diff --git a/.gitignore b/.gitignore
index 770bc07..6cc2103 100644
--- a/.gitignore
+++ b/.gitignore
@@ -233,3 +233,4 @@ targets/comrak/candidate-*/
 targets/comrak/asan*/
 targets/tokei/candidate-*/
 targets/tokei/asan/
+!results/calibration/*.json
diff --git a/CLAUDE.md b/CLAUDE.md
index 62b836a..86ac6a7 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -54,6 +54,7 @@ just build / test / lint    # trust-layer workspace: cargo build/test/clippy+fmt
 just aa [cmd]               # A/A self-test of bench-runner on a shell command
 just compare <a> <b>        # interleaved A/B of two shell commands, bootstrap CI
 just pin-check <target>     # verify corpus MANIFEST.sha256
+just calibrate <cmd> <out>  # automated A/A + regression-injection calibration
 ```
 
 ## Definition of done — one optimization attempt
diff --git a/crates/bench-runner/src/exec.rs b/crates/bench-runner/src/exec.rs
index 9e536ad..a3d2100 100644
--- a/crates/bench-runner/src/exec.rs
+++ b/crates/bench-runner/src/exec.rs
@@ -15,7 +15,7 @@ pub struct MeasuredSamples {
     pub candidate_s: Vec<f64>,
 }
 
-fn time_once(shell_cmd: &str) -> anyhow::Result<f64> {
+fn time_once(shell_cmd: &str, inject_spin_s: f64) -> anyhow::Result<f64> {
     let start = Instant::now();
     let status = Command::new("sh")
         .arg("-c")
@@ -24,6 +24,15 @@ fn time_once(shell_cmd: &str) -> anyhow::Result<f64> {
         .stderr(Stdio::null())
         .status()
         .with_context(|| format!("failed to spawn: {shell_cmd}"))?;
+    if inject_spin_s > 0.0 {
+        // Calibration-only: a real busy-wait inside the timed window, so an
+        // injected slowdown flows through the entire measurement pipeline
+        // (SPEC §3.1 regression-injection self-test).
+        let spin_until = start.elapsed().as_secs_f64() + inject_spin_s;
+        while start.elapsed().as_secs_f64() < spin_until {
+            std::hint::spin_loop();
+        }
+    }
     let elapsed = start.elapsed().as_secs_f64();
     anyhow::ensure!(status.success(), "command exited nonzero: {shell_cmd}");
     Ok(elapsed)
@@ -31,23 +40,26 @@ fn time_once(shell_cmd: &str) -> anyhow::Result<f64> {
 
 /// Run baseline and candidate commands on the interleaved schedule and
 /// return measured wall times per side. For A/A mode, pass the same
-/// command as both sides.
+/// command as both sides. `inject_candidate_spin_s` adds a synthetic
+/// busy-wait to candidate runs only (calibration self-tests; 0.0 in
+/// normal operation).
 pub fn run_interleaved(
     baseline_cmd: &str,
     candidate_cmd: &str,
     runs_per_side: usize,
     warmup: usize,
+    inject_candidate_spin_s: f64,
 ) -> anyhow::Result<MeasuredSamples> {
     let mut out = MeasuredSamples {
         baseline_s: Vec::with_capacity(runs_per_side),
         candidate_s: Vec::with_capacity(runs_per_side),
     };
     for (side, measured) in interleaved(runs_per_side, warmup) {
-        let cmd = match side {
-            Side::Baseline => baseline_cmd,
-            Side::Candidate => candidate_cmd,
+        let (cmd, spin) = match side {
+            Side::Baseline => (baseline_cmd, 0.0),
+            Side::Candidate => (candidate_cmd, inject_candidate_spin_s),
         };
-        let elapsed = time_once(cmd)?;
+        let elapsed = time_once(cmd, spin)?;
         if measured {
             match side {
                 Side::Baseline => out.baseline_s.push(elapsed),
diff --git a/crates/bench-runner/src/main.rs b/crates/bench-runner/src/main.rs
index dae2957..1e79e89 100644
--- a/crates/bench-runner/src/main.rs
+++ b/crates/bench-runner/src/main.rs
@@ -28,6 +28,95 @@ enum Cmd {
         #[arg(long)]
         cmd: String,
     },
+    /// Automated calibration (SPEC §3.1): N A/A sessions must show a
+    /// false-positive rate <5%, and N sessions with a synthetic
+    /// slowdown injected into the candidate must detect it ≥95% of the
+    /// time. Writes a JSON evidence file.
