    /// Get the next build ID.
    pub fn next_id(&self) -> u64 {
        self.next_id.fetch_add(1, Ordering::SeqCst)
    }

    /// Record a completed build.
    ///
    /// Returns a handle to the persistence task if persistence is enabled.
    pub fn record(&self, record: BuildRecord) -> Option<tokio::task::JoinHandle<()>> {
        debug!(
            "Recording build {}: {} ({:?}, {} ms)",
            record.id, record.command, record.location, record.duration_ms
        );

        // Prepare for persistence before locking
        let persistence_task = self
            .persistence_path
            .as_ref()
            .map(|path| (path.clone(), record.clone()));

        // Update memory state under lock
        {
            let mut records = self.records.write().unwrap_or_else(|e| e.into_inner());

            // Evict oldest if at capacity
            if records.len() >= self.capacity {
                records.pop_front();
            }

            records.push_back(record);
        }

        // Persist asynchronously (fire and forget or awaitable)
        if let Some((path, record)) = persistence_task {
            Some(tokio::spawn(async move {
                if let Err(e) = Self::persist_record_async(&path, &record).await {
                    warn!("Failed to persist build record: {}", e);
                }
            }))
        } else {
            None
        }
    }

    /// Register a new active build and return its state.
    pub fn start_active_build(
        &self,
        project_id: String,
        worker_id: String,
        command: String,
        hook_pid: u32,
        slots: u32,
        location: BuildLocation,
    ) -> ActiveBuildState {
        self.start_active_build_with_wrapper(
            project_id, worker_id, command, hook_pid, None, slots, location,
        )
    }

    /// Register an active build while preserving the client durable-lease id.
    #[allow(clippy::too_many_arguments)] // Mirrors the existing explicit active-build registration contract plus correlation id.
    pub fn start_active_build_with_wrapper(
        &self,
        project_id: String,
        worker_id: String,
        command: String,
        hook_pid: u32,
        local_wrapper_id: Option<String>,
        slots: u32,
        location: BuildLocation,
    ) -> ActiveBuildState {
        let id = self.next_id();
        let started_at = Utc::now().to_rfc3339();
        let started_at_mono = Instant::now();
        let state = ActiveBuildState {
            id,
            project_id,
            worker_id,
            command,
            started_at: started_at.clone(),
            hook_process_identity: process_identity(hook_pid),
            started_at_mono,
            hook_pid,
            local_wrapper_id,
            remote_pgid_file: None,
            slots,
            location,
            heartbeat_phase: BuildHeartbeatPhase::SyncUp,
            heartbeat_detail: Some("build_started".to_string()),
            heartbeat_counter: 0,
            heartbeat_percent: None,
            heartbeat_count: 0,
            last_heartbeat_at: started_at.clone(),
            last_heartbeat_mono: started_at_mono,
            last_progress_at: started_at,
            last_progress_mono: started_at_mono,
            detector_hook_alive: true,
            detector_heartbeat_stale: false,
            detector_progress_stale: false,
            detector_confidence: 0.0,
            detector_build_age_secs: 0,
            detector_slots_owned: slots,
            detector_last_evaluated_at: None,
            recovered: false,
        };

        let mut active = self.active.write().unwrap_or_else(|e| e.into_inner());
        active.insert(id, state.clone());
        self.persist_ownership(&active, None)
            .expect("persist active build before exposing ownership");
        state
    }

    /// Try to register a new active build, failing if the same worker already
    /// has a live build for the same project.
    pub fn try_start_active_build(
        &self,
        project_id: String,
        worker_id: String,
        command: String,
        hook_pid: u32,
        slots: u32,
        location: BuildLocation,
    ) -> Option<ActiveBuildState> {
        self.try_start_active_build_with_wrapper(
            project_id, worker_id, command, hook_pid, None, slots, location,
        )
        .expect("persist active build before exposing ownership")
    }

    /// Try to register an active build with a client durable-lease id.
    #[allow(clippy::too_many_arguments)] // Mirrors the existing explicit active-build registration contract plus correlation id.
    pub fn try_start_active_build_with_wrapper(
        &self,
        project_id: String,
        worker_id: String,
        command: String,
        hook_pid: u32,
        local_wrapper_id: Option<String>,
        slots: u32,
        location: BuildLocation,
    ) -> std::io::Result<Option<ActiveBuildState>> {
        let id = self.next_id();
        if id >= QUEUE_ID_NAMESPACE {
            return Err(std::io::Error::other(
                "active build identifier namespace exhausted",
            ));
        }
        let started_at = Utc::now().to_rfc3339();
        let started_at_mono = Instant::now();
        let state = ActiveBuildState {
            id,
            project_id,
            worker_id,
            command,
            started_at: started_at.clone(),
            hook_process_identity: process_identity(hook_pid),
            started_at_mono,
            hook_pid,
            local_wrapper_id,
            remote_pgid_file: None,
            slots,
            location,
            heartbeat_phase: BuildHeartbeatPhase::SyncUp,
            heartbeat_detail: Some("build_started".to_string()),
            heartbeat_counter: 0,
            heartbeat_percent: None,
            heartbeat_count: 0,
            last_heartbeat_at: started_at.clone(),
            last_heartbeat_mono: started_at_mono,
            last_progress_at: started_at,
            last_progress_mono: started_at_mono,
            detector_hook_alive: true,
            detector_heartbeat_stale: false,
            detector_progress_stale: false,
            detector_confidence: 0.0,
            detector_build_age_secs: 0,
            detector_slots_owned: slots,
            detector_last_evaluated_at: None,
            recovered: false,
        };

        let mut active = self.active.write().unwrap_or_else(|e| e.into_inner());
        if self.ownership_failed() {
            return Ok(None);
        }
        if state
            .local_wrapper_id
            .as_deref()
            .is_some_and(|id| self.wrapper_cancelled(id))
        {
            return Ok(None);
        }
        if active.values().any(|existing| {
            existing.project_id == state.project_id && existing.worker_id == state.worker_id
        }) {
            return Ok(None);
        }
        active.insert(id, state.clone());
        self.persist_ownership(&active, None)?;
        Ok(Some(state))
    }

    /// Record a heartbeat/progress update for an active build.
    ///
    /// Returns the updated active state if the build exists.
    pub fn record_build_heartbeat(
        &self,
        heartbeat: BuildHeartbeatRequest,
    ) -> Option<ActiveBuildState> {
        let now = Instant::now();
        let now_rfc3339 = Utc::now().to_rfc3339();

        let mut active = self.active.write().unwrap_or_else(|e| e.into_inner());
        let state = active.get_mut(&heartbeat.build_id)?;

        // Ignore worker mismatch updates to avoid cross-build contamination.
        if state.worker_id != heartbeat.worker_id.as_str() {
            return None;
        }

        // A durable identity can never be replaced or omitted when reattaching.
        if state.local_wrapper_id.is_some() && state.local_wrapper_id != heartbeat.local_wrapper_id
        {
            return None;
        }
        if state.recovered && state.local_wrapper_id.is_none() {
            return None;
        }
        if heartbeat.remote_pgid_file.as_ref().is_some_and(|path| {
            state
                .remote_pgid_file
                .as_ref()
                .is_some_and(|recorded| recorded != path)
        }) {
            return None;
        }
        // A heartbeat describes the original wrapper; it is not a process
        // handoff. A delayed message can outlive that process and its PID can
        // already belong to somebody else. Never capture the new occupant's
        // identity or replace the recorded PID from the message.
        if heartbeat
            .hook_pid
            .filter(|pid| *pid > 0)
            .is_some_and(|pid| pid != state.hook_pid)
        {
            return None;
        }
        if let Some(identity) = state.hook_process_identity.as_ref()
            && process_identity(state.hook_pid).as_ref() != Some(identity)
        {
            return None;
        }
        if state.recovered && state.hook_process_identity.is_none() {
            return None;
        }
        if let Some(local_wrapper_id) = heartbeat.local_wrapper_id {
            state.local_wrapper_id = Some(local_wrapper_id);
        }
        if let Some(remote_pgid_file) = heartbeat
            .remote_pgid_file
            .filter(|path| !path.trim().is_empty())
        {
            state.remote_pgid_file = Some(remote_pgid_file);
        }

        let previous_phase = state.heartbeat_phase.clone();
        let previous_counter = state.heartbeat_counter;
        let previous_percent = state.heartbeat_percent;
        let previous_detail = state.heartbeat_detail.clone();

        state.heartbeat_phase = heartbeat.phase;
        state.heartbeat_detail = heartbeat.detail;
        if let Some(counter) = heartbeat.progress_counter {
            state.heartbeat_counter = state.heartbeat_counter.max(counter);
        }
        if let Some(percent) = heartbeat.progress_percent {
            // `f64::clamp` returns NaN when given NaN — so a worker sending
            // `NaN` would poison the heartbeat state, making every
            // subsequent percent comparison `false` (NaN never compares >)
            // and leaving status displays to render "NaN%". Drop invalid
            // values silently; a progress report with no number is better
            // than a stuck progress bar for the rest of the build.
            if percent.is_finite() {
                state.heartbeat_percent = Some(percent.clamp(0.0, 100.0));
            }
        }
        state.heartbeat_count = state.heartbeat_count.saturating_add(1);
        state.last_heartbeat_at = now_rfc3339.clone();
        state.last_heartbeat_mono = now;

        // Progress evidence can come from phase transitions, increasing counters,
        // percent improvements, or detail updates.
        let counter_progressed = state.heartbeat_counter > previous_counter;
        let percent_progressed = match (previous_percent, state.heartbeat_percent) {
            (Some(before), Some(after)) => after > before + f64::EPSILON,
            (None, Some(_)) => true,
            _ => false,
        };
        let detail_changed = state.heartbeat_detail != previous_detail;
        let phase_changed = state.heartbeat_phase != previous_phase;

        if counter_progressed || percent_progressed || detail_changed || phase_changed {
            state.last_progress_at = now_rfc3339;
            state.last_progress_mono = now;
        }

        let updated = state.clone();
        if let Err(error) = self.persist_ownership(&active, None) {
            warn!("Unable to persist build heartbeat: {error}");
            return None;
        }
        Some(updated)
    }

    /// Record the latest stuck-detector evidence snapshot for an active build.
    pub fn record_stuck_detector_snapshot(
        &self,
        build_id: u64,
        snapshot: StuckDetectorSnapshot,
    ) -> Option<ActiveBuildState> {
        let now_rfc3339 = Utc::now().to_rfc3339();
        let mut active = self.active.write().unwrap_or_else(|e| e.into_inner());
        let state = active.get_mut(&build_id)?;

        state.detector_hook_alive = snapshot.hook_alive;
        state.detector_heartbeat_stale = snapshot.heartbeat_stale;
        state.detector_progress_stale = snapshot.progress_stale;
        state.detector_confidence = snapshot.confidence.clamp(0.0, 1.0);
        state.detector_build_age_secs = snapshot.build_age_secs;
        state.detector_slots_owned = snapshot.slots_owned;
        state.detector_last_evaluated_at = Some(now_rfc3339);

