Method · 12 March 2026
Cold start is not one number
Teams still arrive with a single TTFD printed in a large typeface. The number is not fake. It is just the wrong shape for a launch review. Process birth, class load, first draw, and first useful interaction move independently. When you average them into one trophy, a 40 ms “win” can hide a webview tour that still blocks the ledger.
In class we force a three-slice chart. Slice A is process birth to first draw. Slice B is first draw to first useful interaction — the moment a user can actually tap something that matters. Slice C is anything you deliberately excluded: splash branding, feature education, A/B wrappers. If you cannot name slice C, you are not finished measuring.
The teaching traces in the archive include a Bangkok super-app whose lab TTFD improved after a brand refresh while slice B got worse. The product channel had been celebrating slice A. Office hours that week were uncomfortable, which is the point. Mobile Performance Benchmark Analytics is a reporting craft before it is a tooling craft.
We also record thermal history. A cold start measured on a phone that has been on a charger in air-conditioning is a different experiment from a phone that rode a bag from Asok. Module 03 will make that explicit; module 02 only asks you to write the ambient note in the header. If the header is missing, the trace does not enter the homework set.
None of this requires a new profiler. It requires a refusal to publish a single number when the trace shows three stories. If you want the template, it ships with Trace Reader notes and with the Bench Lead seat. Bring your own capture to office hours; we will not grade a screenshot of a dashboard.