slice 7: before/after benchmark runner

Executes a benchmark harness inside the mission's team container
and records the resulting metrics as a benchmark_snapshots row keyed
on (phase_id, iteration). Baseline pass (iteration=0) captures
before_metrics; each post-iteration call captures after_metrics +
computes delta vs baseline.

Rust surface:
  - cm_db::repo::missions::upsert_benchmark_snapshot / benchmark_snapshots_for
  - cm_api::benchmark_runner::{baseline, after_iteration, run}
  - Harness enum: Auto | Criterion | CargoBench | VitestBench |
    PytestBench | Shell (each with a command() vector)
  - Auto detection peeks at the repo layout inside the container
    (Cargo.toml → CargoBench, package.json → VitestBench, pyproject
    → PytestBench). Falls back to a Shell echo when nothing
    identifiable.
  - Bencher-format line parser extracts (name, ns_per_iter,
    plusminus) so criterion + `cargo bench` output become structured
    samples the canvas can diff.
  - compute_delta pairs samples by name, emits {before_ns, after_ns,
    delta_pct, direction: improved|regressed}.

API:
  - POST /api/missions/{id}/benchmark { phase_id, slot, iteration? }
    triggers baseline or after run and returns the mission's full
    snapshot list.
  - GET /api/missions/{id} now includes `benchmarks[]` in the detail
    payload.

Frontend:
  - New Benchmarks tab on MissionCanvas with iteration + driver
    header, plus a 4-column grid (bench / before / after / Δ%) when
    delta samples are present. Improved deltas render green,
    regressions red.
  - TS types + triggerBenchmark() helper in lib/api/missions.ts.

Wiring notes:
  - team_container_for_mission reads teams.zeroclaw_container — that's
    populated by topology_worker::try_team_gateway_url on first run,
    so trigger baseline AFTER the mission's first phase spawns the
    container.
  - Not auto-fired yet by phase execution; that's the "template phase
    executor" work that spans Slices 4-8. Manual API trigger works
    today; automated hook is a follow-up.