        Some(state.clone())
    }

    /// Complete an active build, moving it into history.
    pub fn finish_active_build(
        &self,
        build_id: u64,
        exit_code: i32,
        duration_ms: Option<u64>,
        bytes_transferred: Option<u64>,
        timing: Option<CommandTimingBreakdown>,
    ) -> Option<BuildRecord> {
        let state = self.active_build(build_id)?;
        self.complete_durable(
            build_id,
            &state.worker_id,
            state.local_wrapper_id.as_deref(),
            BuildCompletion {
                exit_code,
                duration_ms,
                bytes_transferred,
                timing,
                cancellation: None,
            },
        )
        .expect("persist terminal ownership before release")
        .map(|(_, record)| record)
    }

    /// Cancel an active build, moving it into history with a cancel exit code.
    pub fn cancel_active_build(
        &self,
        build_id: u64,
        bytes_transferred: Option<u64>,
        cancellation: Option<BuildCancellationMetadata>,
    ) -> Option<BuildRecord> {
        let state = self.active_build(build_id)?;
        self.complete_durable(
            build_id,
            &state.worker_id,
            state.local_wrapper_id.as_deref(),
            BuildCompletion {
                exit_code: 130,
                duration_ms: None,
                bytes_transferred,
                timing: None,
                cancellation,
            },
        )
        .expect("persist terminal ownership before release")
        .map(|(_, record)| record)
    }

    /// Get a specific active build by ID.
    pub fn active_build(&self, build_id: u64) -> Option<ActiveBuildState> {
        self.active
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .get(&build_id)
            .cloned()
    }

    /// Get all active builds.
    pub fn active_builds(&self) -> Vec<ActiveBuildState> {
        self.active
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .values()
            .cloned()
            .collect()
    }

    /// Check whether a worker already has an active build for the same project.
    pub fn has_active_build_for_project_on_worker(
        &self,
        project_id: &str,
        worker_id: &str,
    ) -> bool {
        self.active
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .values()
            .any(|state| state.project_id == project_id && state.worker_id == worker_id)
    }

    /// Return worker IDs currently running an active build for the given project.
    pub fn active_workers_for_project(&self, project_id: &str) -> HashSet<String> {
        self.active
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .values()
            .filter(|state| state.project_id == project_id)
            .map(|state| state.worker_id.clone())
            .collect()
    }

    // =========================================================================
    // Queue Management
    // =========================================================================

    pub fn wrapper_cancelled(&self, wrapper: &str) -> bool {
        !self.ownership_failed()
            && self
                .cancelled_wrappers
                .read()
                .unwrap_or_else(|e| e.into_inner())
                .contains(wrapper)
    }

    /// Persist cancellation before removing queue visibility or acknowledging it.
    /// A bounded full journal refuses new cancellation rather than forgetting an
    /// identity that might still be delivered by a disconnected wrapper.
    pub fn cancel_wrapper(&self, wrapper: &str) -> std::io::Result<WrapperCancellation> {
        let active = self.active.write().unwrap_or_else(|e| e.into_inner());
        if self.ownership_failed() {
            return Err(std::io::Error::other(
                "durable ownership uncertain; restart required",
            ));
        }
        if let Some(state) = active
            .values()
            .find(|state| state.local_wrapper_id.as_deref() == Some(wrapper))
        {
            return Ok(WrapperCancellation::Active(state.id));
        }
        if let Some(receipt) = self
            .terminal
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .values()
            .find(|receipt| receipt.local_wrapper_id.as_deref() == Some(wrapper))
        {
            return Ok(WrapperCancellation::Completed(Box::new(
                receipt.record.clone(),
            )));
        }
        if !self.wrapper_cancelled(wrapper)
            && !self
                .queued
                .read()
                .unwrap_or_else(|e| e.into_inner())
                .iter()
                .any(|state| state.local_wrapper_id.as_deref() == Some(wrapper))
        {
            return Ok(WrapperCancellation::NotQueued);
        }
        {
            let mut cancelled = self
                .cancelled_wrappers
                .write()
                .unwrap_or_else(|e| e.into_inner());
            if !cancelled.contains(wrapper) && cancelled.len() >= MAX_CANCELLED_WRAPPERS {
                return Err(std::io::Error::other(
                    "queued cancellation journal is full; intent not accepted",
                ));
            }
            cancelled.insert(wrapper.to_owned());
        }
        self.persist_ownership(&active, None)?;
        self.queued
            .write()
            .unwrap_or_else(|e| e.into_inner())
            .retain(|state| state.local_wrapper_id.as_deref() != Some(wrapper));
        Ok(WrapperCancellation::BeforeStart)
    }

    /// Decide queue departure against cancellation while admission is locked.
    /// Once departure wins, cancellation cannot claim a no-start receipt for a
    /// wrapper that may already be handling timeout or local fallback.
    pub fn finish_queued_build(
        &self,
        queue_id: u64,
        wrapper: Option<&str>,
    ) -> std::io::Result<bool> {
        let _active = self.active.write().unwrap_or_else(|e| e.into_inner());
        if self.ownership_failed() {
            return Err(std::io::Error::other(
                "durable ownership uncertain; restart required",
            ));
        }
        let cancelled = wrapper.is_some_and(|id| self.wrapper_cancelled(id));
        self.remove_queued_build(queue_id);
        Ok(cancelled)
    }

    /// Allocate without wrapping into a previously used identity.
    fn next_queue_id(&self) -> Option<u64> {
        self.next_queue_id
            .try_update(Ordering::SeqCst, Ordering::SeqCst, |id| {
                if id < QUEUE_ID_NAMESPACE {
                    None
                } else {
                    id.checked_add(1)
                }
            })
            .ok()
    }

    /// Enqueue a build waiting for an available worker.
    ///
    /// Returns `None` if the queue is full (when max_queue_depth > 0).
    pub fn enqueue_build(
        &self,
        project_id: String,
        command: String,
        hook_pid: u32,
        slots_needed: u32,
        local_wrapper_id: Option<String>,
    ) -> Option<QueuedBuildState> {
        let active = self.active.write().unwrap_or_else(|e| e.into_inner());
        if self.ownership_failed()
            || local_wrapper_id
                .as_deref()
                .is_some_and(|id| self.wrapper_cancelled(id))
        {
            return None;
        }
        let mut queue = self.queued.write().unwrap_or_else(|e| e.into_inner());

        // Check queue depth limit
        if self.max_queue_depth > 0 && queue.len() >= self.max_queue_depth {
            debug!(
                "Queue full ({}/{}), rejecting build for {}",
                queue.len(),
                self.max_queue_depth,
                project_id
            );
            return None;
        }

        let id = self.next_queue_id()?;
        // A crash may waste an ID, but cannot expose an ID before its high-water
        // mark is durable and later cancel a different wrapper after restart.
        self.persist_ownership(&active, None).ok()?;
        let queued_at = Utc::now().to_rfc3339();
        let state = QueuedBuildState {
            id,
            project_id,
            command,
            queued_at,
            queued_at_mono: Instant::now(),
            hook_pid,
            local_wrapper_id,
            slots_needed,
            estimated_start: None,
        };

        queue.push_back(state.clone());
        debug!(
            "Build queued: id={}, position={}, project={}",
            id,
            queue.len(),
            state.project_id
        );

        Some(state)
    }

    /// Dequeue the next build (FIFO).
    ///
    /// Called when a worker becomes available.
    pub fn dequeue_build(&self) -> Option<QueuedBuildState> {
        let mut queue = self.queued.write().unwrap_or_else(|e| e.into_inner());
        let state = queue.pop_front()?;
        debug!(
            "Build dequeued: id={}, waited {:?}, project={}",
            state.id,
            state.queued_at_mono.elapsed(),
            state.project_id
        );
        Some(state)
    }

    /// Remove a specific queued build by ID (e.g., for cancellation).
    pub fn remove_queued_build(&self, queue_id: u64) -> Option<QueuedBuildState> {
        let mut queue = self.queued.write().unwrap_or_else(|e| e.into_inner());
        let pos = queue.iter().position(|b| b.id == queue_id)?;
        queue.remove(pos)
    }

    /// Remove a queued build by hook PID.
    pub fn remove_queued_build_by_pid(&self, hook_pid: u32) -> Option<QueuedBuildState> {
        let mut queue = self.queued.write().unwrap_or_else(|e| e.into_inner());
        let pos = queue.iter().position(|b| b.hook_pid == hook_pid)?;
        queue.remove(pos)
    }

    /// Get all queued builds (in queue order).
    pub fn queued_builds(&self) -> Vec<QueuedBuildState> {
        self.queued
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .iter()
            .cloned()
            .collect()
    }

    /// Get a specific queued build by ID.
    pub fn queued_build(&self, queue_id: u64) -> Option<QueuedBuildState> {
        // Use the same poisoned-lock recovery pattern as the rest of this
        // file: a prior panic under this lock shouldn't propagate a panic
        // into a simple read-only query.
        self.queued
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .iter()
            .find(|b| b.id == queue_id)
            .cloned()
    }

    /// Get the queue position of a build (1-indexed, None if not found).
    pub fn queue_position(&self, queue_id: u64) -> Option<usize> {
        self.queued
            .read()
            .unwrap()
            .iter()
            .position(|b| b.id == queue_id)
            .map(|p| p + 1)
    }

    /// Get the current queue depth.
    pub fn queue_depth(&self) -> usize {
        self.queued.read().unwrap_or_else(|e| e.into_inner()).len()
    }

    /// Check if the queue is empty.
    pub fn queue_is_empty(&self) -> bool {
        self.queued
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .is_empty()
    }

    /// Update estimated start times for all queued builds.
    ///
    /// Uses average build duration from history and active build state.
    pub fn update_queue_estimates(&self) {
        let avg_duration = self.stats().avg_duration_ms;
        let active_count = self.active.read().unwrap_or_else(|e| e.into_inner()).len();

        let mut queue = self.queued.write().unwrap_or_else(|e| e.into_inner());

        // Estimate when each queued build will start
        let now = Utc::now();
        for (i, build) in queue.iter_mut().enumerate() {
            // Simple estimate: position * avg_duration, adjusted for active builds
            let position = i + 1;
            let wait_ms = if active_count > 0 {
                // Assume active builds are half-done on average
                let remaining_active_ms = (avg_duration / 2) as i64;
                let queue_wait_ms = (position as u64 * avg_duration) as i64;
                remaining_active_ms + queue_wait_ms
            } else {
                0 // No active builds, next in queue starts immediately
            };

            let estimated = now + chrono::Duration::milliseconds(wait_ms.max(0));
            build.estimated_start = Some(estimated.to_rfc3339());
        }
    }

    /// Get recent builds (most recent first).
    pub fn recent(&self, limit: usize) -> Vec<BuildRecord> {
        let records = self.records.read().unwrap_or_else(|e| e.into_inner());
        records.iter().rev().take(limit).cloned().collect()
    }

