bde1af57

test: e2e — (a) death asserted across the window; (d) ceiling reasoning; (e) evidence note

a73x   2026-08-11 18:39

test: e2e — (a) death asserted across the window; (d) ceiling reasoning; (e) evidence note

All three from the Task 7 spec review, none changing what the suite covers.

(a) The cold scenario claimed bytes could only be riding QUIC because the
ssh that carried the announce was dead. It observed that death at ONE
instant, before the second marker was typed; it now re-observes it after
the marker has made its round trip, which is the claim the scenario
actually rests on — dead across the whole window, not dead once. The
`kill -0` idiom was checked both ways before being trusted: it exits 1 on
a live pid and 0 on a dead one under set -eu, so it is a check that can
fail rather than a line that always passes.

(d) The ceiling stays 10000ms where the plan suggested deadline+3s, and
the comment now says why instead of leaving the difference looking like an
oversight: it separates bounded from hung and is asked to do nothing
finer. Measured spread is 2026-2129ms over 13 runs, and an overrun does
not arrive a few hundred milliseconds late — it arrives at the
connection's idle timeout, 15000ms, the only other bound in the dial. A
tighter number would catch nothing more and would make the one scenario
that already pays 2s by design the most load-sensitive check here.

(e) Two of its three assertions have never been exercised by a mutation,
because the suite stops at its first failure and every mutation so far
dies earlier. That is now written at the assertions themselves. They are
untested rather than unfalsifiable — the absent-grep is the same shape as
(c)'s, which the deleted-fallback-line mutation killed — and saying so is
worth more than a comment that implies coverage nobody has demonstrated.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>