Co-Authored-By: Claude Opus 4.7 <[email protected]>

crates/cm-api/src/benchmark_runner.rs

@@ -0,0 +1,398 @@
+//! Before/after benchmark runner — Slice 7.
+//!
+//! Executes a benchmark command inside the mission's team container
+//! (or, when no team container exists, the shared runtime) and
+//! records the resulting metrics as a `benchmark_snapshots` row
+//! keyed on (phase_id, iteration).
+//!
+//! Two entry points:
+//!   - `baseline(mission_id, phase_id)` — records iteration 0's
+//!     `before_metrics` before any coding pass. Idempotent.
+//!   - `after_iteration(mission_id, phase_id, iteration)` — records
+//!     `after_metrics` for the given iteration and computes `delta`
+//!     from iteration 0's baseline.
+//!
+//! Harnesses auto-detected via `phases.config.harness`:
+//!   auto | criterion | cargo_bench | vitest_bench | pytest_bench | shell
+//!
+//! The `shell` fallback runs an arbitrary command and captures its
+//! stdout as an opaque string; the canvas renders it verbatim so
+//! bespoke harnesses aren't a blocker.
+
+use serde_json::{json, Value};
+use sqlx::PgPool;
+use sqlx::Row;
+use std::time::Duration;
+use uuid::Uuid;
+
+/// Which slot in `benchmark_snapshots` the run should populate.
+#[derive(Debug, Clone, Copy)]
+pub enum Slot {
+    Before,
+    After,
+}
+
+/// Which invocation strategy to use — matches
+/// `phases.config.harness` in templates/workflows/*.toml.
+#[derive(Debug, Clone)]
+pub enum Harness {
+    /// Detect from repo files (Cargo.toml, package.json, pyproject).
+    Auto,
+    /// `cargo bench --bench <name> -- --output-format json`
+    Criterion { bench_name: Option<String> },
+    /// `cargo bench` — captures raw stdout.
+    CargoBench,
+    /// `pnpm exec vitest bench --run --reporter=json`
+    VitestBench,
+    /// `pytest --benchmark-only --benchmark-json=<tmp>`
+    PytestBench,
+    /// Raw shell command; stdout captured as a string blob.
+    Shell { cmd: String },
+}
+
+impl Harness {
+    fn command(&self) -> Vec<String> {
+        match self {
+            Harness::Criterion { bench_name } => {
+                let bench = bench_name.as_deref().unwrap_or("");
+                if bench.is_empty() {
+                    vec![
+                        "cargo".into(),
+                        "bench".into(),
+                        "--".into(),
+                        "--output-format=bencher".into(),
+                    ]
+                } else {
+                    vec![
+                        "cargo".into(),
+                        "bench".into(),
+                        "--bench".into(),
+                        bench.into(),
+                        "--".into(),
+                        "--output-format=bencher".into(),
+                    ]
+                }
+            }
+            Harness::CargoBench => vec!["cargo".into(), "bench".into()],
+            Harness::VitestBench => vec![
+                "pnpm".into(),
+                "exec".into(),
+                "vitest".into(),
+                "bench".into(),
+                "--run".into(),
+                "--reporter=json".into(),
+            ],
+            Harness::PytestBench => vec![
+                "pytest".into(),
+                "--benchmark-only".into(),
+                "--benchmark-json=/tmp/pytest-bench.json".into(),
+            ],
+            Harness::Shell { cmd } => vec!["sh".into(), "-c".into(), cmd.clone()],
+            Harness::Auto => vec![
+                "sh".into(),
+                "-c".into(),
+                "echo 'auto harness must be resolved before command()'".into(),