    /// Get all builds (most recent first).
    pub fn all(&self) -> Vec<BuildRecord> {
        let records = self.records.read().unwrap_or_else(|e| e.into_inner());
        records.iter().rev().cloned().collect()
    }

    /// Get builds by worker (most recent first).
    pub fn by_worker(&self, worker_id: &str, limit: usize) -> Vec<BuildRecord> {
        let records = self.records.read().unwrap_or_else(|e| e.into_inner());
        records
            .iter()
            .rev()
            .filter(|r| r.worker_id.as_deref() == Some(worker_id))
            .take(limit)
            .cloned()
            .collect()
    }

    /// Get builds by project (most recent first).
    pub fn by_project(&self, project_id: &str, limit: usize) -> Vec<BuildRecord> {
        let records = self.records.read().unwrap_or_else(|e| e.into_inner());
        records
            .iter()
            .rev()
            .filter(|r| r.project_id == project_id)
            .take(limit)
            .cloned()
            .collect()
    }

    /// Get aggregate statistics.
    pub fn stats(&self) -> BuildStats {
        let records = self.records.read().unwrap_or_else(|e| e.into_inner());
        let total = records.len();

        if total == 0 {
            return BuildStats::default();
        }

        let successes = records.iter().filter(|r| build_record_succeeded(r)).count();
        let remote = records
            .iter()
            .filter(|r| r.location == BuildLocation::Remote)
            .count();
        let total_duration: u64 = records.iter().map(|r| r.duration_ms).sum();
        let avg_duration = total_duration / total as u64;

        BuildStats {
            total_builds: total,
            success_count: successes,
            failure_count: total - successes,
            remote_count: remote,
            local_count: total - remote,
            avg_duration_ms: avg_duration,
        }
    }

    /// Calculate saved time statistics from remote builds.
    ///
    /// Estimates what remote builds would have taken locally ONLY from an
    /// observed baseline of successful local builds (`estimate_basis` =
    /// "observed_local_mean"). Without such a baseline no savings are claimed:
    /// the stats report measured remote durations with `estimate_basis` =
    /// "none" and zero estimated savings.
    pub fn saved_time_stats(&self) -> SavedTimeStats {
        let records = self.records.read().unwrap_or_else(|e| e.into_inner());
        let now = Utc::now();
        let today_start = now.date_naive().and_hms_opt(0, 0, 0).unwrap();
        let week_start = today_start - ChronoDuration::days(7);

        // Separate local and remote builds
        let local_builds: Vec<_> = records
            .iter()
            .filter(|r| r.location == BuildLocation::Local && r.exit_code == 0)
            .collect();
        let remote_builds: Vec<_> = records
            .iter()
            .filter(|r| r.location == BuildLocation::Remote && r.exit_code == 0)
            .collect();

        if remote_builds.is_empty() {
            return SavedTimeStats::default();
        }

        // Calculate average local build duration (if we have local builds)
        let avg_local_duration_ms = if !local_builds.is_empty() {
            let total_local: u64 = local_builds.iter().map(|r| r.duration_ms).sum();
            total_local / local_builds.len() as u64
        } else {
            0
        };

        // Calculate totals and time saved
        let mut total_remote_duration_ms: u64 = 0;
        let mut estimated_local_duration_ms: u64 = 0;
        let mut today_remote_ms: u64 = 0;
        let mut today_estimated_local_ms: u64 = 0;
        let mut week_remote_ms: u64 = 0;
        let mut week_estimated_local_ms: u64 = 0;

        for build in &remote_builds {
            let remote_ms = build.duration_ms;
            total_remote_duration_ms += remote_ms;

            // Estimate local duration only from the observed local mean. No
            // invented per-build scaling: without a baseline the estimate is
            // zero and the stats say so via estimate_basis = "none".
            let estimated_local_ms = if avg_local_duration_ms > 0 {
                avg_local_duration_ms
            } else {
                0
            };
            estimated_local_duration_ms += estimated_local_ms;

            // Parse timestamp for daily/weekly aggregation
            if let Ok(completed) = DateTime::parse_from_rfc3339(&build.completed_at) {
                let completed_naive = completed.naive_utc();
                if completed_naive >= today_start.and_utc().naive_utc() {
                    today_remote_ms += remote_ms;
                    today_estimated_local_ms += estimated_local_ms;
                }
                if completed_naive >= week_start.and_utc().naive_utc() {
                    week_remote_ms += remote_ms;
                    week_estimated_local_ms += estimated_local_ms;
                }
            }
        }

        let time_saved_ms = estimated_local_duration_ms.saturating_sub(total_remote_duration_ms);
        let today_saved_ms = today_estimated_local_ms.saturating_sub(today_remote_ms);
        let week_saved_ms = week_estimated_local_ms.saturating_sub(week_remote_ms);

        let avg_speedup = if total_remote_duration_ms > 0 {
            estimated_local_duration_ms as f64 / total_remote_duration_ms as f64
        } else {
            0.0
        };

        let (estimate_basis, local_baseline_builds) = if avg_local_duration_ms > 0 {
            ("observed_local_mean", local_builds.len())
        } else {
            ("none", 0)
        };

        SavedTimeStats {
            total_remote_duration_ms,
            estimated_local_duration_ms,
            time_saved_ms,
            builds_counted: remote_builds.len(),
            avg_speedup,
            today_saved_ms,
            week_saved_ms,
            estimate_basis: estimate_basis.to_string(),
            local_baseline_builds,
        }
    }

    /// Get the number of builds in history.
    pub fn len(&self) -> usize {
        self.records.read().unwrap_or_else(|e| e.into_inner()).len()
    }

    /// Check if history is empty.
    pub fn is_empty(&self) -> bool {
        self.records
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .is_empty()
    }

    /// Clear all build records.
    #[allow(dead_code)] // May be used for testing or admin operations
    pub fn clear(&self) {
        let mut records = self.records.write().unwrap_or_else(|e| e.into_inner());
        records.clear();
    }

    /// Load history from a JSONL file.
    pub fn load_from_file(path: &Path, capacity: usize) -> std::io::Result<Self> {
        let file = match File::open(path) {
            Ok(file) => Some(file),
            Err(error) if error.kind() == std::io::ErrorKind::NotFound => None,
            Err(error) => return Err(error),
        };

        let mut records = VecDeque::with_capacity(capacity);
        let mut max_id = 0u64;

        for line in file
            .into_iter()
            .flat_map(|file| BufReader::new(file).lines())
        {
            let line = line?;
            if line.trim().is_empty() {
                continue;
            }

            match serde_json::from_str::<BuildRecord>(&line) {
                Ok(record) => {
                    max_id = max_id.max(record.id);
                    if records.len() >= capacity {
                        records.pop_front();
                    }
                    records.push_back(record);
                }
                Err(e) => {
                    warn!("Skipping invalid history line: {}", e);
                }
            }
        }

        let ownership_path = path.with_extension("ownership.json");
        let mut active = HashMap::new();
        let mut terminal = HashMap::new();
        let mut cancelled_wrappers = HashSet::new();
        let mut next_queue_id = None;
        let mut ownership_existed = false;
        match File::open(&ownership_path) {
            Ok(file) => {
                ownership_existed = true;
                let snapshot: DurableOwnership = serde_json::from_reader(file)?;
                if snapshot.version != 1 {
                    return Err(std::io::Error::new(
                        std::io::ErrorKind::InvalidData,
                        "unsupported ownership version",
                    ));
                }
                cancelled_wrappers = snapshot.cancelled_wrappers;
                next_queue_id = snapshot.next_queue_id;
                if next_queue_id.is_some_and(|id| id < QUEUE_ID_NAMESPACE) {
                    return Err(std::io::Error::new(
                        std::io::ErrorKind::InvalidData,
                        "invalid queued identifier high-water mark",
                    ));
                }
                if cancelled_wrappers.len() > MAX_CANCELLED_WRAPPERS {
                    return Err(std::io::Error::new(
                        std::io::ErrorKind::InvalidData,
                        "queued cancellation journal exceeds limit",
                    ));
                }
                for mut state in snapshot.active {
                    if state.id == 0
                        || state.id >= QUEUE_ID_NAMESPACE
                        || state.worker_id.is_empty()
                        || active.contains_key(&state.id)
                    {
                        return Err(std::io::Error::new(
                            std::io::ErrorKind::InvalidData,
                            "invalid or duplicate active ownership",
                        ));
                    }
                    for (wall, mono) in [
                        (&state.started_at, &mut state.started_at_mono),
                        (&state.last_heartbeat_at, &mut state.last_heartbeat_mono),
                        (&state.last_progress_at, &mut state.last_progress_mono),
                    ] {
                        let timestamp = DateTime::parse_from_rfc3339(wall).map_err(|error| {
                            std::io::Error::new(std::io::ErrorKind::InvalidData, error)
                        })?;
                        let age = Utc::now()
                            .signed_duration_since(timestamp)
                            .to_std()
                            .unwrap_or_default();
                        *mono = Instant::now().checked_sub(age).unwrap_or_else(Instant::now);
                    }
                    state.recovered = true;
                    max_id = max_id.max(state.id);
                    active.insert(state.id, state);
                }
                for receipt in snapshot.completed {
                    let id = receipt.record.id;
                    if active.contains_key(&id) || terminal.contains_key(&id) {
                        return Err(std::io::Error::new(
                            std::io::ErrorKind::InvalidData,
                            "conflicting terminal ownership",
                        ));
                    }
                    max_id = max_id.max(id);
                    if !records.iter().any(|record| record.id == id) {
                        records.push_back(receipt.record.clone());
                    }
                    terminal.insert(id, receipt);
                }
                while records.len() > capacity {
                    records.pop_front();
                }
            }
            Err(error) if error.kind() == std::io::ErrorKind::NotFound => {}
            Err(error) => return Err(error),
        }
        debug!("Loaded {} build records from {:?}", records.len(), path);

        let epoch_secs = std::time::SystemTime::now()
            .duration_since(std::time::UNIX_EPOCH)
            .unwrap_or_default()
            .as_secs();
        let epoch_id = (epoch_secs << 24) | 1;
        let initial_id = std::cmp::max(max_id.saturating_add(1), epoch_id);
        let queue_epoch = QUEUE_ID_NAMESPACE | epoch_id;
        let next_queue_id = next_queue_id.map_or(queue_epoch, |id| id.max(queue_epoch));

        let history = Self {
            records: RwLock::new(records),
            active: RwLock::new(active),
            queued: RwLock::new(VecDeque::new()),
            capacity,
            max_queue_depth: DEFAULT_MAX_QUEUE_DEPTH,
            next_id: AtomicU64::new(initial_id),
            next_queue_id: AtomicU64::new(next_queue_id),
            persistence_path: Some(path.to_path_buf()),
            terminal: RwLock::new(terminal),
            cancelled_wrappers: RwLock::new(cancelled_wrappers),
            ownership_failed: std::sync::atomic::AtomicBool::new(false),
            #[cfg(test)]
            fail_after_ownership_rename: std::sync::atomic::AtomicBool::new(false),
        };
        // A fresh start has nothing to make durable: the snapshot would hold
        // only the clock-derived queue epoch, which the next start recomputes,
        // and no queue ID has been issued yet. Skipping it keeps an fsync off
        // the path between binding the socket and serving it — on a worker
        // with heavy writeback that fsync stalled startup for 10+ seconds while
        // the bound socket accepted connections nobody answered.
        if ownership_existed {
            history.persist_ownership(
                &history.active.read().unwrap_or_else(|e| e.into_inner()),
                None,
            )?;
        }
        Ok(history)
    }

