diff --git a/docs/BALANCE.md b/docs/BALANCE.md
index 654879c1..bf8feb70 100644
--- a/docs/BALANCE.md
+++ b/docs/BALANCE.md
@@ -350,10 +350,11 @@ DPS, bounty picks from the exposure model, item carriers by effect, a tactician'
 | 同盟 绝境 (4 AI) | 5 → 11 / 40 | 11.85 → 12.22 | 5.1 → 10.9 | 224.6 → 215.2 | 155 → 142 of ≈ 310 picks | 2.00 → 3.05 | 2.3 → 2.4 |
 
 绝境 leaks per alive bot in R12 / R13: solo 8.0 / 9.2 → 5.7 / 4.9, co-op 20.5 / 20.7 → 17.6 / 17.8. Decision time per
-bot prep is unchanged (CPU p50 / p95, the rehearsal included: solo 绝境 520 / 1302 → 531 / 1216 ms, co-op 绝境
-405 / 884 → 404 / 910 ms; the heuristics alone 94 / 234 → 93 / 212 and 71 / 186 → 71 / 185 ms — a heavily loaded host,
-so the absolute values are ≈ 3–4× a quiet one; Σ per match within ± 2 %). The tuning sweeps used seeds 101–148 with the
-rehearsal off; the table is a separate seed range.
+bot prep is unchanged (one thread, back to back, seeds 1–6 — solo 绝境 / co-op 绝境; the rehearsal included): wall clock
+p50 / p95 87 / 175 → 82 / 191 ms and 84 / 172 → 80 / 157 ms, CPU 125 / 315 → 120 / 305 ms and 92 / 198 → 90 / 180 ms;
+the heuristics alone (CPU) 30 / 74 → 26 / 64 ms and 17 / 41 → 16 / 35 ms (the lineup search tries identical pieces
+once and reuses the lineup across refreshes); a 机变 pick 0.07 → 0.14 ms (p50). The tuning sweeps used seeds 101–148
+with the rehearsal off; the table above is a separate seed range.
 
 ---
 
diff --git a/test/match/bot-smarter.test.js b/test/match/bot-smarter.test.js
index 84ce13f5..88e951e0 100644
--- a/test/match/bot-smarter.test.js
+++ b/test/match/bot-smarter.test.js
@@ -4,7 +4,6 @@
 // measured with tools/botbench.mjs (docs/META.md §1.5).
 import { test } from 'node:test';
 import assert from 'node:assert/strict';
-import { performance } from 'node:perf_hooks';
 import { PHASE } from '../../shared/constants.js';
 import { botPickCard, bountyKillChance, itemTarget, arrange, botPrepBegin, rangeTiles } from '../../server/match/bot.js';
 import { parseKey } from '../../server/match/board.js';
@@ -165,12 +164,14 @@ test('cost guard: the prep heuristics of a late-round 4-bot match stay cheap (re
   h.run(() => m.phase === PHASE.PREP && m.round === 10);
   let worst = 0;
   for (const ps of m.alivePlayers()) {
-    const t0 = performance.now();
+    // CPU time (process.cpuUsage) rather than wall clock: a loaded host stretches it far less
+    const u0 = process.cpuUsage();
     assert.equal(botPrepBegin(m, ps), null, 'no rehearsal job with rehearsal off');
-    worst = Math.max(worst, performance.now() - t0);
+    const u = process.cpuUsage(u0);
+    worst = Math.max(worst, (u.user + u.system) / 1000);
   }
-  // ≈ 10–40 ms on a desktop; the bound only catches an accidental blow-up (it is load-sensitive like test/sim/robustness)
-  assert.ok(worst < 400, `worst bot prep ${worst.toFixed(1)} ms`);
+  // ≈ 10–40 ms on a quiet desktop (tools/botbench.mjs); the bound only catches an accidental blow-up
+  assert.ok(worst < 1500, `worst bot prep ${worst.toFixed(1)} ms of CPU`);
   assert.equal(m.errorCount, 0);
   m.dispose();
 });