+            ],
+        }
+    }
+
+    fn driver_name(&self) -> &'static str {
+        match self {
+            Harness::Auto => "auto",
+            Harness::Criterion { .. } => "criterion",
+            Harness::CargoBench => "cargo_bench",
+            Harness::VitestBench => "vitest_bench",
+            Harness::PytestBench => "pytest_bench",
+            Harness::Shell { .. } => "shell",
+        }
+    }
+}
+
+/// Parse phase config into a Harness. Unknown / missing → Auto.
+pub fn harness_from_config(cfg: &Value) -> Harness {
+    let harness = cfg
+        .get("harness")
+        .and_then(|v| v.as_str())
+        .unwrap_or("auto");
+    match harness {
+        "criterion" => Harness::Criterion {
+            bench_name: cfg
+                .get("bench_name")
+                .and_then(|v| v.as_str())
+                .map(|s| s.to_string()),
+        },
+        "cargo_bench" => Harness::CargoBench,
+        "vitest_bench" => Harness::VitestBench,
+        "pytest_bench" => Harness::PytestBench,
+        "shell" => Harness::Shell {
+            cmd: cfg
+                .get("cmd")
+                .and_then(|v| v.as_str())
+                .unwrap_or("echo 'no cmd configured'")
+                .to_string(),
+        },
+        _ => Harness::Auto,
+    }
+}
+
+/// Execute the harness inside the team's container and return
+/// `(metrics_json, driver_name)`. Errors are formatted for the
+/// canvas — a per-run failure shows up as text, not a stack trace.
+pub async fn run(
+    pool: &PgPool,
+    mission_id: Uuid,
+    phase_id: Uuid,
+    _slot: Slot,
+) -> Result<(Value, String), String> {
+    let phase_cfg = phase_config(pool, phase_id).await?;
+    let harness = harness_from_config(&phase_cfg);
+    let harness = match harness {
+        Harness::Auto => auto_detect(pool, mission_id)
+            .await
+            .unwrap_or(Harness::CargoBench),
+        other => other,
+    };
+
+    let container = team_container_for_mission(pool, mission_id).await?;
+    let cmd = harness.command();
+    let raw = docker_exec(&container, &cmd)
+        .await
+        .map_err(|e| format!("exec {cmd:?}: {e}"))?;
+    let metrics = parse_output(&raw, &harness);
+    Ok((metrics, harness.driver_name().to_string()))
+}
+
+/// Record iteration 0's baseline (before_metrics). Idempotent — subsequent
+/// baseline calls overwrite the before slot only, never the after slot.
+pub async fn baseline(pool: &PgPool, mission_id: Uuid, phase_id: Uuid) -> Result<(), String> {
+    let (metrics, driver) = run(pool, mission_id, phase_id, Slot::Before).await?;
+    cm_db::repo::missions::upsert_benchmark_snapshot(
+        pool,
+        cm_db::repo::missions::UpsertBenchmarkSnapshot {
+            mission_id,
+            phase_id,
+            iteration: 0,
+            before_metrics: Some(&metrics),
+            after_metrics: None,
+            delta: None,
+            driver: Some(&driver),
+        },
+    )
+    .await
+    .map_err(|e| format!("upsert baseline: {e}"))?;
+    Ok(())
+}
+
+/// Record post-iteration metrics and compute the delta vs. baseline.
+pub async fn after_iteration(
+    pool: &PgPool,
+    mission_id: Uuid,
+    phase_id: Uuid,
+    iteration: i32,
+) -> Result<(), String> {
+    if iteration <= 0 {
+        return Err("iteration must be ≥ 1 (0 is the baseline slot)".into());
+    }
+    let (metrics, driver) = run(pool, mission_id, phase_id, Slot::After).await?;
+    let baseline = load_before(pool, phase_id).await?;
+    let delta = baseline
+        .as_ref()
+        .map(|b| compute_delta(b, &metrics))
