perf(primer): drop pre-fire usage refresh so the Codex primer fires exactly on the slot

The Codex primer awaited refreshAllCodexAccountUsage + refreshActiveCodexUsage
before firing, purely to feed the old usage-based warm-up gating. With
forceAttempt the primer fires regardless of usage, so those pre-call refreshes
only added variable latency (the few-hundred-ms jitter that pushed the call off
the exact slot) and a needless dependency on the usage API. Fire first at the
slot; refresh once afterwards for observability (never delaying the call). The
hourly usage collector keeps cached usage current.

Verified: typecheck, build, bundle-smoke; usage-primer + codex-warmup suites pass
(primer now fires before any refresh; refreshAll no longer called).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

src/usage-primer.test.ts

@@ -54,7 +54,7 @@ vi.mock('./codex-warmup.js', () => ({
 }));
 
 describe('usage-primer', () => {
-  it('refreshes Codex usage before primer selection and fires regardless of usage level', async () => {
+  it('fires the Codex primer first with forceAttempt and refreshes usage only afterwards', async () => {
     callOrder.length = 0;
     refreshAllCodexAccountUsage.mockClear();
     refreshActiveCodexUsage.mockClear();
@@ -63,7 +63,10 @@ describe('usage-primer', () => {
     const { runCodexPrimerCycle } = await import('./usage-primer.js');
     await runCodexPrimerCycle();
 
-    expect(callOrder).toEqual(['refreshAll', 'refreshActive', 'warmup']);
+    // No pre-call usage refresh (that was the off-slot latency source); the
+    // command fires immediately, then usage is refreshed for observability.
+    expect(callOrder).toEqual(['warmup', 'refreshActive']);
+    expect(refreshAllCodexAccountUsage).not.toHaveBeenCalled();
     expect(runCodexWarmupCycle).toHaveBeenCalledWith(
       expect.objectContaining({
         enabled: true,

src/usage-primer.ts

@@ -3,10 +3,7 @@ import fs from 'fs';
 import path from 'path';
 
 import { CODEX_WARMUP_CONFIG } from './config.js';
-import {
-  refreshActiveCodexUsage,
-  refreshAllCodexAccountUsage,
-} from './codex-usage-collector.js';
+import { refreshActiveCodexUsage } from './codex-usage-collector.js';
 import { runCodexWarmupCycle } from './codex-warmup.js';
 import { logger } from './logger.js';
 import { getAllTokens } from './token-rotation.js';
@@ -153,8 +150,11 @@ async function runClaudePrimerCycle(): Promise<void> {
 
 export async function runCodexPrimerCycle(): Promise<void> {
   try {
-    await refreshAllCodexAccountUsage();
-    await refreshActiveCodexUsage();
+    // Fire immediately at the slot. There is NO pre-call usage refresh on
+    // purpose: forceAttempt fires regardless of usage, so reading usage first
+    // only added variable latency (the few-hundred-ms processing jitter that
+    // pushed the call off the exact slot) and a needless dependency on the
+    // usage API. The hourly usage collector keeps cached usage current.
     const result = await runCodexWarmupCycle(
       {
         ...CODEX_WARMUP_CONFIG,
@@ -164,17 +164,17 @@ export async function runCodexPrimerCycle(): Promise<void> {
         maxD7UsagePct: CODEX_PRIMER_MAX_D7_USAGE_PCT,
       },
       // forceAttempt: fire a real Codex command at *every* fixed slot,
-      // bypassing the opportunistic warm-up skip gates. Without it, two slots
-      // exactly 5h apart fall just under minIntervalMs (the primer fires a few
-      // hundred ms after the slot) and the later slot is silently skipped.
+      // bypassing the opportunistic warm-up skip gates (min-interval, stagger,
+      // failure cooldown, usage/rate-limit filters).
       { ignoreZeroUsageWindow: true, forceAttempt: true },
     );
-    // Best-effort: a slot may land while the trailing 5h window is exhausted,
-    // in which case this records `failed`. We deliberately do NOT try to
-    // re-anchor to the reported reset time: the 5h limit is a trailing rolling
-    // window whose reset slides forward with continued Codex usage, so no
-    // single timed call can pin the reset to a fixed clock time.
     logger.info({ result }, 'Codex primer cycle finished');
+    // Observability only — runs AFTER the call so it never delays it: refresh
+    // so the newly-anchored reset window shows up in usage data/logs. A slot
+    // may still land while the trailing 5h window is exhausted (records
+    // `failed`); we do not chase the reported reset time, since a trailing
+    // window keeps sliding with later usage.
+    await refreshActiveCodexUsage();
   } catch (err) {
     logger.warn({ err }, 'Codex primer cycle threw');
   }