- Shipped
- August 19, 2026 at 4:28 PM UTC
- Author
- Kamo
- Commit
- 4d0d229
It did its job. The question it was added to answer — whether the remaining delay was network, script or paint — turned out to have a fourth answer the probe found by elimination: across 11 real samples there were ZERO client-side navigations. Every click was a full document load, which is the fingerprint of the analytics tracker hijacking links (fixed in f321857), and it is why every server-side measurement kept coming back clean. The same instrumentation confirmed the fix rather than leaving it to feel: /solutions/agencies -> /pricing/offer total=162ms net=16ms block=0ms/0 /pricing/offer -> /blog total=142ms net=20ms block=0ms/0 /blog -> /glossary total=119ms net=26ms block=0ms/0 119-162ms click-to-painted, prefetch serving the payload in 16-26ms, and no main-thread blocking at all across zero long tasks. Removes NavPerfGate, NavPerfProbe and /api/perf-probe. Nothing that produced those numbers should outlive the question, least of all something reporting timings for every visitor.