    /// Persist a single record to the JSONL file (append mode).
    ///
    /// Each completed build spawns its own task that calls this function,
    /// so multiple concurrent writers are normal. We rely on POSIX
    /// `O_APPEND` to serialize writes at the end of the file — but that
    /// guarantee only holds for a single `write()` syscall. The previous
    /// implementation split the record into two writes (JSON bytes, then
    /// `\n`) and two racing writers could produce
    /// `{record_a}{record_b}\n\n` instead of `{record_a}\n{record_b}\n`,
    /// corrupting the JSONL stream and breaking history recovery on
    /// daemon restart. Assemble the whole line in one buffer so a single
    /// `write_all` (which for <PIPE_BUF-sized payloads is a single
    /// syscall on Linux) keeps each line intact under concurrent append.
    async fn persist_record_async(path: &Path, record: &BuildRecord) -> std::io::Result<()> {
        let mut file = AsyncOpenOptions::new()
            .create(true)
            .append(true)
            .open(path)
            .await?;

        let mut line = serde_json::to_vec(record)?;
        line.push(b'\n');
        file.write_all(&line).await?;
        file.flush().await?;
        Ok(())
    }

    /// Compact the persistence file to only contain current records.
    #[allow(dead_code)] // May be used for maintenance operations
    pub fn compact(&self) -> std::io::Result<()> {
        let Some(ref path) = self.persistence_path else {
            return Ok(());
        };

        let records = self.records.read().unwrap_or_else(|e| e.into_inner());
        let temp_path = path.with_extension("tmp");

        {
            let mut file = File::create(&temp_path)?;
            for record in records.iter() {
                writeln!(file, "{}", serde_json::to_string(record)?)?;
            }
        }

        std::fs::rename(temp_path, path)?;
        debug!("Compacted history file: {:?}", path);

        Ok(())
    }

    pub fn terminal_build(&self, build_id: u64, wrapper: &str) -> Option<BuildRecord> {
        self.terminal
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .get(&build_id)
            .filter(|receipt| receipt.local_wrapper_id.as_deref() == Some(wrapper))
            .map(|receipt| receipt.record.clone())
    }
    pub fn has_terminal_build(&self, build_id: u64) -> bool {
        self.terminal
            .read()
            .unwrap_or_else(|e| e.into_inner())
            .contains_key(&build_id)
    }

    /// One locked transition owns both the durable terminal receipt and release.
    pub fn complete_durable(
        &self,
        build_id: u64,
        worker_id: &str,
        wrapper: Option<&str>,
        completion: BuildCompletion,
    ) -> std::io::Result<Option<(ActiveBuildState, BuildRecord)>> {
        let BuildCompletion {
            exit_code,
            duration_ms,
            bytes_transferred,
            timing,
            cancellation,
        } = completion;
        let mut active = self.active.write().unwrap_or_else(|e| e.into_inner());
        let Some(state) = active.get(&build_id) else {
            let terminal = self.terminal.read().unwrap_or_else(|e| e.into_inner());
            if terminal.get(&build_id).is_some_and(|receipt| {
                receipt.record.worker_id.as_deref() == Some(worker_id)
                    && receipt.local_wrapper_id.as_deref() == wrapper
            }) {
                return Ok(None);
            }
            return Err(std::io::Error::new(
                std::io::ErrorKind::NotFound,
                "unknown build or ownership mismatch",
            ));
        };
        if state.worker_id != worker_id || state.local_wrapper_id.as_deref() != wrapper {
            return Err(std::io::Error::new(
                std::io::ErrorKind::PermissionDenied,
                "build ownership mismatch",
            ));
        }
        let record = BuildRecord {
            id: state.id,
            started_at: state.started_at.clone(),
            completed_at: Utc::now().to_rfc3339(),
            project_id: state.project_id.clone(),
            worker_id: Some(state.worker_id.clone()),
            command: state.command.clone(),
            exit_code,
            duration_ms: duration_ms
                .unwrap_or_else(|| state.started_at_mono.elapsed().as_millis() as u64),
            location: state.location,
            bytes_transferred,
            timing,
            cancellation,
        };
        let receipt = TerminalOwnership {
            record: record.clone(),
            local_wrapper_id: state.local_wrapper_id.clone(),
        };
        self.persist_ownership(&active, Some(&receipt))?;
        self.terminal
            .write()
            .unwrap_or_else(|e| e.into_inner())
            .insert(build_id, receipt);
        let state = active.remove(&build_id).expect("locked ownership exists");
        self.record(record.clone());
        Ok(Some((state, record)))
    }

    /// Atomically commit ownership before admitting or acknowledging a job.
    fn persist_ownership(
        &self,
        active: &HashMap<u64, ActiveBuildState>,
        completed: Option<&TerminalOwnership>,
    ) -> std::io::Result<()> {
        if self.ownership_failed() {
            return Err(std::io::Error::other(
                "durable ownership uncertain; restart required",
            ));
        }
        let result = self.write_ownership(active, completed);
        if result.is_err() {
            self.ownership_failed.store(true, Ordering::SeqCst);
        }
        result
    }

    fn write_ownership(
        &self,
        active: &HashMap<u64, ActiveBuildState>,
        completed: Option<&TerminalOwnership>,
    ) -> std::io::Result<()> {
        let Some(path) = &self.persistence_path else {
            return Ok(());
        };
        let path = path.with_extension("ownership.json");
        let terminal = self.terminal.read().unwrap_or_else(|e| e.into_inner());
        let snapshot = DurableOwnership {
            version: 1,
            active: active
                .values()
                .filter(|state| completed.is_none_or(|receipt| receipt.record.id != state.id))
                .cloned()
                .collect(),
            completed: terminal.values().chain(completed).cloned().collect(),
            cancelled_wrappers: self
                .cancelled_wrappers
                .read()
                .unwrap_or_else(|e| e.into_inner())
                .clone(),
            next_queue_id: Some(self.next_queue_id.load(Ordering::SeqCst)),
        };
        let parent = path
            .parent()
            .filter(|parent| !parent.as_os_str().is_empty())
            .unwrap_or(Path::new("."));
        std::fs::create_dir_all(parent)?;
        let temporary = path.with_extension("tmp");
        let mut file = std::fs::OpenOptions::new()
            .write(true)
            .create(true)
            .truncate(true)
            .open(&temporary)?;
        serde_json::to_writer(&mut file, &snapshot)?;
        file.sync_all()?;
        std::fs::rename(&temporary, &path)?;
        #[cfg(test)]
        if self.fail_after_ownership_rename.load(Ordering::SeqCst) {
            return Err(std::io::Error::other(
                "injected post-rename ownership failure",
            ));
        }
        File::open(parent)?.sync_all()?;
        Ok(())
    }
}

impl Default for BuildHistory {
    fn default() -> Self {
        Self::with_default_capacity()
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use rch_common::test_guard;
    use std::time::{SystemTime, UNIX_EPOCH};
    use tempfile::TempDir;

    fn now_iso() -> String {
        let since_epoch = SystemTime::now()
            .duration_since(UNIX_EPOCH)
            .unwrap()
            .as_secs();
        format!("2024-01-01T00:00:{}Z", since_epoch % 60)
    }

    fn make_build_record(id: u64) -> BuildRecord {
        BuildRecord {
            id,
            started_at: now_iso(),
            completed_at: now_iso(),
            project_id: "test-project".to_string(),
            worker_id: None,
            command: "cargo build".to_string(),
            exit_code: 0,
            duration_ms: 100,
            location: BuildLocation::Local,
            bytes_transferred: None,
            timing: None,
            cancellation: None,
        }
    }

    #[tokio::test]
    async fn active_build_survives_history_restart() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");
        let history = BuildHistory::new(10).with_persistence(path.clone());
        // Ensure the persistence file exists through the normal completed
        // history path, so a missing file cannot mask loss of the active job.
        history
            .record(make_build_record(1))
            .expect("persistence task")
            .await
            .unwrap();
        let running = history.start_active_build_with_wrapper(
            "restart-project".to_string(),
            "worker-1".to_string(),
            "cargo check".to_string(),
            std::process::id(),
            Some("restart-wrapper".to_string()),
            2,
            BuildLocation::Remote,
        );
        drop(history);

        let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
        assert!(
            recovered.active_build(running.id).is_some(),
            "restart lost active build {}; surviving work must remain tracked",
            running.id
        );
    }

    #[test]
    fn test_ring_buffer_capacity() {
        let _guard = test_guard!();
        let history = BuildHistory::new(3);

        for i in 1..=5 {
            history.record(make_build_record(i));
        }

        let recent = history.recent(10);
        assert_eq!(recent.len(), 3); // Capped at capacity
        assert_eq!(recent[0].id, 5); // Most recent first
        assert_eq!(recent[2].id, 3); // Oldest retained
    }

    #[test]
    fn test_recent_ordering() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);
        history.record(make_build_record(1));
        history.record(make_build_record(2));
        history.record(make_build_record(3));

        let recent = history.recent(2);
        assert_eq!(recent.len(), 2);
        assert_eq!(recent[0].id, 3); // Most recent first
        assert_eq!(recent[1].id, 2);
    }

    #[test]
    fn test_by_worker_filter() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        let mut record1 = make_build_record(1);
        record1.worker_id = Some("worker-1".to_string());
        history.record(record1);

        let mut record2 = make_build_record(2);
        record2.worker_id = Some("worker-2".to_string());
        history.record(record2);

        let mut record3 = make_build_record(3);
        record3.worker_id = Some("worker-1".to_string());
        history.record(record3);

        let worker1_builds = history.by_worker("worker-1", 10);
        assert_eq!(worker1_builds.len(), 2);
        assert!(
            worker1_builds
                .iter()
                .all(|b| b.worker_id.as_deref() == Some("worker-1"))
        );
    }

    #[test]
    fn test_by_project_filter() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        let mut record1 = make_build_record(1);
        record1.project_id = "proj-a".to_string();
        history.record(record1);

        let mut record2 = make_build_record(2);
        record2.project_id = "proj-b".to_string();
        history.record(record2);

        let mut record3 = make_build_record(3);
        record3.project_id = "proj-a".to_string();
        history.record(record3);

        let proj_a_builds = history.by_project("proj-a", 10);
        assert_eq!(proj_a_builds.len(), 2);
        assert!(proj_a_builds.iter().all(|b| b.project_id == "proj-a"));
    }

