test(chapter): [PKG-CHAPTER-05] correlate current-major restart timing failure

refactor/changes/2026-09-16-PKG-CHAPTER-05-major-timing.md

@@ -0,0 +1,38 @@
+# PKG-CHAPTER-05 current-major restart timing
+
+The registered core 6.0.0 and Chapter 2.0.0 candidates were checked with the
+unchanged installed quality-switch matrix: four old/new combinations across
+Chromium, Firefox and Windows WebKit, two workers, no retries. The run passed
+11 cases and failed one in 39,777.596 ms. The failed case used candidate core
+with published Chapter on WebKit 26.6. Both archives and the exact installed
+map are recorded in [the evidence](../baselines/chapter-major-timing-validation.json).
+
+This failure now has a directly correlated event/trace observation. Quality
+selection was recorded at page time 2982 ms and the required restart at 3601
+ms, a delay of 619 ms. The first poll returned an empty event list at 3567 ms.
+The next browser evaluation, trace call `call@241`, took 9989.901 ms and returned
+the correct restart list at page time 13643 ms. By then the existing 7000 ms
+poll deadline had expired. Later state showed the requested URL, paused video,
+position 3.0123779 seconds and no media error.
+
+Thus this particular failure is not evidence that restart itself was emitted
+late or missing. It demonstrates that observation through the browser crossed
+the deadline while the event was already recorded. The trace does not identify
+whether the underlying delay is media, rendering, OS scheduling or automation;
+the prior native-video five-second gaps remain relevant but do not prove a
+single common cause. No production workaround or altered wait policy is justified
+by this result alone.
+
+The full report, screenshot, trace and inline attachments were archived before
+another browser run. The committed record keeps relevant per-case states and
+the slow trace call; its report hash retains access to the full manifest and
+other attachments without duplicating them in every case. Historical published
+core WebKit position-reset assertions remain unchanged and are not counted as
+new restoration fixes.
+
+No source, test behavior, timeout, dependency or public API changed. The risk
+CHAPTER-TIMING-01 stays open and PKG-CHAPTER-05 stays doing. Required real devices
+and other combinations remain outstanding. Future diagnosis should target the
+observed browser stall rather than adding an artificial restart delay or rerunning
+until green. The registered build inputs remain valid because only evidence and
+status documentation changed. No formal review or remote operation occurred.