+    Calibrate {
+        #[arg(long)]
+        cmd: String,
+        #[arg(long, default_value_t = 20)]
+        sessions: usize,
+        /// Injected relative slowdown for the detection test.
+        #[arg(long, default_value_t = 0.05)]
+        slowdown: f64,
+        /// Output JSON evidence path.
+        #[arg(long)]
+        out: PathBuf,
+    },
+}
+
+fn calibrate(
+    cfg: &bench_runner::config::AcceptConfig,
+    cmd: &str,
+    sessions: usize,
+    slowdown: f64,
+    out: &PathBuf,
+) -> Result<()> {
+    let fp = EnvFingerprint::collect("caller-provided (wrap cmd in taskset)", "system-default");
+    // Baseline median to size the injected busy-wait.
+    let probe = exec::run_interleaved(cmd, cmd, 5, 1, 0.0)?;
+    let inject_s = stats::median(&probe.baseline_s) * slowdown;
+
+    let mut aa_sessions = Vec::new();
+    let mut inj_sessions = Vec::new();
+    let (mut false_positives, mut detections) = (0usize, 0usize);
+    for i in 0..sessions {
+        // Vary the bootstrap seed per session; resampling with one fixed
+        // seed across sessions would correlate the very statistic we are
+        // trying to validate.
+        let seed = cfg.bootstrap_seed.wrapping_add(i as u64);
+        let aa = exec::run_interleaved(cmd, cmd, cfg.runs_per_side, cfg.warmup_runs, 0.0)?;
+        let aa_ci = stats::bootstrap_ratio_ci(
+            &aa.baseline_s, &aa.candidate_s, cfg.bootstrap_iters, cfg.confidence, seed,
+        );
+        if decide(aa_ci, cfg.threshold) == Verdict::Accepted {
+            false_positives += 1;
+        }
+        aa_sessions.push(serde_json::json!({"median": aa_ci.median, "lo": aa_ci.lo, "hi": aa_ci.hi}));
+
+        let inj =
+            exec::run_interleaved(cmd, cmd, cfg.runs_per_side, cfg.warmup_runs, inject_s)?;
+        let inj_ci = stats::bootstrap_ratio_ci(
+            &inj.baseline_s, &inj.candidate_s, cfg.bootstrap_iters, cfg.confidence, seed,
+        );
+        // Detected = the harness resolves a regression: CI entirely < 1.
+        if inj_ci.hi < 1.0 {
+            detections += 1;
+        }
+        inj_sessions.push(serde_json::json!({"median": inj_ci.median, "lo": inj_ci.lo, "hi": inj_ci.hi}));
+        eprintln!(
+            "session {}/{sessions}: A/A [{:.4},{:.4}] fp={false_positives}; inj [{:.4},{:.4}] det={detections}",
+            i + 1, aa_ci.lo, aa_ci.hi, inj_ci.lo, inj_ci.hi
+        );
+    }
+
+    let fp_rate = false_positives as f64 / sessions as f64;
+    let det_rate = detections as f64 / sessions as f64;
+    let pass = fp_rate < 0.05 && det_rate >= 0.95;
+    let evidence = serde_json::json!({
+        "workload_cmd": cmd,
+        "sessions": sessions,
+        "runs_per_side": cfg.runs_per_side,
+        "threshold": cfg.threshold,
+        "injected_slowdown": slowdown,
+        "injected_busy_wait_s": inject_s,
+        "aa_false_positive_rate": fp_rate,
+        "aa_sessions": aa_sessions,
+        "injection_detection_rate": det_rate,
+        "injection_sessions": inj_sessions,
+        "acceptance": {"fp_lt": 0.05, "detection_ge": 0.95, "pass": pass},
+        "env_fingerprint": fp,
+    });
+    std::fs::write(out, serde_json::to_string_pretty(&evidence)?)?;
+    println!(
+        "calibration: A/A false-positive rate {fp_rate:.3} (<0.05 required), \
+         injected-{:.0}%-slowdown detection rate {det_rate:.3} (>=0.95 required) -> {}",
+        slowdown * 100.0,
+        if pass { "PASS" } else { "FAIL" }
+    );
+    anyhow::ensure!(pass, "calibration acceptance criteria not met");
+    Ok(())
 }
 
 fn main() -> Result<()> {