    #[test]
    fn test_stats_calculation() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // 2 successes, 1 failure, 2 remote, 1 local
        let mut record1 = make_build_record(1);
        record1.exit_code = 0;
        record1.location = BuildLocation::Remote;
        record1.duration_ms = 1000;
        history.record(record1);

        let mut record2 = make_build_record(2);
        record2.exit_code = 0;
        record2.location = BuildLocation::Remote;
        record2.duration_ms = 2000;
        history.record(record2);

        let mut record3 = make_build_record(3);
        record3.exit_code = 1;
        record3.location = BuildLocation::Local;
        record3.duration_ms = 500;
        history.record(record3);

        let stats = history.stats();
        assert_eq!(stats.total_builds, 3);
        assert_eq!(stats.success_count, 2);
        assert_eq!(stats.failure_count, 1);
        assert_eq!(stats.remote_count, 2);
        assert_eq!(stats.local_count, 1);
        assert_eq!(stats.avg_duration_ms, 1166); // (1000+2000+500)/3
    }

    #[test]
    fn test_stats_preserve_terminal_outcomes_across_command_kinds() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // Exit 101 does not tell us whether compilation or assertions failed.
        let mut failing_test = make_build_record(1);
        failing_test.command =
            "env cargo test -p frankenterm-core some_test -- --nocapture".to_string();
        failing_test.exit_code = 101;
        history.record(failing_test);

        // A real build command that failed to compile — genuine build failure.
        let mut failing_build = make_build_record(2);
        failing_build.command = "cargo build --release".to_string();
        failing_build.exit_code = 1;
        history.record(failing_build);

        // A different nonzero test exit is also a failed command.
        let mut test_compile_error = make_build_record(3);
        test_compile_error.command = "cargo test --workspace".to_string();
        test_compile_error.exit_code = 1;
        history.record(test_compile_error);

        // A passing test command.
        let mut passing_test = make_build_record(4);
        passing_test.command = "cargo nextest run".to_string();
        passing_test.exit_code = 0;
        history.record(passing_test);

        let stats = history.stats();
        assert_eq!(stats.total_builds, 4);
        assert_eq!(
            stats.success_count, 1,
            "only a successful terminal command belongs in the success numerator"
        );
        assert_eq!(stats.failure_count, 3);
    }

    #[test]
    fn test_stats_never_promote_nonzero_exits_to_success() {
        let _guard = test_guard!();
        for command in [
            "cargo test",
            "cargo bench",
            "cargo nextest run",
            "bun test",
            "cargo build",
            "cargo check",
            "env cargo test --workspace",
        ] {
            let history = BuildHistory::new(256);
            for exit_code in 0..=255 {
                let mut record = make_build_record(exit_code as u64);
                record.command = command.to_string();
                record.exit_code = exit_code;
                history.record(record);
            }
            let stats = history.stats();
            assert_eq!(stats.total_builds, 256, "{command}");
            assert_eq!(stats.success_count, 1, "{command}");
            assert_eq!(stats.failure_count, 255, "{command}");
            assert_eq!(history.recent(1)[0].exit_code, 255);
        }
    }

    #[test]
    fn test_empty_history() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        assert!(history.recent(10).is_empty());
        assert!(history.by_worker("any", 10).is_empty());

        let stats = history.stats();
        assert_eq!(stats.total_builds, 0);
        assert_eq!(stats.avg_duration_ms, 0);
    }

    #[test]
    fn test_next_id() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // IDs use a timestamp-based epoch, so we verify monotonic sequence
        // rather than exact values.
        let id1 = history.next_id();
        let id2 = history.next_id();
        let id3 = history.next_id();
        assert!(id1 > 0, "first ID should be positive");
        assert_eq!(id2, id1 + 1, "IDs should be sequential");
        assert_eq!(id3, id2 + 1, "IDs should be sequential");
    }

    #[test]
    fn test_try_start_active_build_blocks_same_project_on_same_worker() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        let first = history.try_start_active_build(
            "proj-a".to_string(),
            "worker-1".to_string(),
            "cargo test".to_string(),
            1111,
            2,
            BuildLocation::Remote,
        );
        assert!(first.is_some(), "first active build should be registered");

        let second = history.try_start_active_build(
            "proj-a".to_string(),
            "worker-1".to_string(),
            "cargo test".to_string(),
            2222,
            2,
            BuildLocation::Remote,
        );
        assert!(
            second.is_none(),
            "same project on same worker must be rejected"
        );

        let different_worker = history.try_start_active_build(
            "proj-a".to_string(),
            "worker-2".to_string(),
            "cargo test".to_string(),
            3333,
            2,
            BuildLocation::Remote,
        );
        assert!(
            different_worker.is_some(),
            "same project on a different worker should still be allowed"
        );

        let different_project = history.try_start_active_build(
            "proj-b".to_string(),
            "worker-1".to_string(),
            "cargo test".to_string(),
            4444,
            2,
            BuildLocation::Remote,
        );
        assert!(
            different_project.is_some(),
            "different project on the same worker should still be allowed"
        );
    }

    #[test]
    fn test_active_workers_for_project_tracks_current_workers() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        history.start_active_build(
            "proj-a".to_string(),
            "worker-1".to_string(),
            "cargo build".to_string(),
            1111,
            2,
            BuildLocation::Remote,
        );
        history.start_active_build(
            "proj-a".to_string(),
            "worker-2".to_string(),
            "cargo build".to_string(),
            2222,
            2,
            BuildLocation::Remote,
        );
        history.start_active_build(
            "proj-b".to_string(),
            "worker-3".to_string(),
            "cargo build".to_string(),
            3333,
            2,
            BuildLocation::Remote,
        );

        let active_workers = history.active_workers_for_project("proj-a");
        assert_eq!(active_workers.len(), 2);
        assert!(active_workers.contains("worker-1"));
        assert!(active_workers.contains("worker-2"));
        assert!(!active_workers.contains("worker-3"));
    }

    #[test]
    fn test_cancel_active_build_records_cancellation_metadata() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);
        let active = history.start_active_build(
            "proj".to_string(),
            "worker-a".to_string(),
            "cargo test".to_string(),
            0,
            4,
            BuildLocation::Remote,
        );

        let metadata = BuildCancellationMetadata {
            operation_id: "cancel-1".to_string(),
            origin: "timeout".to_string(),
            reason_code: "timeout".to_string(),
            decision_path: vec![
                "requested".to_string(),
                "term_sent".to_string(),
                "remote_kill_sent".to_string(),
                "completed".to_string(),
            ],
            escalation_stage: "remote_kill".to_string(),
            escalation_count: 1,
            remote_kill_attempted: true,
            cleanup_ok: true,
            history_cancelled: true,
            final_state: "completed".to_string(),
            worker_health: Some(rch_common::BuildCancellationWorkerHealth {
                status: "healthy".to_string(),
                speed_score: 91.2,
                used_slots: 0,
                available_slots: 8,
                pressure_state: "healthy".to_string(),
                pressure_reason_code: "healthy".to_string(),
            }),
        };

        let cancelled = history
            .cancel_active_build(active.id, None, Some(metadata.clone()))
            .expect("cancelled build record");
        assert_eq!(cancelled.exit_code, 130);
        assert_eq!(
            cancelled
                .cancellation
                .as_ref()
                .expect("cancellation metadata")
                .operation_id,
            metadata.operation_id
        );
        assert!(history.active_build(active.id).is_none());

        let recent = history.recent(1);
        assert_eq!(recent.len(), 1);
        assert_eq!(
            recent[0]
                .cancellation
                .as_ref()
                .expect("persisted cancellation metadata")
                .escalation_stage,
            "remote_kill"
        );
    }

    #[test]
    fn test_record_build_heartbeat_updates_progress_metadata() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);
        let build = history.start_active_build(
            "proj".to_string(),
            "worker-a".to_string(),
            "cargo build".to_string(),
            1234,
            4,
            BuildLocation::Remote,
        );
        let initial_progress_at = build.last_progress_at.clone();

        let updated = history
            .record_build_heartbeat(BuildHeartbeatRequest {
                build_id: build.id,
                worker_id: rch_common::WorkerId::new("worker-a"),
                hook_pid: Some(1234),
                local_wrapper_id: Some("rchw-test".to_string()),
                remote_pgid_file: Some("/tmp/rch/proj/hash/.rch-run/1.pgid".to_string()),
                phase: BuildHeartbeatPhase::Execute,
                detail: Some("Compiling".to_string()),
                progress_counter: Some(3),
                progress_percent: Some(25.0),
            })
            .expect("active build should be updated");

        assert_eq!(updated.heartbeat_phase, BuildHeartbeatPhase::Execute);
        assert_eq!(updated.heartbeat_counter, 3);
        assert_eq!(updated.heartbeat_percent, Some(25.0));
        assert_eq!(updated.heartbeat_count, 1);
        assert_eq!(
            updated.remote_pgid_file.as_deref(),
            Some("/tmp/rch/proj/hash/.rch-run/1.pgid")
        );
        assert_ne!(updated.last_progress_at, initial_progress_at);
    }

    #[test]
    fn test_record_build_heartbeat_rejects_worker_mismatch() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);
        let build = history.start_active_build(
            "proj".to_string(),
            "worker-a".to_string(),
            "cargo check".to_string(),
            9999,
            2,
            BuildLocation::Remote,
        );

        let updated = history.record_build_heartbeat(BuildHeartbeatRequest {
            build_id: build.id,
            worker_id: rch_common::WorkerId::new("worker-b"),
            hook_pid: Some(9999),
            local_wrapper_id: None,
            remote_pgid_file: None,
            phase: BuildHeartbeatPhase::Execute,
            detail: Some("Unexpected".to_string()),
            progress_counter: Some(1),
            progress_percent: Some(10.0),
        });
        assert!(
            updated.is_none(),
            "mismatched worker heartbeat must be ignored"
        );
    }

