todlando: W3 run 35995451280 @98ae312e GREEN 5/5 (Linux 3285/3285, Windows 3322/3322). NOT handing off yet — two gaps CI cannot see:
(1) int adapter_peer_e2e has NOT executed anywhere (thin lane = kind(lib)+kind(bin); it's a test binary). It spawns two real daemons.
(2) docs-site/src/cli/reference.md is not regenerated for --via-subnet / --remote (grep: 0 hits); drift gate is not in the thin lane, golden would red.
Ask: a box window on hfenduleam for `cargo run -p xtask -- gen` + `cargo nextest run -p spt --test adapter_peer_e2e` (--jobs 2, own pool claimed). Then rebase onto main tip after #256/IR-146 and hand off.
F1 raw first-run lines (nextest, test prints nothing else):
Linux  11:57:09.1144815Z PASS [0.017s] (2171/3285) spt-runtime entry_exec::tests::a_command_climbing_out_of_the_install_is_not_an_entry
Win    12:05:26.4588352Z PASS [0.296s] (2234/3322) spt-runtime entry_exec::tests::a_command_climbing_out_of_the_install_is_not_an_entry
That test's first assert is the premise (install/../victim is_file && starts_with(install)) — it passed on both OS, so the lexical hole is MEASURED, then the canonical fix held. Existing entry_exec arms (declared_entries_*, force_heals_* on Linux) also PASS.