The kickoff states that a slow server and an absent server are both normal, and
asks what a client shows in each case. They are different conditions with
different recoveries, and software that collapses them shows a person a spinner
for ninety seconds and then an error that says nothing.
The organisation publishes a number that makes this concrete. The first usable
tile appears in under 1.2 seconds from a cold start. A server that has not
answered by then has already missed the budget whether or not it eventually
answers.
What the record must answer
The thresholds, as numbers, at which the core changes what it reports. An answer
is late. An answer is not coming. The server is unreachable. Each threshold
carries the reason for its value rather than a round number chosen because it
looked tidy.
What the core reports at each stage, and that it reports progressively rather
than only at the end, so a client can show something truthful at 300 ms without
guessing.
What the core can still serve from cache while a request is outstanding, and how
a client asks for that rather than for the network.
The recovery path in each case. What the core retries by itself, what it waits
for, and what needs a person.
Done when
The record names each threshold with its value and its reason, states what the
core reports at each, and lets a reader derive what a client would put on screen
at 0.3 s, at 1.2 s, at 5 s, and at no answer at all.
The kickoff states that a slow server and an absent server are both normal, and
asks what a client shows in each case. They are different conditions with
different recoveries, and software that collapses them shows a person a spinner
for ninety seconds and then an error that says nothing.
The organisation publishes a number that makes this concrete. The first usable
tile appears in under 1.2 seconds from a cold start. A server that has not
answered by then has already missed the budget whether or not it eventually
answers.
What the record must answer
The thresholds, as numbers, at which the core changes what it reports. An answer
is late. An answer is not coming. The server is unreachable. Each threshold
carries the reason for its value rather than a round number chosen because it
looked tidy.
What the core reports at each stage, and that it reports progressively rather
than only at the end, so a client can show something truthful at 300 ms without
guessing.
What the core can still serve from cache while a request is outstanding, and how
a client asks for that rather than for the network.
The recovery path in each case. What the core retries by itself, what it waits
for, and what needs a person.
Done when
The record names each threshold with its value and its reason, states what the
core reports at each, and lets a reader derive what a client would put on screen
at 0.3 s, at 1.2 s, at 5 s, and at no answer at all.