    #[cfg(target_os = "linux")]
    mod heartbeat_identity {
        use super::*;

        struct OwnedWrapper(std::process::Child);

        impl OwnedWrapper {
            fn start() -> Self {
                Self(
                    std::process::Command::new("/bin/sleep")
                        .arg("60")
                        .spawn()
                        .unwrap(),
                )
            }

            fn assert_running(&mut self) {
                assert!(self.0.try_wait().unwrap().is_none());
            }
        }

        impl Drop for OwnedWrapper {
            fn drop(&mut self) {
                let _ = self.0.kill();
                let _ = self.0.wait();
            }
        }

        fn register(history: &BuildHistory, pid: u32) -> ActiveBuildState {
            history.start_active_build_with_wrapper(
                "heartbeat-owner-project".to_owned(),
                "heartbeat-owner-worker".to_owned(),
                "controlled heartbeat fixture".to_owned(),
                pid,
                Some("exact-heartbeat-wrapper".to_owned()),
                2,
                BuildLocation::Remote,
            )
        }

        fn heartbeat(build: &ActiveBuildState, pid: Option<u32>) -> BuildHeartbeatRequest {
            BuildHeartbeatRequest {
                build_id: build.id,
                worker_id: rch_common::WorkerId::new(&build.worker_id),
                hook_pid: pid,
                local_wrapper_id: build.local_wrapper_id.clone(),
                remote_pgid_file: Some("/tmp/rch/heartbeat-owner/.rch-run/owned.pgid".to_owned()),
                phase: BuildHeartbeatPhase::Execute,
                detail: Some("owned wrapper making progress".to_owned()),
                progress_counter: Some(3),
                progress_percent: Some(25.0),
            }
        }

        fn persistent_history(root: &TempDir) -> (BuildHistory, PathBuf) {
            let path = root.path().join("history.jsonl");
            std::fs::write(&path, b"").unwrap();
            (BuildHistory::new(10).with_persistence(path.clone()), path)
        }

        fn persist_process_identity(
            history: &BuildHistory,
            build_id: u64,
            identity: Option<String>,
        ) {
            let mut active = history.active.write().unwrap();
            active.get_mut(&build_id).unwrap().hook_process_identity = identity;
            history.persist_ownership(&active, None).unwrap();
        }

        fn assert_unchanged(history: &BuildHistory, before: &ActiveBuildState) {
            let after = history.active_build(before.id).unwrap();
            assert_eq!(
                serde_json::to_value(&after).unwrap(),
                serde_json::to_value(before).unwrap()
            );
            assert_eq!(after.started_at_mono, before.started_at_mono);
            assert_eq!(after.last_heartbeat_mono, before.last_heartbeat_mono);
            assert_eq!(after.last_progress_mono, before.last_progress_mono);
        }

        #[test]
        fn live_heartbeat_preserves_established_process_identity() {
            let mut owner = OwnedWrapper::start();
            let history = BuildHistory::new(10);
            let build = register(&history, owner.0.id());
            let identity = process_identity(owner.0.id()).unwrap();
            assert_eq!(
                build.hook_process_identity.as_deref(),
                Some(identity.as_str())
            );

            let updated = history
                .record_build_heartbeat(heartbeat(&build, Some(owner.0.id())))
                .unwrap();
            assert_eq!(updated.hook_process_identity, Some(identity));
            assert_eq!(updated.hook_pid, build.hook_pid);
            assert_eq!(updated.local_wrapper_id, build.local_wrapper_id);
            assert_eq!(updated.heartbeat_counter, 3);
            assert_eq!(updated.heartbeat_count, 1);
            assert_eq!(updated.heartbeat_phase, BuildHeartbeatPhase::Execute);
            owner.assert_running();
        }

        #[test]
        fn heartbeat_cannot_replace_wrapper_pid_even_with_exact_wrapper_id() {
            let mut owner = OwnedWrapper::start();
            let mut unrelated = OwnedWrapper::start();
            let root = TempDir::new().unwrap();
            let (history, path) = persistent_history(&root);
            let build = register(&history, owner.0.id());
            let journal = std::fs::read(path.with_extension("ownership.json")).unwrap();

            assert!(
                history
                    .record_build_heartbeat(heartbeat(&build, Some(unrelated.0.id())))
                    .is_none()
            );
            assert_unchanged(&history, &build);
            assert_eq!(
                std::fs::read(path.with_extension("ownership.json")).unwrap(),
                journal
            );
            owner.assert_running();
            unrelated.assert_running();
        }

        #[test]
        fn delayed_recovered_heartbeat_rejects_modelled_pid_reuse_without_adoption() {
            let mut unrelated = OwnedWrapper::start();
            let actual = process_identity(unrelated.0.id()).unwrap();
            let (boot, ticks) = actual.rsplit_once(':').unwrap();
            let ticks = ticks.parse::<u64>().unwrap();
            let prior_ticks = if ticks > 0 { ticks - 1 } else { 1 };
            let prior_identity = format!("{boot}:{prior_ticks}");
            assert_ne!(prior_identity, actual);

            let root = TempDir::new().unwrap();
            let (history, path) = persistent_history(&root);
            let build = register(&history, unrelated.0.id());
            // Model PID reuse by recording the prior owner's distinct start
            // ticks. The unrelated current occupant is a real live process;
            // this test does not claim to force operating-system PID reuse.
            persist_process_identity(&history, build.id, Some(prior_identity));
            drop(history);
            let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
            let before = recovered.active_build(build.id).unwrap();
            assert!(before.recovered);
            let journal = std::fs::read(path.with_extension("ownership.json")).unwrap();

            for pid in [Some(unrelated.0.id()), None] {
                assert!(
                    recovered
                        .record_build_heartbeat(heartbeat(&before, pid))
                        .is_none(),
                    "a delayed heartbeat must not authorize the current PID occupant"
                );
                assert_unchanged(&recovered, &before);
                assert_eq!(
                    std::fs::read(path.with_extension("ownership.json")).unwrap(),
                    journal
                );
            }
            assert_eq!(process_identity(unrelated.0.id()).unwrap(), actual);
            unrelated.assert_running();
        }

        #[test]
        fn recovered_heartbeat_accepts_matching_live_process_without_rebinding() {
            let mut owner = OwnedWrapper::start();
            let root = TempDir::new().unwrap();
            let (history, path) = persistent_history(&root);
            let build = register(&history, owner.0.id());
            assert!(build.hook_process_identity.is_some());
            drop(history);
            let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
            assert!(recovered.active_build(build.id).unwrap().recovered);

            let updated = recovered
                .record_build_heartbeat(heartbeat(&build, Some(owner.0.id())))
                .unwrap();
            assert_eq!(updated.hook_process_identity, build.hook_process_identity);
            assert_eq!(updated.hook_pid, build.hook_pid);
            assert_eq!(updated.local_wrapper_id, build.local_wrapper_id);
            assert!(updated.recovered);
            assert_eq!(updated.heartbeat_count, 1);
            drop(recovered);

            let reopened = BuildHistory::load_from_file(&path, 10).unwrap();
            let persisted = reopened.active_build(build.id).unwrap();
            assert_eq!(persisted.hook_process_identity, build.hook_process_identity);
            assert_eq!(persisted.heartbeat_count, 1);
            assert_eq!(persisted.heartbeat_counter, 3);
            owner.assert_running();
        }

        #[test]
        fn heartbeat_never_captures_identity_for_preexisting_unverified_pid() {
            let mut unrelated = OwnedWrapper::start();
            let history = BuildHistory::new(10);
            let build = register(&history, unrelated.0.id());
            persist_process_identity(&history, build.id, None);
            let before = history.active_build(build.id).unwrap();
            assert!(!before.recovered);

            let _ = history.record_build_heartbeat(heartbeat(&before, Some(unrelated.0.id())));
            let after = history.active_build(build.id).unwrap();
            assert!(after.hook_process_identity.is_none());
            assert_eq!(after.hook_pid, before.hook_pid);
            assert!(!after.recovered);
            unrelated.assert_running();
        }

        #[test]
        fn recovered_heartbeat_refuses_missing_process_identity() {
            let mut unrelated = OwnedWrapper::start();
            let root = TempDir::new().unwrap();
            let (history, path) = persistent_history(&root);
            let build = register(&history, unrelated.0.id());
            persist_process_identity(&history, build.id, None);
            drop(history);
            let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
            let before = recovered.active_build(build.id).unwrap();
            assert!(before.recovered);
            let journal = std::fs::read(path.with_extension("ownership.json")).unwrap();

            assert!(
                recovered
                    .record_build_heartbeat(heartbeat(&before, Some(unrelated.0.id())))
                    .is_none()
            );
            assert_unchanged(&recovered, &before);
            assert_eq!(
                std::fs::read(path.with_extension("ownership.json")).unwrap(),
                journal
            );
            unrelated.assert_running();
        }
    }

    #[test]
    fn test_record_stuck_detector_snapshot_updates_active_build() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);
        let build = history.start_active_build(
            "proj".to_string(),
            "worker-a".to_string(),
            "cargo test".to_string(),
            4321,
            6,
            BuildLocation::Remote,
        );

        let updated = history
            .record_stuck_detector_snapshot(
                build.id,
                StuckDetectorSnapshot {
                    hook_alive: false,
                    heartbeat_stale: true,
                    progress_stale: true,
                    confidence: 0.91,
                    build_age_secs: 140,
                    slots_owned: 6,
                },
            )
            .expect("active build should exist");

        assert!(!updated.detector_hook_alive);
        assert!(updated.detector_heartbeat_stale);
        assert!(updated.detector_progress_stale);
        assert_eq!(updated.detector_confidence, 0.91);
        assert_eq!(updated.detector_build_age_secs, 140);
        assert_eq!(updated.detector_slots_owned, 6);
        assert!(updated.detector_last_evaluated_at.is_some());
    }

    #[tokio::test]
    async fn test_thread_safety() {
        use std::sync::Arc;

        let history = Arc::new(BuildHistory::new(100));

        let handles: Vec<_> = (0..10)
            .map(|i| {
                let h = Arc::clone(&history);
                tokio::spawn(async move {
                    for j in 0..10 {
                        h.record(make_build_record((i * 10 + j) as u64));
                    }
                })
            })
            .collect();

        for handle in handles {
            handle.await.unwrap();
        }

        let recent = history.recent(200);
        assert_eq!(recent.len(), 100); // All 100 recorded
    }

    #[tokio::test]
    async fn test_persisted_stats_do_not_invent_compilation_success() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");
        let history = BuildHistory::new(10).with_persistence(path.clone());
        let exits = [0, 1, 101, 130, 137, 143];
        for (id, exit_code) in exits.into_iter().enumerate() {
            let mut record = make_build_record(id as u64);
            record.command = "cargo test".to_string();
            record.exit_code = exit_code;
            if let Some(handle) = history.record(record) {
                handle.await.unwrap();
            }
        }
        let loaded = BuildHistory::load_from_file(&path, 10).unwrap();
        for stats in [history.stats(), loaded.stats()] {
            assert_eq!(stats.total_builds, exits.len());
            assert_eq!(stats.success_count, 1);
            assert_eq!(stats.failure_count, exits.len() - 1);
        }
        let persisted_exits: Vec<_> = loaded
            .recent(10)
            .iter()
            .rev()
            .map(|record| record.exit_code)
            .collect();
        assert_eq!(persisted_exits, exits);
    }

    #[tokio::test]
    async fn test_persistence_save_load() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");

        // Create and populate history
        let history = BuildHistory::new(5).with_persistence(path.clone());
        for i in 1..=3 {
            if let Some(handle) = history.record(make_build_record(i)) {
                handle.await.unwrap();
            }
        }

        // Load into new instance
        let loaded = BuildHistory::load_from_file(&path, 5).unwrap();
        let recent = loaded.recent(10);

        assert_eq!(recent.len(), 3);
        assert_eq!(recent[0].id, 3);
    }

    #[tokio::test]
    async fn test_persistence_append_mode() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");

        // First session
        {
            let history = BuildHistory::new(10).with_persistence(path.clone());
            if let Some(handle) = history.record(make_build_record(1)) {
                handle.await.unwrap();
            }
            if let Some(handle) = history.record(make_build_record(2)) {
                handle.await.unwrap();
            }
        }