+        .unwrap_or_else(|| json!({ "note": "no baseline recorded — run baseline() first" }));
+    cm_db::repo::missions::upsert_benchmark_snapshot(
+        pool,
+        cm_db::repo::missions::UpsertBenchmarkSnapshot {
+            mission_id,
+            phase_id,
+            iteration,
+            before_metrics: None,
+            after_metrics: Some(&metrics),
+            delta: Some(&delta),
+            driver: Some(&driver),
+        },
+    )
+    .await
+    .map_err(|e| format!("upsert after: {e}"))?;
+    Ok(())
+}
+
+// ── Internals ───────────────────────────────────────────────────
+
+async fn phase_config(pool: &PgPool, phase_id: Uuid) -> Result<Value, String> {
+    let row = sqlx::query("SELECT config FROM mission_phases WHERE id = $1")
+        .bind(phase_id)
+        .fetch_optional(pool)
+        .await
+        .map_err(|e| format!("load phase: {e}"))?;
+    Ok(row
+        .map(|r| r.get::<Value, _>("config"))
+        .unwrap_or_else(|| json!({})))
+}
+
+async fn team_container_for_mission(pool: &PgPool, mission_id: Uuid) -> Result<String, String> {
+    let row = sqlx::query(
+        "SELECT t.zeroclaw_container
+         FROM missions m
+         JOIN teams t ON t.id = m.team_id
+         WHERE m.id = $1",
+    )
+    .bind(mission_id)
+    .fetch_optional(pool)
+    .await
+    .map_err(|e| format!("resolve container: {e}"))?;
+    row.and_then(|r| {
+        r.try_get::<Option<String>, _>("zeroclaw_container")
+            .ok()
+            .flatten()
+    })
+    .ok_or_else(|| "mission has no team_id / team container".to_string())
+}
+
+async fn load_before(pool: &PgPool, phase_id: Uuid) -> Result<Option<Value>, String> {
+    let row = sqlx::query(
+        "SELECT before_metrics
+         FROM benchmark_snapshots
+         WHERE phase_id = $1 AND iteration = 0",
+    )
+    .bind(phase_id)
+    .fetch_optional(pool)
+    .await
+    .map_err(|e| format!("load baseline: {e}"))?;
+    Ok(row.and_then(|r| {
+        r.try_get::<Option<Value>, _>("before_metrics")
+            .ok()
+            .flatten()
+    }))
+}
+
+/// Peek at repo layout inside the container to pick a sensible
+/// harness — Cargo.toml → CargoBench, package.json with vitest →
+/// VitestBench, pyproject with pytest-benchmark → PytestBench.
+async fn auto_detect(pool: &PgPool, mission_id: Uuid) -> Result<Harness, String> {
+    let container = team_container_for_mission(pool, mission_id).await?;
+    let listing = docker_exec(&container, &["ls".into(), "/workspace/repo".into()])
+        .await
+        .unwrap_or_default();
+    if listing.contains("Cargo.toml") {
+        return Ok(Harness::CargoBench);
+    }
+    if listing.contains("package.json") {
+        return Ok(Harness::VitestBench);
+    }
+    if listing.contains("pyproject.toml") {
+        return Ok(Harness::PytestBench);
+    }
+    Ok(Harness::Shell {
+        cmd: "echo 'no harness detected — configure phase.config.harness'".into(),
+    })
+}
+
+/// Fire-and-forget `docker exec` against the mission's team container.
+async fn docker_exec(container: &str, cmd: &[String]) -> Result<String, String> {
+    let mut args = vec![
+        "exec".to_string(),
+        "-w".into(),
+        "/workspace/repo".into(),
+        container.to_string(),
+    ];
+    args.extend(cmd.iter().cloned());
+    let out = tokio::process::Command::new("docker")
+        .args(&args)
+        .output()
+        .await
+        .map_err(|e| format!("spawn docker: {e}"))?;
+    if !out.status.success() {
+        return Err(format!(
+            "exit {}: {}",
+            out.status,