@@ -40,6 +129,9 @@ fn main() -> Result<()> {
     let (baseline, candidate, aa_mode) = match &cli.command {
         Cmd::Compare { baseline, candidate } => (baseline.clone(), candidate.clone(), false),
         Cmd::Aa { cmd } => (cmd.clone(), cmd.clone(), true),
+        Cmd::Calibrate { cmd, sessions, slowdown, out } => {
+            return calibrate(&cfg, cmd, *sessions, *slowdown, out);
+        }
     };
 
     let fp = EnvFingerprint::collect("none", "system-default");
@@ -50,7 +142,7 @@ fn main() -> Result<()> {
     );
 
     let samples =
-        exec::run_interleaved(&baseline, &candidate, cfg.runs_per_side, cfg.warmup_runs)?;
+        exec::run_interleaved(&baseline, &candidate, cfg.runs_per_side, cfg.warmup_runs, 0.0)?;
     let ci = stats::bootstrap_ratio_ci(
         &samples.baseline_s,
         &samples.candidate_s,
diff --git a/justfile b/justfile
index 9a6a647..9453cf5 100644
--- a/justfile
+++ b/justfile
@@ -27,3 +27,9 @@ compare baseline candidate:
 # (wired through diff-test CLI once it exists; manual check for now)
 pin-check target:
     cd corpora/{{target}} && sh gen-corpus.sh
+
+# Automated calibration: N A/A sessions (<5% false-positive required) +
+# N injected-slowdown sessions (>=95% detection required), JSON evidence
+calibrate cmd out sessions="20":
+    cargo run -p bench-runner -- --config config/accept.toml calibrate \
+        --cmd "{{cmd}}" --sessions {{sessions}} --out "{{out}}"
diff --git a/results/calibration/2026-07-06-comrak-full-calibration.json b/results/calibration/2026-07-06-comrak-full-calibration.json
new file mode 100644
index 0000000..e2a6c0c
--- /dev/null
+++ b/results/calibration/2026-07-06-comrak-full-calibration.json
@@ -0,0 +1,228 @@
+{
+  "aa_false_positive_rate": 0.0,
+  "aa_sessions": [
+    {
+      "hi": 1.015998931989443,
+      "lo": 0.9905153352677495,
+      "median": 1.0012776949264486
+    },
+    {
+      "hi": 1.0055502756953207,
+      "lo": 0.981646945965943,
+      "median": 0.9908293785706317
+    },
+    {
+      "hi": 1.0204094554652672,
+      "lo": 0.9846287241295675,
+      "median": 1.002373764499574
+    },
+    {
+      "hi": 1.0047915300411685,
+      "lo": 0.9860341673780392,
+      "median": 0.9964988810591069
+    },
+    {
+      "hi": 1.0102726750450453,
+      "lo": 0.9876236508283872,
+      "median": 0.9947054654657848
+    },
+    {
+      "hi": 1.0298441811992136,
+      "lo": 0.9876526600106184,
+      "median": 1.0086506615393056
+    },
+    {
+      "hi": 1.0099440547974916,
+      "lo": 0.9822805275362518,
+      "median": 0.993850347123508
+    },
+    {
+      "hi": 1.00385962045072,
+      "lo": 0.9848165937114653,
+      "median": 0.9939030705425225
+    },
+    {
+      "hi": 1.0115679493048169,
+      "lo": 0.9909357476627676,
+      "median": 1.0008452273219608
+    },
+    {
+      "hi": 1.013718263355025,
+      "lo": 0.977768078322329,
+      "median": 0.9983105725894315
+    },
+    {
+      "hi": 1.0070675650128182,
+      "lo": 0.9868898707776302,
+      "median": 0.9979255224721003
+    },
+    {
+      "hi": 1.01473324312686,
+      "lo": 0.984779294044258,
+      "median": 0.9942794507746874
+    },
+    {
+      "hi": 1.0484508437256985,
+      "lo": 0.9663266833242238,
+      "median": 1.0028654978002207
+    },
+    {
+      "hi": 1.0153007892092838,
+      "lo": 0.9903096574162721,
+      "median": 1.0049001338403383
+    },
+    {
+      "hi": 1.0092343435725986,
+      "lo": 0.9922852873631642,
+      "median": 1.0034947624362798
+    },
+    {
+      "hi": 1.0125833696661481,
+      "lo": 0.9858629698133391,