        // Second session - load and add more
        {
            let history = BuildHistory::load_from_file(&path, 10).unwrap();
            // Use next_id to ensure we don't duplicate IDs
            let id = history.next_id();
            if let Some(handle) = history.record(make_build_record(id)) {
                handle.await.unwrap();
            }
        }

        // Third session - verify all records
        let history = BuildHistory::load_from_file(&path, 10).unwrap();
        assert_eq!(history.len(), 3);
    }

    #[tokio::test]
    async fn test_compaction() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");

        // Create history with 3 records but capacity 2
        let history = BuildHistory::new(2).with_persistence(path.clone());
        for i in 1..=3 {
            if let Some(handle) = history.record(make_build_record(i)) {
                handle.await.unwrap();
            }
        }

        // Compact
        history.compact().unwrap();

        // Verify file only has 2 records
        let loaded = BuildHistory::load_from_file(&path, 10).unwrap();
        assert_eq!(loaded.len(), 2);
    }

    #[test]
    fn test_clear() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);
        history.record(make_build_record(1));
        history.record(make_build_record(2));

        assert_eq!(history.len(), 2);

        history.clear();

        assert_eq!(history.len(), 0);
        assert!(history.is_empty());
    }

    // =========================================================================
    // Queue Tests
    // =========================================================================

    #[test]
    fn queued_cancellation_survives_restart_and_blocks_only_same_wrapper() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");
        let history = BuildHistory::new(10).with_persistence(path.clone());
        history
            .enqueue_build(
                "project".into(),
                "cargo build".into(),
                0,
                2,
                Some("cancelled".into()),
            )
            .unwrap();
        history
            .enqueue_build(
                "other".into(),
                "cargo build".into(),
                0,
                2,
                Some("other".into()),
            )
            .unwrap();
        assert!(matches!(
            history.cancel_wrapper("cancelled").unwrap(),
            WrapperCancellation::BeforeStart
        ));
        assert_eq!(history.queue_depth(), 1);
        assert!(matches!(
            history.cancel_wrapper("cancelled").unwrap(),
            WrapperCancellation::BeforeStart
        ));
        let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
        assert!(recovered.wrapper_cancelled("cancelled"));
        assert!(
            recovered
                .try_start_active_build_with_wrapper(
                    "project".into(),
                    "worker".into(),
                    "cargo build".into(),
                    0,
                    Some("cancelled".into()),
                    2,
                    BuildLocation::Remote
                )
                .unwrap()
                .is_none()
        );
        assert!(
            recovered
                .try_start_active_build_with_wrapper(
                    "project".into(),
                    "worker".into(),
                    "cargo build".into(),
                    0,
                    Some("other".into()),
                    2,
                    BuildLocation::Remote
                )
                .unwrap()
                .is_some()
        );
    }

    #[test]
    fn queued_cancellation_and_admission_have_exactly_one_winner() {
        use std::sync::{Arc, Barrier};
        for _ in 0..32 {
            let history = Arc::new(BuildHistory::new(10));
            history
                .enqueue_build(
                    "project".into(),
                    "cargo build".into(),
                    0,
                    2,
                    Some("racing".into()),
                )
                .unwrap();
            let barrier = Arc::new(Barrier::new(2));
            let admission_history = history.clone();
            let admission_barrier = barrier.clone();
            let admission = std::thread::spawn(move || {
                admission_barrier.wait();
                admission_history
                    .try_start_active_build_with_wrapper(
                        "project".into(),
                        "worker".into(),
                        "cargo build".into(),
                        0,
                        Some("racing".into()),
                        2,
                        BuildLocation::Remote,
                    )
                    .unwrap()
            });
            barrier.wait();
            let cancellation = history.cancel_wrapper("racing").unwrap();
            let admitted = admission.join().unwrap();
            match cancellation {
                WrapperCancellation::BeforeStart => {
                    assert!(admitted.is_none());
                    assert!(history.active_builds().is_empty());
                    assert!(history.wrapper_cancelled("racing"));
                }
                WrapperCancellation::Active(build_id) => {
                    assert_eq!(admitted.unwrap().id, build_id);
                    assert!(!history.wrapper_cancelled("racing"));
                    assert_eq!(history.active_builds().len(), 1);
                }
                WrapperCancellation::Completed(_) => panic!("no command completed in this race"),
                WrapperCancellation::NotQueued => panic!("the queue entry was not removed"),
            }
        }
    }

    #[test]
    fn queued_departure_and_cancellation_cannot_both_authorize_their_outcome() {
        use std::sync::{Arc, Barrier};
        for _ in 0..32 {
            let history = Arc::new(BuildHistory::new(10));
            let queued = history
                .enqueue_build(
                    "project".into(),
                    "cargo build".into(),
                    0,
                    2,
                    Some("racing".into()),
                )
                .unwrap();
            let barrier = Arc::new(Barrier::new(2));
            let leaving_history = history.clone();
            let leaving_barrier = barrier.clone();
            let departure = std::thread::spawn(move || {
                leaving_barrier.wait();
                leaving_history
                    .finish_queued_build(queued.id, Some("racing"))
                    .unwrap()
            });
            barrier.wait();
            let cancelled = history.cancel_wrapper("racing").unwrap();
            let must_stop = departure.join().unwrap();
            match cancelled {
                WrapperCancellation::BeforeStart => assert!(must_stop),
                WrapperCancellation::NotQueued => assert!(!must_stop),
                _ => panic!("no active or completed build in departure race"),
            }
            assert_eq!(history.queue_depth(), 0);
        }
    }

    #[test]
    fn queued_cancellation_persistence_failure_never_acknowledges_or_drops_queue() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");
        let history = BuildHistory::new(10).with_persistence(path.clone());
        history
            .enqueue_build(
                "project".into(),
                "cargo build".into(),
                0,
                2,
                Some("cancelled".into()),
            )
            .unwrap();
        history
            .fail_after_ownership_rename
            .store(true, Ordering::SeqCst);
        assert!(history.cancel_wrapper("cancelled").is_err());
        assert!(history.ownership_failed());
        assert_eq!(history.queue_depth(), 1);
        assert!(
            history
                .try_start_active_build_with_wrapper(
                    "project".into(),
                    "worker".into(),
                    "cargo build".into(),
                    0,
                    Some("different".into()),
                    2,
                    BuildLocation::Remote
                )
                .unwrap()
                .is_none()
        );
        let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
        assert!(recovered.wrapper_cancelled("cancelled"));
    }

    #[test]
    fn queued_cancellation_full_journal_retains_old_intents() {
        let history = BuildHistory::new(10);
        *history.cancelled_wrappers.write().unwrap() = (0..MAX_CANCELLED_WRAPPERS)
            .map(|i| format!("wrapper-{i}"))
            .collect();
        history
            .enqueue_build(
                "project".into(),
                "cargo build".into(),
                0,
                2,
                Some("new-wrapper".into()),
            )
            .unwrap();
        assert!(history.cancel_wrapper("new-wrapper").is_err());
        assert!(!history.wrapper_cancelled("new-wrapper"));
        assert!(matches!(
            history.cancel_wrapper("wrapper-0").unwrap(),
            WrapperCancellation::BeforeStart
        ));
        assert_eq!(
            history.cancelled_wrappers.read().unwrap().len(),
            MAX_CANCELLED_WRAPPERS
        );
    }

    #[test]
    fn queued_ids_survive_restart_and_clock_rollback_without_reuse() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");
        let history = BuildHistory::new(10).with_persistence(path.clone());
        // Model a previous daemon that allocated beyond the current wall-clock
        // seed. Restart must honor the durable high-water, even after rollback.
        let future = history.next_queue_id.load(Ordering::SeqCst) + (7200 << 24);
        history.next_queue_id.store(future, Ordering::SeqCst);
        let old = history
            .enqueue_build(
                "old".into(),
                "cargo build".into(),
                0,
                2,
                Some("old-wrapper".into()),
            )
            .unwrap();
        assert_eq!(old.id, future);
        let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
        let new = recovered
            .enqueue_build(
                "new".into(),
                "cargo build".into(),
                0,
                2,
                Some("new-wrapper".into()),
            )
            .unwrap();
        assert_eq!(new.id, old.id + 1);
        assert!(new.id >= QUEUE_ID_NAMESPACE);
        assert!(recovered.queued_build(old.id).is_none());
        assert_eq!(
            recovered
                .queued_build(new.id)
                .unwrap()
                .local_wrapper_id
                .as_deref(),
            Some("new-wrapper")
        );
        assert!(matches!(
            recovered.cancel_wrapper("old-wrapper").unwrap(),
            WrapperCancellation::NotQueued
        ));
        assert_eq!(recovered.queue_depth(), 1);
    }

    #[test]
    fn fresh_load_defers_ownership_write_but_existing_snapshot_is_rewritten() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");
        let ownership = path.with_extension("ownership.json");

        // Fresh start: nothing recovered, so no durable write before serving.
        let fresh = BuildHistory::load_from_file(&path, 10).unwrap();
        assert!(!ownership.exists());
        assert!(!path.with_extension("tmp").exists());

        // The first ownership change still becomes durable immediately.
        fresh
            .try_start_active_build_with_wrapper(
                "active".into(),
                "worker".into(),
                "cargo build".into(),
                0,
                None,
                2,
                BuildLocation::Remote,
            )
            .unwrap()
            .unwrap();
        assert!(ownership.exists());

        // A restart that finds a snapshot re-persists it (queue epoch advance).
        std::fs::write(&ownership, br#"{"version":1,"active":[],"completed":[]}"#).unwrap();
        BuildHistory::load_from_file(&path, 10).unwrap();
        let rewritten = std::fs::read_to_string(&ownership).unwrap();
        assert!(rewritten.contains("next_queue_id"), "{rewritten}");
    }

    #[test]
    fn queued_ids_migrate_legacy_snapshot_into_separate_namespace() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");
        std::fs::write(
            path.with_extension("ownership.json"),
            br#"{"version":1,"active":[],"completed":[]}"#,
        )
        .unwrap();
        let history = BuildHistory::load_from_file(&path, 10).unwrap();
        let queued = history
            .enqueue_build("queued".into(), "cargo build".into(), 0, 2, None)
            .unwrap();
        let active = history
            .try_start_active_build_with_wrapper(
                "active".into(),
                "worker".into(),
                "cargo build".into(),
                0,
                None,
                2,
                BuildLocation::Remote,
            )
            .unwrap()
            .unwrap();
        assert!(queued.id >= QUEUE_ID_NAMESPACE);
        assert!(active.id < QUEUE_ID_NAMESPACE);
        let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
        assert!(
            recovered
                .enqueue_build("next".into(), "cargo build".into(), 0, 2, None)
                .unwrap()
                .id
                > queued.id
        );
    }