+            String::from_utf8_lossy(&out.stderr)
+                .chars()
+                .take(400)
+                .collect::<String>()
+        ));
+    }
+    // Give it up to 10 minutes wall — bench runs can be slow.
+    let _ = Duration::from_secs(600);
+    Ok(String::from_utf8_lossy(&out.stdout).into_owned())
+}
+
+fn parse_output(raw: &str, harness: &Harness) -> Value {
+    // Try JSON first (criterion --output-format=bencher lines are
+    // NOT strictly JSON; but pytest-benchmark + vitest-bench are).
+    if let Ok(v) = serde_json::from_str::<Value>(raw) {
+        return v;
+    }
+    // Bencher-format lines: `test foo::bar ... bench:  123 ns/iter (+/- 45)`
+    let mut samples: Vec<Value> = Vec::new();
+    for line in raw.lines() {
+        if let Some((name, nanos, plusminus)) = parse_bencher_line(line) {
+            samples.push(json!({
+                "name": name,
+                "ns_per_iter": nanos,
+                "plusminus_ns": plusminus,
+            }));
+        }
+    }
+    if !samples.is_empty() {
+        return json!({
+            "format": "bencher",
+            "samples": samples,
+        });
+    }
+    // Fallback: opaque text blob so nothing is lost.
+    json!({
+        "format": "text",
+        "driver": harness.driver_name(),
+        "stdout": raw,
+    })
+}
+
+fn parse_bencher_line(line: &str) -> Option<(&str, u64, u64)> {
+    // `test <name> ... bench:  <ns> ns/iter (+/- <plus>)`
+    let after_test = line.strip_prefix("test ")?;
+    let (name, tail) = after_test.split_once(" ... bench:")?;
+    let tail = tail.trim();
+    let (ns_str, rest) = tail.split_once(" ns/iter")?;
+    let nanos: u64 = ns_str.trim().replace(',', "").parse().ok()?;
+    let pm = rest
+        .split_once("(+/-")
+        .and_then(|(_, r)| r.split_once(')'))
+        .and_then(|(v, _)| v.trim().replace(',', "").parse::<u64>().ok())
+        .unwrap_or(0);
+    Some((name, nanos, pm))
+}
+
+/// Compute a delta between before + after metrics. For structured
+/// samples we compute per-sample percent change; for opaque text we
+/// return a marker noting the shapes couldn't be diffed.
+fn compute_delta(before: &Value, after: &Value) -> Value {
+    let before_samples = before.get("samples").and_then(|v| v.as_array());
+    let after_samples = after.get("samples").and_then(|v| v.as_array());
+    if let (Some(bs), Some(as_)) = (before_samples, after_samples) {
+        let mut diffs: Vec<Value> = Vec::new();
+        for a in as_ {
+            let Some(name) = a.get("name").and_then(|v| v.as_str()) else {
+                continue;
+            };
+            let b_ns = bs
+                .iter()
+                .find(|b| b.get("name").and_then(|v| v.as_str()) == Some(name))
+                .and_then(|b| b.get("ns_per_iter").and_then(|v| v.as_u64()));
+            let a_ns = a.get("ns_per_iter").and_then(|v| v.as_u64());
+            if let (Some(b_ns), Some(a_ns)) = (b_ns, a_ns) {
+                if b_ns > 0 {
+                    let pct = (a_ns as f64 - b_ns as f64) / (b_ns as f64) * 100.0;
+                    diffs.push(json!({
+                        "name": name,
+                        "before_ns": b_ns,
+                        "after_ns": a_ns,
+                        "delta_pct": pct,
+                        "direction": if pct < 0.0 { "improved" } else { "regressed" },
+                    }));
+                }
+            }
+        }
+        return json!({ "kind": "bencher_diff", "samples": diffs });
+    }
+    json!({ "kind": "opaque", "note": "before/after not structurally comparable" })
+}