+      "median": 0.9945866155256149
+    },
+    {
+      "hi": 1.0322221482398073,
+      "lo": 0.9951458973996147,
+      "median": 1.0075699560145404
+    },
+    {
+      "hi": 1.009761041226371,
+      "lo": 0.9898464904505755,
+      "median": 0.998425451880562
+    },
+    {
+      "hi": 1.016631057133635,
+      "lo": 0.9805588949385654,
+      "median": 0.997950429996102
+    },
+    {
+      "hi": 1.0200791563569052,
+      "lo": 0.9873713898804045,
+      "median": 1.000136749596359
+    }
+  ],
+  "acceptance": {
+    "detection_ge": 0.95,
+    "fp_lt": 0.05,
+    "pass": true
+  },
+  "env_fingerprint": {
+    "aslr": "mode=system-default randomize_va_space=2",
+    "cpu_model": "Intel(R) Xeon(R) Processor @ 2.80GHz",
+    "cpu_pinning": "caller-provided (wrap cmd in taskset)",
+    "governor": "unknown",
+    "kernel": "6.18.5",
+    "smt": "0",
+    "turbo": "unknown"
+  },
+  "injected_busy_wait_s": 0.011125513,
+  "injected_slowdown": 0.05,
+  "injection_detection_rate": 0.95,
+  "injection_sessions": [
+    {
+      "hi": 0.9705730747977983,
+      "lo": 0.9384443885194832,
+      "median": 0.9570129481885624
+    },
+    {
+      "hi": 0.9691260796731174,
+      "lo": 0.9309798719977085,
+      "median": 0.9411908680460361
+    },
+    {
+      "hi": 0.9775705854771548,
+      "lo": 0.9464129998680949,
+      "median": 0.9569584325473247
+    },
+    {
+      "hi": 0.9738958239067871,
+      "lo": 0.9490577184781656,
+      "median": 0.9624758687422929
+    },
+    {
+      "hi": 0.9692636780671424,
+      "lo": 0.9448230573540036,
+      "median": 0.9611951204744579
+    },
+    {
+      "hi": 0.960038069546381,
+      "lo": 0.9412872028306517,
+      "median": 0.953204396987671
+    },
+    {
+      "hi": 0.970357721583405,
+      "lo": 0.9312571815624449,
+      "median": 0.9497958159533386
+    },
+    {
+      "hi": 1.0189632738942447,
+      "lo": 0.8979330813534421,
+      "median": 0.9650177023489201
+    },
+    {
+      "hi": 0.9715620167742915,
+      "lo": 0.9336764011030461,
+      "median": 0.9512964417642272
+    },
+    {
+      "hi": 0.9768953182714468,
+      "lo": 0.9385015313970146,
+      "median": 0.9538566142021077
+    },
+    {
+      "hi": 0.9733246543232819,
+      "lo": 0.9452506173210715,
+      "median": 0.9558422793746748
+    },
+    {
+      "hi": 0.968562587511655,
+      "lo": 0.9364861001420586,
+      "median": 0.9461689697819208
+    },
+    {
+      "hi": 0.9832173753801465,
+      "lo": 0.9352025813901989,
+      "median": 0.9618202496291594
+    },
+    {
+      "hi": 0.9701329322900167,
+      "lo": 0.9395861281724339,
+      "median": 0.954612721677948
+    },
+    {
+      "hi": 0.9921703664327655,
+      "lo": 0.9357083455043089,
+      "median": 0.958758177852707
+    },
+    {
+      "hi": 0.976082080482069,
+      "lo": 0.9328145868694007,
+      "median": 0.947453955805073
+    },
+    {
+      "hi": 0.9795318432851117,
+      "lo": 0.9317744456440425,
+      "median": 0.9607784721426458
+    },
+    {
+      "hi": 0.9751440256938523,
+      "lo": 0.9491637711903045,
+      "median": 0.9646173778660493
+    },
+    {
+      "hi": 0.9730959153169779,
+      "lo": 0.928638531103063,
+      "median": 0.9490375911197446
+    },
+    {
+      "hi": 0.9630989374168489,
+      "lo": 0.9484798599580982,
+      "median": 0.9548094809061651
+    }
+  ],
+  "runs_per_side": 30,
+  "sessions": 20,
+  "threshold": 0.02,
+  "workload_cmd": "taskset -c 2 sh -c 'targets/comrak/workspace/target/release/comrak < corpora/comrak/progit.md > /dev/null'"
+}
\ No newline at end of file