    #[test]
    fn queued_id_allocation_failure_stays_invisible_and_exhaustion_never_wraps() {
        let tmp = TempDir::new().unwrap();
        let path = tmp.path().join("history.jsonl");
        let history = BuildHistory::new(10).with_persistence(path.clone());
        let skipped = history.next_queue_id.load(Ordering::SeqCst);
        history
            .fail_after_ownership_rename
            .store(true, Ordering::SeqCst);
        assert!(
            history
                .enqueue_build("failed".into(), "cargo build".into(), 0, 2, None)
                .is_none()
        );
        assert!(history.ownership_failed());
        assert_eq!(history.queue_depth(), 0);
        let recovered = BuildHistory::load_from_file(&path, 10).unwrap();
        let queued = recovered
            .enqueue_build("next".into(), "cargo build".into(), 0, 2, None)
            .unwrap();
        assert!(queued.id > skipped);
        recovered.next_queue_id.store(u64::MAX, Ordering::SeqCst);
        assert!(
            recovered
                .enqueue_build("overflow".into(), "cargo build".into(), 0, 2, None)
                .is_none()
        );
        assert_eq!(recovered.next_queue_id.load(Ordering::SeqCst), u64::MAX);
        assert_eq!(recovered.queue_depth(), 1);
    }

    #[test]
    fn test_enqueue_dequeue_fifo() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // Enqueue three builds
        let b1 = history
            .enqueue_build("proj-a".into(), "cargo build".into(), 1001, 4, None)
            .unwrap();
        let b2 = history
            .enqueue_build("proj-b".into(), "cargo test".into(), 1002, 8, None)
            .unwrap();
        let b3 = history
            .enqueue_build("proj-c".into(), "cargo check".into(), 1003, 2, None)
            .unwrap();

        assert_eq!(history.queue_depth(), 3);

        // Dequeue in FIFO order
        let d1 = history.dequeue_build().unwrap();
        assert_eq!(d1.id, b1.id);
        assert_eq!(d1.project_id, "proj-a");

        let d2 = history.dequeue_build().unwrap();
        assert_eq!(d2.id, b2.id);

        let d3 = history.dequeue_build().unwrap();
        assert_eq!(d3.id, b3.id);

        assert!(history.queue_is_empty());
        assert!(history.dequeue_build().is_none());
    }

    #[test]
    fn test_queue_depth_limit() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10).with_max_queue_depth(2);

        // First two should succeed
        assert!(
            history
                .enqueue_build("proj-a".into(), "build".into(), 1, 4, None)
                .is_some()
        );
        assert!(
            history
                .enqueue_build("proj-b".into(), "build".into(), 2, 4, None)
                .is_some()
        );

        // Third should fail
        assert!(
            history
                .enqueue_build("proj-c".into(), "build".into(), 3, 4, None)
                .is_none()
        );

        // Dequeue one, then third should succeed
        history.dequeue_build();
        assert!(
            history
                .enqueue_build("proj-c".into(), "build".into(), 3, 4, None)
                .is_some()
        );
    }

    #[test]
    fn test_queue_position() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        let b1 = history
            .enqueue_build("proj-a".into(), "build".into(), 1, 4, None)
            .unwrap();
        let b2 = history
            .enqueue_build("proj-b".into(), "build".into(), 2, 4, None)
            .unwrap();
        let b3 = history
            .enqueue_build("proj-c".into(), "build".into(), 3, 4, None)
            .unwrap();

        // Positions are 1-indexed
        assert_eq!(history.queue_position(b1.id), Some(1));
        assert_eq!(history.queue_position(b2.id), Some(2));
        assert_eq!(history.queue_position(b3.id), Some(3));
        assert_eq!(history.queue_position(999), None);
    }

    #[test]
    fn test_remove_queued_build() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        let b1 = history
            .enqueue_build("proj-a".into(), "build".into(), 1001, 4, None)
            .unwrap();
        let b2 = history
            .enqueue_build("proj-b".into(), "build".into(), 1002, 4, None)
            .unwrap();
        let b3 = history
            .enqueue_build("proj-c".into(), "build".into(), 1003, 4, None)
            .unwrap();

        // Remove middle build by ID
        let removed = history.remove_queued_build(b2.id).unwrap();
        assert_eq!(removed.project_id, "proj-b");
        assert_eq!(history.queue_depth(), 2);

        // Remove by PID
        let removed = history.remove_queued_build_by_pid(1003).unwrap();
        assert_eq!(removed.id, b3.id);
        assert_eq!(history.queue_depth(), 1);

        // Only b1 remains
        let d = history.dequeue_build().unwrap();
        assert_eq!(d.id, b1.id);
    }

    #[test]
    fn test_queued_builds_list() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        history.enqueue_build("proj-a".into(), "build".into(), 1, 4, None);
        history.enqueue_build("proj-b".into(), "test".into(), 2, 8, None);

        let queued = history.queued_builds();
        assert_eq!(queued.len(), 2);
        assert_eq!(queued[0].project_id, "proj-a");
        assert_eq!(queued[1].project_id, "proj-b");
    }

    #[test]
    fn test_queued_build_lookup() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        let b = history
            .enqueue_build("proj-a".into(), "cargo build".into(), 1001, 4, None)
            .unwrap();

        let found = history.queued_build(b.id).unwrap();
        assert_eq!(found.command, "cargo build");
        assert_eq!(found.hook_pid, 1001);
        assert_eq!(found.slots_needed, 4);

        assert!(history.queued_build(999).is_none());
    }

    #[test]
    fn test_queue_unlimited_depth() {
        let _guard = test_guard!();
        // max_queue_depth = 0 means unlimited
        let history = BuildHistory::new(10).with_max_queue_depth(0);

        // Should be able to enqueue many
        for i in 0..1000 {
            assert!(
                history
                    .enqueue_build(format!("proj-{}", i), "build".into(), i, 4, None)
                    .is_some()
            );
        }

        assert_eq!(history.queue_depth(), 1000);
    }

    // =========================================================================
    // Saved Time Stats Tests
    // =========================================================================

    #[test]
    fn test_saved_time_stats_empty_history() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);
        let stats = history.saved_time_stats();

        assert_eq!(stats.builds_counted, 0);
        assert_eq!(stats.time_saved_ms, 0);
        assert_eq!(stats.total_remote_duration_ms, 0);
        assert_eq!(stats.estimated_local_duration_ms, 0);
    }

    #[test]
    fn test_saved_time_stats_only_local_builds() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // Add local builds only
        for i in 1..=3 {
            let mut record = make_build_record(i);
            record.location = BuildLocation::Local;
            record.duration_ms = 1000;
            history.record(record);
        }

        let stats = history.saved_time_stats();
        assert_eq!(stats.builds_counted, 0);
        assert_eq!(stats.time_saved_ms, 0);
    }

    #[test]
    fn test_saved_time_stats_without_local_baseline_makes_no_up_estimate() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // Remote builds only: there is no observed local baseline, so no local
        // estimate may be invented (the old code fabricated a 2.0x factor).
        for i in 1..=3 {
            let mut record = make_build_record(i);
            record.location = BuildLocation::Remote;
            record.worker_id = Some("worker-1".to_string());
            record.duration_ms = 1000;
            history.record(record);
        }

        let stats = history.saved_time_stats();
        assert_eq!(stats.builds_counted, 3);
        assert_eq!(stats.total_remote_duration_ms, 3000);
        assert_eq!(stats.estimate_basis, "none");
        assert_eq!(stats.local_baseline_builds, 0);
        assert_eq!(stats.estimated_local_duration_ms, 0);
        assert_eq!(stats.time_saved_ms, 0);
        assert_eq!(stats.today_saved_ms, 0);
        assert_eq!(stats.week_saved_ms, 0);
        assert_eq!(stats.avg_speedup, 0.0);
    }

    #[test]
    fn test_saved_time_stats_observed_local_mean_backs_estimate() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        for i in 1..=2 {
            let mut record = make_build_record(i);
            record.location = BuildLocation::Local;
            record.duration_ms = 2000;
            history.record(record);
        }
        let mut remote = make_build_record(3);
        remote.location = BuildLocation::Remote;
        remote.worker_id = Some("worker-1".to_string());
        remote.duration_ms = 1000;
        history.record(remote);

        let stats = history.saved_time_stats();
        assert_eq!(stats.builds_counted, 1);
        assert_eq!(stats.estimate_basis, "observed_local_mean");
        assert_eq!(stats.local_baseline_builds, 2);
        assert_eq!(stats.estimated_local_duration_ms, 2000);
        assert_eq!(stats.time_saved_ms, 1000);
        assert!((stats.avg_speedup - 2.0).abs() < 0.01);
    }

    #[test]
    fn test_saved_time_stats_mixed_builds() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // Add local builds
        for i in 1..=2 {
            let mut record = make_build_record(i);
            record.location = BuildLocation::Local;
            record.duration_ms = 2000; // Local takes 2s
            history.record(record);
        }

        // Add remote builds
        for i in 3..=4 {
            let mut record = make_build_record(i);
            record.location = BuildLocation::Remote;
            record.worker_id = Some("worker-1".to_string());
            record.duration_ms = 1000; // Remote takes 1s
            history.record(record);
        }

        let stats = history.saved_time_stats();
        assert_eq!(stats.builds_counted, 2);
        assert_eq!(stats.total_remote_duration_ms, 2000);
        // With avg local duration 2000ms: estimated local = 2 * 2000 = 4000
        assert_eq!(stats.estimated_local_duration_ms, 4000);
        assert_eq!(stats.time_saved_ms, 2000);
        assert!((stats.avg_speedup - 2.0).abs() < 0.01);
    }

    #[test]
    fn test_saved_time_stats_failed_builds_excluded() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // Add successful remote build
        let mut record1 = make_build_record(1);
        record1.location = BuildLocation::Remote;
        record1.worker_id = Some("worker-1".to_string());
        record1.duration_ms = 1000;
        record1.exit_code = 0;
        history.record(record1);

        // Add failed remote build
        let mut record2 = make_build_record(2);
        record2.location = BuildLocation::Remote;
        record2.worker_id = Some("worker-1".to_string());
        record2.duration_ms = 5000;
        record2.exit_code = 1;
        history.record(record2);

        let stats = history.saved_time_stats();
        // Only successful remote builds are counted
        assert_eq!(stats.builds_counted, 1);
        assert_eq!(stats.total_remote_duration_ms, 1000);
    }

    #[test]
    fn test_saved_time_stats_no_negative_savings() {
        let _guard = test_guard!();
        let history = BuildHistory::new(10);

        // Add fast local builds (500ms)
        for i in 1..=2 {
            let mut record = make_build_record(i);
            record.location = BuildLocation::Local;
            record.duration_ms = 500;
            history.record(record);
        }

        // Add slow remote builds (1000ms - slower than local!)
        for i in 3..=4 {
            let mut record = make_build_record(i);
            record.location = BuildLocation::Remote;
            record.worker_id = Some("worker-1".to_string());
            record.duration_ms = 1000;
            history.record(record);
        }

        let stats = history.saved_time_stats();
        // estimated_local = 500ms * 2 = 1000ms
        // time_saved = max(0, 1000 - 2000) = 0 (no negative savings)
        assert_eq!(stats.time_saved_ms, 0);
    }
}