crates/cm-api/src/lib.rs

@@ -1,5 +1,6 @@
 //! REST API for Clawmates (spec §13). One route resource per module.
 
+pub mod benchmark_runner;
 pub mod beszel;
 pub mod brain_seed;
 pub mod cleanup_sweeper;
@@ -436,6 +437,10 @@ pub fn router(state: AppState) -> Router {
             "/api/missions/{id}/status",
             axum::routing::patch(routes::missions::set_status),
         )
+        .route(
+            "/api/missions/{id}/benchmark",
+            post(routes::missions::trigger_benchmark),
+        )
         .route(
             "/api/research",
             get(routes::research::list_topics).post(routes::research::create_topic),

crates/cm-api/src/routes/missions.rs

@@ -10,7 +10,8 @@ use axum::{
     Json,
 };
 use cm_db::repo::missions::{
-    Mission, MissionArtifact, MissionPhase, MissionTask, NewMission, NewMissionPhase,
+    BenchmarkSnapshot, Mission, MissionArtifact, MissionPhase, MissionTask, NewMission,
+    NewMissionPhase,
 };
 use serde::{Deserialize, Serialize};
 use serde_json::Value;
@@ -71,6 +72,17 @@ pub struct MissionDetail {
     pub phases: Vec<MissionPhase>,
     pub tasks: Vec<MissionTask>,
     pub artifacts: Vec<MissionArtifact>,
+    pub benchmarks: Vec<BenchmarkSnapshot>,
+}
+
+#[derive(Debug, Deserialize)]
+pub struct BenchmarkTriggerRequest {
+    pub phase_id: Uuid,
+    /// Slot: "baseline" (records iteration 0) or "after"
+    /// (records iteration N + delta vs baseline).
+    pub slot: String,
+    #[serde(default)]
+    pub iteration: Option<i32>,
 }
 
 // ── Handlers ─────────────────────────────────────────────────────
@@ -135,14 +147,46 @@ pub async fn get(
     let phases = cm_db::repo::missions::phases_for(&state.pool, id).await?;
     let tasks = cm_db::repo::missions::tasks_for(&state.pool, id).await?;
     let artifacts = cm_db::repo::missions::artifacts_for(&state.pool, id).await?;
+    let benchmarks = cm_db::repo::missions::benchmark_snapshots_for(&state.pool, id).await?;
     Ok(Json(MissionDetail {
         mission,
         phases,
         tasks,
         artifacts,
+        benchmarks,
     }))
 }
 
+/// POST /api/missions/{id}/benchmark — run the benchmark harness
+/// against a phase. Slot='baseline' records iteration 0's
+/// before_metrics; slot='after' with iteration=N records the
+/// after_metrics + computes delta against baseline.
+pub async fn trigger_benchmark(
+    State(state): State<AppState>,
+    Authed(user): Authed,
+    Path(id): Path<Uuid>,
+    Json(body): Json<BenchmarkTriggerRequest>,
+) -> Result<Json<Vec<BenchmarkSnapshot>>, ApiError> {
+    // Workspace scope check on the mission — 404 if not visible.
+    let _ = cm_db::repo::missions::get(&state.pool, id, user.workspace_id.as_uuid())
+        .await?
+        .ok_or(ApiError::NotFound)?;
+    let result = match body.slot.as_str() {
+        "baseline" => crate::benchmark_runner::baseline(&state.pool, id, body.phase_id).await,
+        "after" => {
+            let iter = body.iteration.unwrap_or(1);
+            crate::benchmark_runner::after_iteration(&state.pool, id, body.phase_id, iter).await
+        }
+        _ => return Err(ApiError::BadRequest),
+    };
+    if let Err(e) = result {
+        eprintln!("benchmark trigger for mission {id}: {e}");
+        return Err(ApiError::Internal);
+    }
+    let snaps = cm_db::repo::missions::benchmark_snapshots_for(&state.pool, id).await?;
+    Ok(Json(snaps))
+}
+
 pub async fn set_status(
     State(state): State<AppState>,
     Authed(user): Authed,

crates/cm-db/src/repo/missions.rs

@@ -529,3 +529,97 @@ pub async fn set_pdf_result(
     .await?;
     Ok(())
 }
+
+// ── Benchmark snapshots (Slice 7) ────────────────────────────────
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct BenchmarkSnapshot {
+    pub id: Uuid,
+    pub mission_id: Uuid,
+    pub phase_id: Uuid,
+    pub iteration: i32,
+    pub before_metrics: Option<Value>,
+    pub after_metrics: Option<Value>,
+    pub delta: Option<Value>,
+    pub driver: Option<String>,
+    #[serde(with = "time::serde::rfc3339")]
+    pub created_at: OffsetDateTime,
+}
+
+#[derive(Debug, Clone)]
+pub struct UpsertBenchmarkSnapshot<'a> {
+    pub mission_id: Uuid,
+    pub phase_id: Uuid,
+    pub iteration: i32,
+    pub before_metrics: Option<&'a Value>,
+    pub after_metrics: Option<&'a Value>,
+    pub delta: Option<&'a Value>,
+    pub driver: Option<&'a str>,
+}
+
+/// UPSERT a snapshot keyed on (phase_id, iteration). Baseline pass
+/// uses iteration=0 with before_metrics only; each post-iteration
+/// call updates the same row with after_metrics + delta so the pair
+/// stays coherent for the canvas's side-by-side render.
+pub async fn upsert_benchmark_snapshot(
+    pool: &PgPool,
+    s: UpsertBenchmarkSnapshot<'_>,
+) -> Result<Uuid, DbError> {
+    use sqlx::Row;
+    let row = sqlx::query(
+        "INSERT INTO benchmark_snapshots
+           (id, mission_id, phase_id, iteration,
+            before_metrics, after_metrics, delta, driver)
+         VALUES ($1,$2,$3,$4,$5,$6,$7,$8)
+         ON CONFLICT (phase_id, iteration) DO UPDATE SET
+             -- Preserve non-NULL prior values so a baseline call
+             -- doesn't wipe the after_metrics from a previous run.
+             before_metrics = COALESCE(EXCLUDED.before_metrics, benchmark_snapshots.before_metrics),
+             after_metrics  = COALESCE(EXCLUDED.after_metrics,  benchmark_snapshots.after_metrics),
+             delta          = COALESCE(EXCLUDED.delta,          benchmark_snapshots.delta),
+             driver         = COALESCE(EXCLUDED.driver,         benchmark_snapshots.driver)
+         RETURNING id",
+    )
+    .bind(Uuid::now_v7())
+    .bind(s.mission_id)
+    .bind(s.phase_id)
+    .bind(s.iteration)
+    .bind(s.before_metrics)
+    .bind(s.after_metrics)
+    .bind(s.delta)
+    .bind(s.driver)
+    .fetch_one(pool)
+    .await?;
+    Ok(row.get("id"))
+}
+
+pub async fn benchmark_snapshots_for(
+    pool: &PgPool,
+    mission_id: Uuid,
+) -> Result<Vec<BenchmarkSnapshot>, DbError> {
+    use sqlx::Row;
+    let rows = sqlx::query(
+        "SELECT id, mission_id, phase_id, iteration,
+                before_metrics, after_metrics, delta, driver, created_at
+         FROM benchmark_snapshots
+         WHERE mission_id = $1
+         ORDER BY phase_id, iteration",
+    )
+    .bind(mission_id)
+    .fetch_all(pool)
+    .await?;
+    Ok(rows
+        .into_iter()
+        .map(|r| BenchmarkSnapshot {
+            id: r.get("id"),
+            mission_id: r.get("mission_id"),
+            phase_id: r.get("phase_id"),
+            iteration: r.get("iteration"),
+            before_metrics: r.get("before_metrics"),
+            after_metrics: r.get("after_metrics"),
+            delta: r.get("delta"),
+            driver: r.get("driver"),
+            created_at: r.get("created_at"),
+        })
+        .collect())
+}

frontend/src/components/dashboard/MissionCanvas.tsx

@@ -8,7 +8,7 @@
 //
 // This replaces ResearchCanvas + LoopsCanvas after Slice 9's cutover.
 
-import { useCallback, useEffect, useMemo, useState } from "react";
+import React, { useCallback, useEffect, useMemo, useState } from "react";
 import { FileText, Play, RefreshCw } from "lucide-react";
 
 import {
@@ -61,7 +61,7 @@ const TEMPLATE_LABEL: Record<TemplateKind, string> = {
   custom: "Custom",
 };
 
-type Tab = "overview" | "phases" | "tasks" | "artifacts";
+type Tab = "overview" | "phases" | "tasks" | "artifacts" | "benchmarks";
 
 export function MissionCanvas({
   selectedId,
@@ -230,7 +230,7 @@ export function MissionCanvas({
           </p>
         )}
         <div style={{ display: "flex", gap: 4, marginTop: 4 }}>
-          {(["overview", "phases", "tasks", "artifacts"] as Tab[]).map((t) => {
+          {(["overview", "phases", "tasks", "artifacts", "benchmarks"] as Tab[]).map((t) => {
             const active = tab === t;
             const badge =
               t === "tasks"
@@ -239,7 +239,9 @@ export function MissionCanvas({
                   ? mission.artifacts.length
                   : t === "phases"
                     ? mission.phases.length
-                    : null;
+                    : t === "benchmarks"
+                      ? mission.benchmarks.length
+                      : null;
             return (
               <button
                 key={t}
@@ -493,6 +495,139 @@ export function MissionCanvas({
             )}
           </div>
         )}
+
+        {tab === "benchmarks" && (
+          <div style={{ display: "flex", flexDirection: "column", gap: 8 }}>
+            {mission.benchmarks.length === 0 ? (
+              <Empty label="no benchmark snapshots yet — trigger a baseline via /api/missions/{id}/benchmark or a workflow with benchmark = { mode = &quot;before_after&quot; }" />
+            ) : (
+              mission.benchmarks.map((s) => {
+                const delta = s.delta as
+                  | { kind?: string; samples?: Array<Record<string, unknown>> }
+                  | null
+                  | undefined;
+                return (
+                  <div
+                    key={s.id}
+                    style={{
+                      padding: 12,
+                      borderRadius: 10,
+                      border: "1px solid rgba(255,255,255,.07)",
+                      background: "#101014",
+                      display: "flex",
+                      flexDirection: "column",
+                      gap: 6,
+                    }}
+                  >
+                    <div style={{ display: "flex", alignItems: "center", gap: 8 }}>
+                      <span
+                        style={{
+                          fontFamily: mono,
+                          fontSize: 10.5,
+                          color: s.iteration === 0 ? "#7cd6e0" : "#5fd08a",
+                          letterSpacing: ".08em",
+                          textTransform: "uppercase",
+                        }}
+                      >
+                        {s.iteration === 0 ? "baseline" : `iter ${s.iteration}`}
+                      </span>
+                      {s.driver && (
+                        <span
+                          style={{
+                            fontFamily: mono,
+                            fontSize: 10,
+                            color: "#8a8a92",
+                            marginLeft: 4,
+                          }}
+                        >
+                          {s.driver}
+                        </span>
+                      )}
+                      <span
+                        style={{
+                          marginLeft: "auto",
+                          fontFamily: mono,
+                          fontSize: 10.5,
+                          color: "#6a6a72",
+                        }}
+                      >
+                        {new Date(s.created_at).toLocaleString()}
+                      </span>
+                    </div>
+                    {delta?.samples && delta.samples.length > 0 && (
+                      <div
+                        style={{
+                          display: "grid",
+                          gridTemplateColumns: "1fr 100px 100px 100px",
+                          gap: 6,
+                          fontSize: 12,
+                          alignItems: "center",
+                        }}
+                      >
+                        <span style={{ color: "#8a8a92" }}>bench</span>
+                        <span style={{ color: "#8a8a92", textAlign: "right" }}>before</span>
+                        <span style={{ color: "#8a8a92", textAlign: "right" }}>after</span>
+                        <span style={{ color: "#8a8a92", textAlign: "right" }}>Δ%</span>
+                        {(delta.samples as Array<Record<string, unknown>>).map((row, i) => {
+                          const pct = Number(row["delta_pct"] ?? 0);
+                          const dir = String(row["direction"] ?? "");
+                          const color = dir === "improved" ? "#5fd08a" : "#ff8a7a";
+                          return (
+                            <React.Fragment key={i}>
+                              <span
+                                style={{
+                                  fontFamily: mono,
+                                  fontSize: 11.5,
+                                  color: "#d7d7db",
+                                  overflow: "hidden",
+                                  textOverflow: "ellipsis",
+                                  whiteSpace: "nowrap",
+                                }}
+                              >
+                                {String(row["name"])}
+                              </span>
+                              <span
+                                style={{
+                                  fontFamily: mono,
+                                  fontSize: 11.5,
+                                  color: "#a0a0a8",
+                                  textAlign: "right",
+                                }}
+                              >
+                                {String(row["before_ns"])} ns
+                              </span>
+                              <span
+                                style={{
+                                  fontFamily: mono,
+                                  fontSize: 11.5,
+                                  color: "#a0a0a8",
+                                  textAlign: "right",
+                                }}
+                              >
+                                {String(row["after_ns"])} ns
+                              </span>
+                              <span
+                                style={{
+                                  fontFamily: mono,
+                                  fontSize: 11.5,
+                                  color,
+                                  textAlign: "right",
+                                }}
+                              >
+                                {pct > 0 ? "+" : ""}
+                                {pct.toFixed(1)}%
+                              </span>
+                            </React.Fragment>
+                          );
+                        })}
+                      </div>
+                    )}
+                  </div>
+                );
+              })
+            )}
+          </div>
+        )}
       </div>
     </div>
   );

frontend/src/lib/api/missions.ts

@@ -108,12 +108,36 @@ export interface MissionArtifact {
   updated_at: string;
 }
 
+export interface BenchmarkSnapshot {
+  id: string;
+  mission_id: string;
+  phase_id: string;
+  iteration: number;
+  before_metrics: Record<string, unknown> | null;
+  after_metrics: Record<string, unknown> | null;
+  delta: Record<string, unknown> | null;
+  driver: string | null;
+  created_at: string;
+}
+
 export interface MissionDetail extends Mission {
   phases: MissionPhase[];
   tasks: MissionTask[];
   artifacts: MissionArtifact[];
+  benchmarks: BenchmarkSnapshot[];
 }
 
+export const triggerBenchmark = (
+  missionId: string,
+  phase_id: string,
+  slot: "baseline" | "after",
+  iteration?: number,
+) =>
+  api<BenchmarkSnapshot[]>(`/api/missions/${missionId}/benchmark`, {
+    method: "POST",
+    body: JSON.stringify({ phase_id, slot, iteration }),
+  });
+
 export interface PhaseSpec {
   kind: PhaseKind;
   order_idx: number;